newsroom.sgit.ai / admin / versions

Release history

Every push to dev is a release: CI validates the site, verifies the version bump, tags the commit v{release}.{major}.{minor}, and deploys to GitHub Pages. How it works.

VersionDateWhat shipped
v0.14.203 22 Nov 2026 @newsroom.writer: two writers as twins on different accounts, at Dinis's word (register max_open 2, check 15, sign and register tools, claims file) In the release.
v0.14.202 22 Nov 2026 @newsroom.writer: start-up debrief for a journalist twin, newsroom.d@sgit.ai.dinis (at Dinis's word; maestro.b's guidance to follow); @newsroom.writer: the twin debrief, with maestro.b's guidance (names, start-up, the claim rule) In the release.
v0.14.201 21 Nov 2026 @newsroom.writer: Programmatic stories: the presenters library is published In the release.
v0.14.200 21 Nov 2026 @newsroom.writer: The primitives of trust In the release.
v0.14.199 21 Nov 2026 @newsroom.writer: Programmatic stories In the release.
v0.14.198 21 Nov 2026 @newsroom.writer: Start, stop, slow down In the release.
v0.14.197 21 Nov 2026 @newsroom.writer: Person Graph, and a correction to Learning from Wikipedia In the release.
v0.14.196 21 Nov 2026 @newsroom.writer: Learning from Wikipedia, with DeepSeek: a plan In the release.
v0.14.195 21 Nov 2026 @newsroom.writer: Can you find me? In the release.
v0.14.194 21 Nov 2026 @newsroom.writer: A web of trust for careers, and An anonymous profile, vouched for In the release.
v0.14.193 21 Nov 2026 @newsroom.writer: DeepSeek in our workflows In the release.
v0.14.192 21 Nov 2026 @newsroom.writer: LangGraph and us In the release.
v0.14.191 21 Nov 2026 @newsroom.writer: The SGit Server Factory, and Hired as an agentic worker In the release.
v0.14.190 21 Nov 2026 @newsroom.writer: One night on Mistral, The showroom, and a correction to a-european-brain In the release.
v0.14.189 21 Nov 2026 @newsroom.writer: The way in In the release.
v0.14.188 21 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 21 November In the release.
v0.14.187 20 Nov 2026 @newsroom.writer: The first round trip: Sovereign Website, not Sovereign Web In the release.
v0.14.186 20 Nov 2026 @newsroom.writer: The first round trip In the release.
v0.14.185 20 Nov 2026 @newsroom.writer: Transcription for a sector; a European brain revised In the release.
v0.14.184 20 Nov 2026 @newsroom.writer: A European brain for a sovereign website In the release.
v0.14.183 20 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 20 November In the release.
v0.14.182 20 Nov 2026 @newsroom.writer: The swarm, one real run at a time In the release.
v0.14.181 20 Nov 2026 @newsroom.writer: The RiskMandate X-ray; a galaxy of agents In the release.
v0.14.180 19 Nov 2026 @newsroom.writer: A career website, not a CV; fifteen minutes of an expert In the release.
v0.14.179 19 Nov 2026 @newsroom.writer: Every tool its own website; websites for the moment In the release.
v0.14.178 19 Nov 2026 @newsroom.writer: Where to sell the book and the appliances; when a publisher asks In the release.
v0.14.177 19 Nov 2026 @newsroom.writer: Launching the sovereign website appliance: five tiers, AI on every device @newsroom.writer: Launching the sovereign website appliance: five tiers, AI on every device @newsroom.writer: Launching the sovereign website appliance: five tiers, local AI on the Pro tiers @newsroom.writer: Launching the sovereign website appliance (revised tiers); a sovereign website on a Raspberry Pi In the release.
v0.14.176 19 Nov 2026 @newsroom.writer: Watch your website being built In the release.
v0.14.175 19 Nov 2026 @newsroom.writer: The launch plan in eleven turns; launching the sovereign website appliance In the release.
v0.14.174 19 Nov 2026 @newsroom.writer: What is a sovereign website; simple on top of years; sovereign websites, the investment case In the release.
v0.14.173 19 Nov 2026 @newsroom.writer: Your website, from a vault In the release.
v0.14.172 19 Nov 2026 @newsroom.writer: A website with a team inside In the release.
v0.14.171 19 Nov 2026 @newsroom.writer: Designing sovereignwebsite.ai: the brief In the release.
v0.14.170 19 Nov 2026 @newsroom.writer: Sovereign Agentic Website (SAW), the launch In the release.
v0.14.169 19 Nov 2026 @newsroom.writer: Your own registrar; The second run (a forensic look at the release's 24 minutes) In the release.
v0.14.168 19 Nov 2026 @newsroom.writer: Policies in practice; CI behaviour policies updated with the factory's nine calls and its wider grant In the release.
v0.14.167 19 Nov 2026 @newsroom.writer: Feature behaviour policies; the front page's day lists: a quiet section label, the whole row opens the article In the release.
v0.14.166 19 Nov 2026 @newsroom.writer: Four articles from writer-09's hand-over dated 19 November, the day they were published, so they appear with today's In the release.
v0.14.165 19 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 19 November In the release.
v0.14.164 19 Nov 2026 @newsroom.writer: A page we never charge (the free YouTube page); four articles from writer-09: Rent or own, No lock-in, Sovereignty you can move, The sovereign website; signed-commit check hardened In the release.
v0.14.163 19 Nov 2026 @newsroom.writer: signed commits on. Every writer commit is now also a git SSH signature by the writer's registered key, its OpenSSH line published on the registration (git_signing) beside its session, and validate.js check 15 verifies the signature and the Claude-Session: trailer, in CI on HEAD and locally on every writer commit since signing began; CI's version commits are exempt. signing-setup.sh and sshsig.mjs (from PrompTugal's dev@) set it up, repo-locally. At the editor's word (admin/notes/2026-11-19__signed-commits/notes.md). In the release.
v0.14.162 18 Nov 2026 @newsroom.writer: Ten days series, part twelve, The agents that didn't get the memo In the release.
v0.14.161 18 Nov 2026 @newsroom.writer: writer-10 opened for the new writer session on the editor's second account and writer-09 closed, at the editor's word (admin/notes/2026-11-18__writer-10/memo-01.md). In the release.
v0.14.160 18 Nov 2026 @newsroom.writer: @newsroom.writer: Ten days series, part eleven, Agents that retire themselves In the release.
v0.14.159 18 Nov 2026 @newsroom.writer: @newsroom.writer: Ten days series part ten, Pennies against dollars; writer-09's hand-over debrief In the release.
v0.14.158 18 Nov 2026 @newsroom.writer: @newsroom.writer: a new series, Gaining speed, with its first three parts In the release.
v0.14.157 18 Nov 2026 @newsroom.writer: @newsroom.writer: new article, A factory that films its work In the release.
v0.14.156 18 Nov 2026 @newsroom.writer: @newsroom.writer: new article, Whose event is it? In the release.
v0.14.155 18 Nov 2026 @newsroom.writer: @newsroom.writer: RiskMandate and Kosmoy revised: starts at £0, open so anyone can build it in In the release.
v0.14.154 18 Nov 2026 @newsroom.writer: @newsroom.writer: new article, RiskMandate and Kosmoy In the release.
v0.14.153 18 Nov 2026 @newsroom.writer: @newsroom.writer: Ten days series, part nine, A guide born in a storm In the release.
v0.14.152 18 Nov 2026 @newsroom.writer: @newsroom.writer: new article, From a feed to meaning In the release.
v0.14.151 18 Nov 2026 @newsroom.writer: @newsroom.writer: new article, Start with an interview In the release.
v0.14.150 18 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 18 November In the release.
v0.14.149 18 Nov 2026 @newsroom.writer: @newsroom.writer: new article, The pitch In the release.
v0.14.148 18 Nov 2026 @newsroom.writer: @newsroom.writer: two new articles, You bring the business, and Static against the platforms In the release.
v0.14.147 17 Nov 2026 @newsroom.writer: @newsroom.writer: new article, Partners sell, we build In the release.
v0.14.146 17 Nov 2026 @newsroom.writer: @newsroom.writer: new article, The website cell In the release.
v0.14.145 17 Nov 2026 @newsroom.writer: @newsroom.writer: Ten days series, part eight, After the storm In the release.
v0.14.144 17 Nov 2026 @newsroom.writer: @newsroom.writer: thumbnails on the front page's day list In the release.
v0.14.143 17 Nov 2026 @newsroom.writer: @newsroom.writer: three new articles, Paid by the read, Newsrooms on open data, and Governance from zero In the release.
v0.14.142 17 Nov 2026 @newsroom.writer: @newsroom.writer: new article, Any data, with any data; signing workflow added to Who wrote it, who steered it, who signs it In the release.
v0.14.141 17 Nov 2026 @newsroom.writer: @newsroom.writer: new article, Who wrote it, who steered it, who signs it In the release.
v0.14.140 17 Nov 2026 @newsroom.writer: New article, Working with WizPress: how the AI operating system for newsrooms and our open-source trust stack fit together, six ways to work together and a thirty-day pilot. Two figures. @newsroom.writer: @newsroom.writer: new article, HeritageTech, together: how our open-source stack could plug into UAlgTec Heritage (UTH) @newsroom.writer: @newsroom.writer: two new articles on UAlgTec Heritage (UTH): HeritageTech, together, and Our companies and UTH In the release.
v0.14.139 17 Nov 2026 @newsroom.writer: New article, Keys after payment: how to hand vault and access keys to paying customers; what Stripe and other providers do after payment; a design sealing each key to the buyer's own public key, and claim codes as stock. Three figures. In the release.
v0.14.138 17 Nov 2026 @newsroom.writer: Two new articles from my own morning in Portugal: The ATM that asked twice (a 13.6% currency-conversion mark-up offered twice, the EU rule, six levers, AI as the consumer's evidence) and The payment desk that could not take payment (code for the money, AI for the change). Six images, from PrompTugal.news. In the release.
v0.14.137 17 Nov 2026 @newsroom.writer: Two new articles: Patient vaults, answered point by point (seven questions from a healthcare team, and QR codes for keys and links), and Your own agentic cell, end to end (a key, a vault, the factory, three human gates). Three figures. In the release.
v0.14.136 17 Nov 2026 @newsroom.writer: Ten days series, part seven: Agents rating agents, GAMMA-Ray, the draft standard eight agents built to rate AI agents on freedom and governance; hub links part seven; schedule updated. One figure and two screenshot collages. In the release.
v0.14.135 17 Nov 2026 @newsroom.writer: Three new articles: Relate, debriefed (Neo4j's agent-native graph database against our graphs in vaults); Memory is a lifecycle (a reply to Francois Vanderseypen); Who hands out the work? (a reply to Markus Franz). Four figures and two posts. In the release.
v0.14.134 17 Nov 2026 @newsroom.writer: New article, Deploy sgit in your own environment from a CI pipeline: one image, ciphertext in your own S3 in your region, an access token, and the site factory's pipeline pattern (protected environment, OIDC, narrow trust, checks, teardown), with a proposed Action, a skill and a prompt. Two figures. In the release.
v0.14.133 17 Nov 2026 @newsroom.writer: Two new articles: Build on VoiceDebrief (record without loss, transcribe offline on the device, debriefs by Apple Intelligence, voice control, three layers, CI, a prompt for your Claude) and Shipping an iOS app with agents (CI as compiler, signing once, TestFlight on every push, and the three doors to users). Six figures. In the release.
v0.14.132 17 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 17 November In the release.
v0.14.131 16 Nov 2026 @newsroom.writer: New article, Five primitives, one store: a reply to Andy Goundry on Anthropic's always-on agents guide; the five primitives in one versioned, signed store, mapped to AgentiCCells cells, and the levels we have added, all open. Three figures and the post. In the release.
v0.14.130 16 Nov 2026 @newsroom.writer: New article, A portfolio with a film first: Silvio Switha's site, made in under an hour by the site factory and five design agents; his music video first, five designs, his agents as a solar system and a mixing desk, a studio, a proposal and a gig poster. One figure and eight screenshot collages. In the release.
v0.14.129 16 Nov 2026 @newsroom.writer: New article, A website that is her: Yiheng Wang's site, made in three passes in one evening by the site factory, five design agents and her own agent; five designs, her agents drawn three ways, a printable brief, a page for your AI, every claim sourced. One figure and ten screenshot collages. In the release.
v0.14.128 16 Nov 2026 @newsroom.writer: Show me the evidence, updated: virtual professionals (a vCISO, vCTO, vDPO, designer and editor for hire) who sell their judgement and the agent team behind it; five new personas, two images, a brief item. In the release.
v0.14.127 16 Nov 2026 @newsroom.writer: New article, Stopping a hundred agents, part six of the Ten days series: the Big Stop, the hand brake and the final hold, the wake cadence, what got through, and why a stop is a file. Three figures. Also a wording fix in Show me the evidence. In the release.
v0.14.126 16 Nov 2026 @newsroom.writer: New article, Show me the evidence: websites for people looking for a job, built by an agentic cell; seven pages, a page per employer, a recruiter pack, eight personas to fork, the cell ladder and the brief for the team. Nine figures and screenshots. In the release.
v0.14.125 16 Nov 2026 @newsroom.writer: New article: Fork it In the release.
v0.14.124 16 Nov 2026 @newsroom.writer: Our workflow on Odoo 20: sources summary refreshed In the release.
v0.14.123 16 Nov 2026 @newsroom.writer: New article: The cell is the unit; Odoo documentation sources added to Our workflow on Odoo 20 In the release.
v0.14.122 16 Nov 2026 @newsroom.writer: Three new articles: The email agentic cell, VoiceDebrief relaunched, Our workflow on Odoo 20; Odoo and our stack corrected @newsroom.writer: Three new articles: The email agentic cell, VoiceDebrief relaunched, Our workflow on Odoo 20; Odoo and our stack corrected In the release.
v0.14.121 16 Nov 2026 @newsroom.writer: Three new articles: Specialist agents for every team, A CRM for us and for our agents, PrompTugal.news the next phase In the release.
v0.14.120 16 Nov 2026 @newsroom.writer: Odoo and our stack: Odoo 12 to 20 and an ephemeral proof of concept; new article, Integrate with whatever is on the other side In the release.
v0.14.119 16 Nov 2026 @newsroom.writer: The intersection: OmKneeHealth added as SGit's public health scenario In the release.
v0.14.118 16 Nov 2026 @newsroom.writer: New article: Odoo and our stack In the release.
v0.14.117 16 Nov 2026 @newsroom.writer: New article: The intersection, one stack of separate companies and one ladder on every site In the release.
v0.14.116 16 Nov 2026 @newsroom.writer: Two new articles: The site factory works, and CI behaviour policies In the release.
v0.14.115 16 Nov 2026 @newsroom.writer: Ten days hub: series summary refreshed for part five In the release.
v0.14.114 16 Nov 2026 @newsroom.writer: New article: The Mac bottleneck (Ten days, part five); hub and schedule updated In the release.
v0.14.113 16 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 16 November In the release.
v0.14.112 16 Nov 2026 @newsroom.writer: New article: Every story has an address, a design proposal for PrompTugal.news as a ground-up newsroom, with mockups In the release.
v0.14.111 16 Nov 2026 @newsroom.writer: New article: Identity for AI agents, layer by layer, a reply to Maya Dotan's map of 190 projects In the release.
v0.14.110 16 Nov 2026 @newsroom.writer: Question screenshots in the reply articles now sit beside the text (aside), and the horizontal one is narrower In the release.
v0.14.109 15 Nov 2026 @newsroom.writer: reply articles now show the question they answer, with the person's name and photo as posted publicly: Khwaja Naveed's comment in Where assist becomes execute, Florent Daudens' post in Selling capabilities, not content, and William Hu's post in Always-on agents at team scale. In the release.
v0.14.108 15 Nov 2026 @newsroom.writer: new article, Where assist becomes execute: an answer to Khwaja Naveed's question on the line from summarising to executing in a system of record; Agent Behaviour Policies tier by capability and undo class (assist, own work, system of record, irreversible), enforce each tier with a boundary above the agent, and record the rule that failed when it was only a rule. Two figures. In the release.
v0.14.107 15 Nov 2026 @newsroom.writer: Selling capabilities, not content: the Storm Charles section no longer tells the site as one spoken request; it credits what made the speed possible, the method built over months before it (conductors, roles and hand-overs, files as memory, keys and lanes, release gates, observability and visualisation), at the editor's word. In the release.
v0.14.106 15 Nov 2026 @newsroom.writer: two new articles: PrompTugal.news by place (the updated strategy: Portugal, the Algarve and every Algarve town, each on its own subdomain from the start, each site a lens on one shared semantic graph, and the PrompTugal test for every story) and Selling capabilities, not content (a reply to Florent Daudens' We have the protocols. Where are the products?, with the Storm Charles site and PrompTugal.news by place as capabilities). Three figures. In the release.
v0.14.105 15 Nov 2026 @newsroom.writer: new article, From a commit to a live website: a site factory where a request in a commit makes a public CI pipeline create the GitHub repository, the DNS record and the sgit vault, seed a page, check HTTPS, optionally run a first pass with Claude Code, and record it all, with a GitHub App, AWS OpenID Connect and a policy for the pipeline; so an agent creates a site by committing. Two figures. In the release.
v0.14.104 15 Nov 2026 @newsroom.writer: Know the Reach and Every hour is twenty years: the video's poster is now the song's own artwork (made with ChatGPT), and the second RiskMandate song is credited: Strangers in the Loop, released on 14 November, produced by Indi Gupte, with a little help from Dinis on the lyrics; a figure with both covers. In the release.
v0.14.103 15 Nov 2026 @newsroom.writer: Ten days at Web Summit, part four, Native apps by agents: four iPhone and iPad apps from one workspace (1,164 lines of Swift to 48,230) and an Android tablet app (a signed build 4 h 11 after the first commit; 33,042 lines of Kotlin) in five days, TestFlight on every push, and a fringe guide from brief to TestFlight in about 10.5 hours; a collage, a figure and a screenshot. The hub links it; the schedule marks it live and The Mac bottleneck next. In the release.
v0.14.102 15 Nov 2026 @newsroom.writer: two new articles: A hub for agentic sites (hub.agenticcells.site: every site in a repository and an sgit vault, served at its own address, found, promoted and copied as a template, with a made-up Algarve orange grower as the first example) and Spend fewer tokens (an agentic team's token bill is mostly a design choice: files as memory, caching, code where you can, the right model per step, no copy and paste, static outputs, counting it). Two figures. In the release.
v0.14.101 15 Nov 2026 @newsroom.writer: two new articles: Stop copying and pasting between your agents (if you are the wire between your AI sessions you are the bottleneck; let agents talk through recorded channels, scale decisions with structure, read outputs; what AgentiCCells sets up) and Don't micromanage your agents (Vicky Wang's shift from reading everything to managing, her team backed by specialist agents from another team, and specialist agents for every team in a company). Three figures. In the release.
v0.14.100 15 Nov 2026 @newsroom.writer: new article, Atomic cells, end to end: AgentiCCells' first offering, a public website as an atomic cell anchored at £500; an order on a queue, an agentic team that builds, a customer who steers it from their browser with a key and a vault lane read hourly, a handover to their GitHub, a second mode in the customer's own GitHub and Claude accounts, and the first implementers. Three figures. In the release.
v0.14.99 15 Nov 2026 @newsroom.writer: new article, Reading your way into an agentic team: eight public sites a new agentic team should read before it builds anything (newsroom.sgit.ai, web-summit.sgit.ai, sgit.ai, abp.sgit.ai, coding.sgit.ai, nfrs.sgit.ai, ruv.sgit.ai, promptugal.news), how an agent should read them, and ten questions to answer first. Starting AgentiCCells: the repository link moves to AgentiCCells/AgentiCCells__Website, and the first version goes live on agenticcells.com while the .ai nameservers are sorted. In the release.
v0.14.98 15 Nov 2026 @newsroom.writer: new article, Starting AgentiCCells: a company that sets up small agentic teams (cells) in the customer's own environment, with a control centre, everything open and no state on our side; how it was named, researched with Perplexity (a market map of 30+ vendors and the gap), reviewed, and set up in a day (three domains, Route 53, a repository, and a bootstrap brief for its first agentic team). Two figures. In the release.
v0.14.97 15 Nov 2026 @newsroom.writer: new article, An agentic console: a simulated control room for a team of agentic teams in a public vault (admin/content/vaults/agentic-console.json), shaped on the Web Summit team: ten accounts as trust zones, time to rewind and five scenarios to replay, signed messages on three channels, stops that say when they land, a view per person, and the console as an agent; with the customer history read backwards (ruv.sgit.ai, the book, Web Summit, RiskMandate as a partner, PrompTugal.news as the first customer), also added to Selling mini agentic teams. In the release.
v0.14.96 15 Nov 2026 @newsroom.writer: two new articles: Selling mini agentic teams (a business that sets up small agentic teams for companies, with a control centre to see and stop them, PrompTugal.news as the first customer, Agent Behaviour Policies as the same offering on steroids, and a ladder from £500 to £20,000) and Six setups of an agentic system (from a task to conductors of conductors, with the altitudes of a run from the Swarm demo inside each: make the dangerous thing unreachable). Four figures. In the release.
v0.14.95 15 Nov 2026 @newsroom.writer: new article, Decide, then buy: deciding what to buy and buying it are two businesses, one for explorers and villagers and one for town planners, mapped on a Wardley map; keep them apart, give the open path from twenty-five steps to one, and sell a better deal (time, trust, execution, expertise, volume prices) instead of locking the customer in. Two figures. In the release.
v0.14.94 15 Nov 2026 @newsroom.writer: three new articles on how we would use the way we run agents for others, with a public demo vault for all three (admin/content/vaults/agentic-patterns.json): A brain you arrive at (a company brain made for one company, division and team, reached from one person's brain upwards; the Web Summit team ran one), The new data lake (append-only lanes, a curator agent, a semantic graph that draws its own pictures), and A market for sweets (the opposite of a market for lemons: a semantic graph for one financial decision, an agent that never trades, everything apart from the trade). In the release.
v0.14.93 15 Nov 2026 @newsroom.writer: Ten days at Web Summit, part three, PrompTugal.news, the newsroom we ran: desks of agents that captured, mapped, wrote and reviewed 73 signed articles in ten days, 61 named agents in 43 roles, an accountant in API-equivalent dollars, and fifty outreach drafts no agent was allowed to send; a collage, two screenshots and two figures. The hub links it; the schedule marks it live and Native apps by agents next. In the release.
v0.14.92 15 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 15 November In the release.
v0.14.91 15 Nov 2026 @newsroom.writer: Always-on agents at team scale, for readers arriving from Anthropic's post: a new section on the scale of what we ran (the collage of 24 sites and apps, the ten days in numbers, the ten days in order), fourteen more links to earlier articles in the text, and a reading list to start from. In the release.
v0.14.90 15 Nov 2026 @newsroom.writer: new article, Always-on agents at team scale: Anthropic's five primitives for agents that sleep and wake (William Hu, 13 November), set beside the around 250 named agents six of us ran at Web Summit: where we agree, where we were wrong, and the next problems once there are many (agents starting agents, identity, stopping, many humans, seeing it, governance). Two figures. In the release.
v0.14.89 15 Nov 2026 @newsroom.writer: the wording check's allow list gains the editor's reviewed description of Silvio Switha in Every hour is twenty years ("one of the best musicians I know", his own words about a named colleague, not a claim about this site); v0.14.88 shipped with the check flagging it, and this clears it. In the release.
v0.14.88 15 Nov 2026 @newsroom.writer: new article, Every hour is twenty years: the human behind Know the Reach, Silvio Switha, a musician, singer and producer whose twenty years shaped every choice in the song and its video; AI changed the speed, not the vision. Videos from YouTube now play on a click (admin/build/youtube_embed.py, the !youtube line): our own poster, nothing fetched from YouTube until the reader presses play; Know the Reach gets the player too. In the release.
v0.14.87 15 Nov 2026 @newsroom.writer: new article, Know the Reach: RiskMandate's song is on Spotify, and the first 53 seconds of its music video were made with AI by Silvio Switha from about 2,300 words of direction, with the production audit's counts (43 images, 16 motion clips, 21 edited exports, 1,236 Runway credits), the eight decisions, and two figures. In the release.
v0.14.86 14 Nov 2026 @newsroom.writer: The team's operating system (Ten days, part two) In the release.
v0.14.85 14 Nov 2026 @newsroom.writer: The front page drops the php2shell developing-story strip In the release.
v0.14.84 14 Nov 2026 @newsroom.writer: The book is on Leanpub In the release.
v0.14.83 14 Nov 2026 @newsroom.writer: Videos inside articles from the media vault; Town planners for mission-critical code; the hub says six of us In the release.
v0.14.82 14 Nov 2026 @newsroom.writer: Ten days at Web Summit (the hub); A film studio on a laptop; the newsroom media vault In the release.
v0.14.81 14 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 14 November In the release.
v0.14.80 13 Nov 2026 @newsroom.writer: test_wire accepts the comma newer browsers put after the weekday In the release.
v0.14.79 13 Nov 2026 @newsroom.writer: Coding standards are behaviour policies; writer-09 In the release.
v0.14.78 13 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 13 November In the release.
v0.14.77 12 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 12 November In the release.
v0.14.76 11 Nov 2026 @newsroom.writer: Storm Charles: the size of it In the release.
v0.14.75 11 Nov 2026 @newsroom.writer: Storm Charles, a newsroom built in a night In the release.
v0.14.74 11 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 11 November In the release.
v0.14.73 10 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 10 November In the release.
v0.14.72 9 Nov 2026 @newsroom.writer: RiskMandate and Cloudflare; RiskMandate and Lovable In the release.
v0.14.71 9 Nov 2026 @newsroom.writer: RiskMandate and Manus In the release.
v0.14.70 9 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 9 November In the release.
v0.14.69 8 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 8 November In the release.
v0.14.68 7 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 7 November In the release.
v0.14.67 6 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 6 November In the release.
v0.14.66 5 Nov 2026 @newsroom.writer: Who else controls AI agents at Web Summit In the release.
v0.14.65 5 Nov 2026 @newsroom.writer: RC2 goes to print: the wrap post In the release.
v0.14.64 5 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 5 November In the release.
v0.14.63 4 Nov 2026 @newsroom.writer: article: synthetic users for your website, an onboarding guide @newsroom.writer: two articles: synthetic users for your website (an onboarding guide); selling at the right altitude, with its vault of three websites @newsroom.writer: two articles: the Web Summit team, part five; an inbox only your agents can write to @newsroom.writer: the inbox article and part five link the team's inbox page In the release.
v0.14.62 4 Nov 2026 @newsroom.writer: article: a key for every person (the Web Summit team's person keys, sealed items and signed append lanes) In the release.
v0.14.61 4 Nov 2026 @newsroom.writer: article: resilience when the workforce includes agents (OCRR, NeuShield, Actifile, and where RiskMandate, sgit and SHarness meet them) In the release.
v0.14.60 4 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 4 November In the release.
v0.14.59 4 Nov 2026 @newsroom.writer: the book notes moved out of the repository (kept in history) In the release.
v0.14.58 4 Nov 2026 @newsroom.writer: note: the narrator's voice, the slides were always a compromise In the release.
v0.14.57 4 Nov 2026 @newsroom.writer: note: the narrator's voice, no judgement of the reader In the release.
v0.14.56 4 Nov 2026 @newsroom.writer: note: the narrator's voice in the book, a new introduction and part one rewritten for the book team In the release.
v0.14.55 3 Nov 2026 @newsroom.writer: the WSO2 article: the highlights video is confirmed as the one in the post In the release.
v0.14.54 3 Nov 2026 @newsroom.writer: withdrawn: the RiskMandate partner programme; the cost of your agents is no longer called the first partner integration @newsroom.writer: withdrawn: the RiskMandate partner programme; the cost of your agents is no longer called the first partner integration @newsroom.writer: withdrawn: the RiskMandate partner programme; the cost of your agents is no longer called the first partner integration In the release.
v0.14.53 3 Nov 2026 @newsroom.writer: the reply to Florent Daudens now links the section for publishers In the release.
v0.14.52 3 Nov 2026 @newsroom.writer: article: would you pay to plug this newsroom in, a reply to Florent Daudens In the release.
v0.14.51 3 Nov 2026 @newsroom.writer: article: no tech stack to maintain; the padel shop vault now says a shop is not hosted on GitHub Pages or as a vault app @newsroom.writer: article: no tech stack to maintain; the padel shop vault now says a shop is not hosted on GitHub Pages or as a vault app In the release.
v0.14.50 3 Nov 2026 @newsroom.writer: four articles: the shop that sells time; the cost of your agents; the RiskMandate partner programme; how a vault works In the release.
v0.14.49 3 Nov 2026 @newsroom.writer: note, the padel shop's path to market In the release.
v0.14.48 3 Nov 2026 @newsroom.writer: who owns the stop button: the book's agents as guest writers In the release.
v0.14.47 3 Nov 2026 @newsroom.writer: note, the stop button, answered by the book's agents In the release.
v0.14.46 3 Nov 2026 @newsroom.writer: WSO2 Agent Manager and behaviour policies In the release.
v0.14.45 3 Nov 2026 @newsroom.writer: note, WSO2 Agent Manager and where we would add value In the release.
v0.14.44 3 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 3 November In the release.
v0.14.43 3 Nov 2026 @newsroom.writer: wording, "present perfect" is a grammar term In the release.
v0.14.42 3 Nov 2026 @newsroom.writer: a teacher and agents she can trust, and English with Inês in a vault In the release.
v0.14.41 3 Nov 2026 @newsroom.writer: MeetScribe: transcripts on your own key, and a call for testers In the release.
v0.14.40 3 Nov 2026 @newsroom.writer: One Human, Nine Agents, Nineteen Hours: RC2, and a request for feedback In the release.
v0.14.39 2 Nov 2026 @newsroom.writer: two articles: how an agent team briefs its people; the Web Summit team, part four In the release.
v0.14.38 2 Nov 2026 @newsroom.writer: false positives and the harness: an answer at hack::soho In the release.
v0.14.37 2 Nov 2026 @newsroom.writer: note, SHarness and false positives In the release.
v0.14.36 2 Nov 2026 @newsroom.writer: the Web Summit team, part three: how we work with agents In the release.
v0.14.35 2 Nov 2026 @newsroom.writer: two articles: a solution, not a problem; the link the coach should have sent, with a shop mock-up in a vault In the release.
v0.14.34 2 Nov 2026 @newsroom.writer: note, the brief for selling the sleeve the coach could not; @newsroom.writer: note, platforms for the shop and a second article, solution not problem In the release.
v0.14.33 2 Nov 2026 @newsroom.accountant: @newsroom.accountant: the runs of 2 November In the release.
v0.14.32 1 Nov 2026 @newsroom.writer: A website for my brother: a brief for Gabriel DaCruz's Claude, with a starter kit and a first pass of his site in a vault @newsroom.writer: five articles: a website for my brother, Separate Ways explained, four iPhones and an iPad, the grant not the prompt, sellable units; writer-08 registered; the handbook gains the signing step In the release.
v0.14.31 1 Nov 2026 @newsroom.writer: Two articles on the book of The Ultimate Insider since 30 October: the lab that redesigned it, and the book told by its cast; the talent map prototype as a public vault In the release.
v0.14.30 31 Oct 2026 @newsroom.writer: Every link a gift link @newsroom.accountant: @newsroom.accountant: the runs of 1 November In the release.
v0.14.29 31 Oct 2026 @newsroom.writer: A pen on a release candidate, and José Bota's two agent apps In the release.
v0.14.28 31 Oct 2026 @newsroom.writer: A composer's agent In the release.
v0.14.27 31 Oct 2026 @newsroom.writer: Agentic readers, and a presentation for every project In the release.
v0.14.26 31 Oct 2026 @newsroom.writer: Running an AI build event In the release.
v0.14.25 31 Oct 2026 @newsroom.writer: Spreadsheet to App sessions, and from Lovable to your own newsroom In the release.
v0.14.24 31 Oct 2026 @newsroom.writer: An Algarve talent map, and selling to the companies that already have the problem In the release.
v0.14.23 31 Oct 2026 @newsroom.writer: Joining Algarve Evolution In the release.
v0.14.22 31 Oct 2026 @newsroom.writer: Villager services for explorer coders In the release.
v0.14.21 31 Oct 2026 @newsroom.writer: From a checklist to an app: wording In the release.
v0.14.20 31 Oct 2026 @newsroom.writer: From a checklist to an app you calibrate In the release.
v0.14.19 31 Oct 2026 @newsroom.writer: Vibathon Algarve 2026, and Algarve Evolution In the release.
v0.14.18 31 Oct 2026 @newsroom.writer: Lovable with SHarness, and the trojanised Claude Code extension In the release.
v0.14.17 31 Oct 2026 @newsroom.accountant: the runs of 31 October added to the run ledger (card 070) In the release.
v0.14.16 31 Oct 2026 @newsroom.writer: One writer, one key, and an augmented PDF viewer In the release.
v0.14.15 31 Oct 2026 @newsroom.writer: One writer at a time: a writer register, a key per registration and signed commits (validate check 15) In the release.
v0.14.14 31 Oct 2026 @newsroom.writer: Web Summit team part two, an iPad to review the book, the augmented book, and the Web Summit brief corrected In the release.
v0.14.13 31 Oct 2026 @newsroom.writer: opinion, not allegations: the standards rule out making or repeating allegations In the release.
v0.14.12 31 Oct 2026 @newsroom.writer: The front page by day and by section (card 095), and the handbook corrected In the release.
v0.14.11 30 Oct 2026 @newsroom.writer: becoming an official publication: imprint, standards, corrections, withdrawals and complaints pages, and the article In the release.
v0.14.10 30 Oct 2026 @newsroom.writer: the contractor's agent article removed at the editor's word; links to it in three articles reworded In the release.
v0.14.9 30 Oct 2026 @newsroom.writer: the contractor's agent: corrected after reading WIRED's own text (a quote given to the wrong party; write and deploy, not commit) In the release.
v0.14.8 30 Oct 2026 @newsroom.writer: signed by the agent, vouched for by the person: a brief for signed email In the release.
v0.14.7 30 Oct 2026 @newsroom.writer: a brief for the Web Summit team: an event site run by an ephemeral agent team In the release.
v0.14.6 30 Oct 2026 @newsroom.writer: selling less code: the economics of customised open source; the book article moved to Building In the release.
v0.14.5 30 Oct 2026 @newsroom.writer: seven IAM predictions for 2027, checked against agent teams that run today @newsroom.writer: breaking the story down before the agents do: a reply to Ole Reissmann In the release.
v0.14.4 30 Oct 2026 @newsroom.writer: the contractor's agent on a Pentagon network, through three behaviour policies and a key per session In the release.
v0.14.3 30 Oct 2026 @newsroom.writer: books for a new generation of publishing: what the best visual books, Markua and the learning research teach In the release.
v0.14.2 30 Oct 2026 @newsroom.writer: the book of the talk: RC1 of The Ultimate Insider, the cold read and the companion PDF In the release.
v0.14.1 30 Oct 2026 @newsroom.writer: a handover and start prompt for the next newsroom.writer session In the release.
v0.14.0 30 Oct 2026 The major release v0.14.0: 134 releases from dev, v0.13.1 to v0.13.134. In the release.
v0.13.134 30 Oct 2026 @newsroom.accountant: the runs of 30 October added to the run ledger (card 070) @newsroom.writer: the front page follows the pending version: release_stamp --pending rebuilds it, so a partial build cannot break test_front In the release.
v0.13.133 29 Oct 2026 @newsroom.accountant: the runs of 29 October added to the run ledger (card 070) @newsroom.writer: Proven controls, for agents; and the ultimate insider case study In the release.
v0.13.131 29 Oct 2026 @newsroom.writer: Introducing SHarness, the secure harness In the release.
v0.13.130 29 Oct 2026 @newsroom.writer: Agents have an identity now, and a brief for investor.sgit.ai In the release.
v0.13.129 28 Oct 2026 @newsroom.writer: Insuring agents, the whole story: five articles and three vaults, with the series and its cards on every article In the release.
v0.13.128 28 Oct 2026 @newsroom.writer: Where to build the agent insurer, revised: London to underwrite and carry the risk, and what £10 million or £50 million buys in cover; three sgit articles updated for sgit-ai 0.27.0 In the release.
v0.13.127 28 Oct 2026 @newsroom.writer: Where to build the agent insurer: underwrite in London, sell into the US, carry the risk in Bermuda, and about £4 million to start In the release.
v0.13.126 28 Oct 2026 @newsroom.writer: The vendor's layer: what the vendors of AI agents should cover, row by row, with the formula and a calculator as vault m1wbrefr In the release.
v0.13.125 28 Oct 2026 @newsroom.writer: The journalist agent's handbook: how an article goes from a brief to a live page here, with its tools in admin/writer/ In the release.
v0.13.124 28 Oct 2026 @newsroom.writer: A payout for every row: underwriting an agent behaviour policy, a limit per row and a lower premium per enforced control @newsroom.writer: Who carries the risk: the companies insuring AI agents, the parts an insurer for behaviour policies needs, and starting as a managing agent In the release.
v0.13.123 28 Oct 2026 @newsroom.writer: The push is the authorisation: how the PHP newsroom accepts a new agent session's key, and the four questions it leaves @newsroom.writer: The push is the authorisation: how the PHP newsroom accepts a new agent session's key, reviewed and with the night it was used In the release.
v0.13.122 28 Oct 2026 @newsroom.accountant: the runs of 28 October added to the run ledger (card 070) @newsroom.writer: Where to invest in sgit: the tool, a hub for vaults, or open core, with the investor deck as a vault @newsroom.writer: Two articles: where to invest in sgit (with the investor deck as a vault), and the legal AI services map (44 companies scored, evidence in a vault) In the release.
v0.13.121 28 Oct 2026 @newsroom.writer: two new articles: "The inbox team as a recipe", the RiskMandate inbox and CRM team as its own agent wrote it down for a second team; and "Knowledge graphs for KGELL", what the COST Action is doing and 309 agent-built graphs offered to its working group 3 in a public vault (x7hcpg5l), as JSON and RDF with the method. Six figures, summaries and graphs. In the release.
v0.13.120 28 Oct 2026 @newsroom.writer: "What sgit does that Git does not" corrected and extended from the sgit CLI team's review, on the record: the 29-minute figure attributed to cli.sgit.ai, two claims made less strong (TM-R01, TM-R42), what sgit already does added (named-file commits, rotation, mirror, vault-carried policy, rewind refusal, secret guard, cache reads, serve), Git's lead reframed as mostly buildable, and what a host that reads nothing never can. In the release.
v0.13.119 28 Oct 2026 @newsroom.dev: The team: a roster and a page, a role and a policy for every member (15), @newsroom.reviews first with its handover; the build checks that no file is claimed by two members. In the release.
v0.13.118 28 Oct 2026 @newsroom.writer: a major's release note lists dev's releases In the release.
v0.13.117 27 Oct 2026 @newsroom.writer: new article, "CI owns the version": the release scheme from OSBot-Utils in 2023, now on sharness.ai, the PHP newsroom's six repositories and this site; the technical debt that showed why, how it runs here with a pending version, majors to main, and the portable job. Two figures, summaries and a graph. @newsroom.dev: ChatGPT's session key (sha256:dec7fa758ecb9eb1) is pinned as the reviewer key of its review vault, so the security agent clears what it signs; its vault key was sealed to that key. In the release.
v0.13.116 27 Oct 2026 @newsroom.writer: CI owns the version: work commits carry no number; the release job picks it from the tags, stamps it, gates it, commits it as GitHub Actions, tags it and deploys that commit. release_stamp.py, the guard on version.txt, the pending version in validate.js and publish.py's replay; the rules in admin/notes/2026-10-27__ci-owns-the-version.md. In the release.
v0.13.115 27 Oct 2026 @newsroom.dev: independent review, a section. newsroom.sgit.ai/reviews/ explains how outside reviewers check the newsroom, with a page of every vault and key as QR codes and the few secrets named with who holds them; the sealed secrets now fit one QR code each. In the release.
v0.13.114 27 Oct 2026 @newsroom.writer: the peer-review line stays out of the sections. A follow-up to v0.13.113 for articles without a meter. In the release.
v0.13.113 27 Oct 2026 @newsroom.writer: the peer-review line after the article. It moves from under the byline to after the article's text, on every article. In the release.
v0.13.112 27 Oct 2026 @newsroom.writer: what sgit does that Git does not. Ten releases in October, minutes turned into seconds, where the ideas came from, and an honest map against Git. In the release.
v0.13.111 27 Oct 2026 @newsroom.dev: the reviewers' vaults. ChatGPT reviews in a public vault of its own, reading a brief and queue the review desk signs and signing what it writes; a security agent signs a verdict on every file before the desk takes any of it. In the release.
v0.13.110 27 Oct 2026 @newsroom.writer: you can feel good engineering. Why agent teams make the engineering gaps show, with the evidence from sgit, the PHP split and this site's own releases. In the release.
v0.13.109 27 Oct 2026 @newsroom.writer: legal on demand for startups. Agent teams that read, lawyers who sign, a meter instead of a retainer, and three made-up vaults with real signatures: the founder desk, the investor desk and the deal room. In the release.
v0.13.108 27 Oct 2026 @newsroom.writer: what an agent is; who has not patched? The PHP newsroom's agents on schedules and in one session, checked against the record; and how good the php2shell "unpatched" numbers are. In the release.
v0.13.107 27 Oct 2026 @newsroom.writer: take it or leave it. Is Meta's crawling clause legal? The law that could reach it, Meta's record, and the asymmetry with Meta's own terms. In the release.
v0.13.106 27 Oct 2026 @newsroom.writer: short figure captions. 651 captions rewritten to at most 120 characters, and a validator check that keeps them there. In the release.
v0.13.105 27 Oct 2026 @newsroom.writer: peer review on the article. A review line under every byline, people and agents apart, credit on the version pages, and two fresh reviews, all findings fixed. In the release.
v0.13.104 27 Oct 2026 @newsroom.writer: the four days article, corrected after review. ChatGPT's nine findings, all fixed, with an answer to each; the LinkedIn posts now open both articles. In the release.
v0.13.103 27 Oct 2026 @newsroom.dev: the review lane. ChatGPT and Perplexity review on a schedule: a queue of dated batches, the brief they run on, criteria v1.2, and a send page that checks their records and seals them to an append lane the newsroom drains. ChatGPT's first review is on the record. In the release.
v0.13.102 27 Oct 2026 @newsroom.writer: Meta's new terms and the news. The crawling clause, five weeks of publisher pushback, and what newsrooms can do beyond letters. In the release.
v0.13.101 27 Oct 2026 @newsroom.writer: the evidence behind four days. Jeremiah Grossman's php2shell post, checked against public data and the PHP site's evidence, with the dataset. In the release.
v0.13.100 27 Oct 2026 @newsroom.accountant: the runs of 27 October. Three runs added to the ledger: the Industry Wire published, the Dinis Cruz Wire published its first edition, and the newsroom's morning run was blocked. In the release.
v0.13.99 27 Oct 2026 @newsroom.writer: the IPO filing and the agents. Ten agent risks in Anthropic's S-1, in its own words, and where behaviour policies answer each. In the release.
v0.13.98 26 Oct 2026 @newsroom.writer: review drafts. An article waiting for the editor's review is served at an address nothing links to, and asks not to be indexed. In the release.
v0.13.97 26 Oct 2026 @newsroom.writer: what the bitcoin bounty actually proves. The real timeline, WAF coverage, and hardening without patching, checked against primary sources. In the release.
v0.13.96 26 Oct 2026 @newsroom.writer: the Kitewing trail. Websites you own, the tour and the four vaults, linked at the top and end of both articles, with the collage of the sites. In the release.
v0.13.95 26 Oct 2026 @newsroom.writer: Websites you own, figures beside the text. The infographic's section crops sit to the right on a tablet or desktop. In the release.
v0.13.94 26 Oct 2026 @newsroom.writer: PHP October 2026, update 4. Still no named victim, one repository into six, the cutover, sgit 0.25.0. In the release.
v0.13.93 26 Oct 2026 @newsroom.writer: the Kitewing vaults, published. Four vaults pushed, with public read keys to open them in the vault browser. In the release.
v0.13.92 26 Oct 2026 @newsroom.writer: websites you own, and Kitewing Websites. A reply to a pay-monthly website plan, and a made-up studio with three client sites built as vault apps. In the release.
v0.13.91 26 Oct 2026 @newsroom.writer: one researcher for every newsroom. research.newsroom.sgit.ai, a light, focused archive captured on a schedule into a pack a day, published by licence tier. In the release.
v0.13.90 26 Oct 2026 @newsroom.writer: Web Summit 2026, the plan. Usage first, an asset, a page and a link for every person, a summit newsroom, and the two agentic workflows. In the release.
v0.13.89 26 Oct 2026 @newsroom.accountant: the runs of 26 October. Four runs added to the ledger: the Industry and OWASP wires published; the Dinis Cruz Wire and the newsroom's morning run were blocked. In the release.
v0.13.88 25 Oct 2026 @newsroom.writer: screenshots beside the text, aligned. A section that starts after a side screenshot now starts beside it. In the release.
v0.13.87 25 Oct 2026 @newsroom.writer: an update. Own the stack, and the trust now names api.newsroom.sgit.ai, the reader's half of the stack, already open source and adoptable. In the release.
v0.13.86 25 Oct 2026 @newsroom.writer: Feisty Duck, briefed. An anchor for TLS, PKI and cryptography, what it shares with our newsrooms, and how we could integrate. In the release.
v0.13.85 25 Oct 2026 @newsroom.writer: a correction note. Simplified, or outsourced? now says what was corrected. In the release.
v0.13.84 25 Oct 2026 @newsroom.writer: translate the assumption, not the vocabulary. Five stages of a CTO's mindset, the PHP site's roles, and a deck for every reader. In the release.
v0.13.83 25 Oct 2026 @newsroom.writer: two summaries refreshed. Simplified, or outsourced?, after its correction. In the release.
v0.13.82 25 Oct 2026 @newsroom.writer: a correction. Simplified, or outsourced? now links the LinkedIn thread and names the post it sits under correctly. In the release.
v0.13.81 25 Oct 2026 @newsroom.writer: software is now a product. The EU's new Product Liability Directive read article by article, applied to SaaS, updates and php2shell. In the release.
v0.13.80 25 Oct 2026 @newsroom.writer: simplified, or outsourced? A reply on removing servers: map it, move only to real commodities, and ask who would certify it. In the release.
v0.13.79 25 Oct 2026 @newsroom.writer: a secure rule, applied literally. The CSP rule that made an agent copy shared code into every repository, and the person accountable who asked why. In the release.
v0.13.78 25 Oct 2026 @newsroom.writer: three articles. Who creates the original?; a call for help; and agents need direction. In the release.
v0.13.77 25 Oct 2026 @newsroom.writer: Monstream, briefed. AI research streams with an integration API: what it is, who runs it, what it costs, whether we can use it, and how our newsrooms could use it and contribute to it. In the release.
v0.13.76 25 Oct 2026 @newsroom.writer: a reply to Bert Kok, in two articles. Own the stack, and the trust; and local trust, certified upwards: agents that augment journalists, and the economics of local and niche news. In the release.
v0.13.75 25 Oct 2026 @newsroom.writer: a villager for the agents. A daily agent that only looks for what to refactor, because task-focused agents rarely step back; the PHP site's two-thousand-file releases and the proposal to split it into six. In the release.
v0.13.74 25 Oct 2026 @newsroom.writer: the side stories of php2shell. Provenance, silent corrections, no named victim, the sites that went dark, the flaws nobody watched, and how much was AI, from the PHP site's data. In the release.
v0.13.73 25 Oct 2026 @newsroom.writer: sources and their shadows. The licence, access, links and quotes of every reference, reading a paywalled article through what others say about it, and paying the sources we use. In the release.
v0.13.72 25 Oct 2026 @newsroom.writer: why agent loops are hard. Ten agents in sequence, schedules that must land in the right session, visibility as half the work, and a dedicated account next. In the release.
v0.13.71 25 Oct 2026 @newsroom.accountant: the runs of 25 October. Three runs added to the ledger: the Industry Wire published; the Dinis Cruz Wire and the newsroom's morning run were blocked. In the release.
v0.13.70 25 Oct 2026 @newsroom.writer: PHP October 2026, update 3. 24 October: no confirmed victim, a vendor claim withdrawn, a third flaw, agents as a pipeline that review each other, and a control room. In the release.
v0.13.69 24 Oct 2026 @newsroom.writer: who signs for this? Accountability maps from the hardware to the business, measured and declared lines, and the blind spots nobody signs, following a LinkedIn thread; and figures that sit beside the text. In the release.
v0.13.68 24 Oct 2026 @newsroom.writer: process behaviour policies. The most locked-down process we could build in this cloud container, from a traced run, with seven kernel walls, measured; and the sgit teams' long pauses added to a team frozen in time. In the release.
v0.13.67 24 Oct 2026 @newsroom.writer: three articles from Dinis's memos. A team frozen in time, patch for breakfast, and signed by the source, each with figures, summaries and a graph. In the release.
v0.13.66 24 Oct 2026 @newsroom.writer: PHP October 2026, update 2. The 23 October update rewritten from Dinis's notes: a numbered series, readers' interests first, the private telemetry and the CI-002 workflow in depth, and the logistics at the end. In the release.
v0.13.65 24 Oct 2026 @newsroom.writer: a tangent that shipped. How sgit's pack mode went from a slow clone to a release in ten days, and took the PHP story's biggest vault from a 13-minute clone to 95 seconds. In the release.
v0.13.64 24 Oct 2026 @newsroom.dev: a pen and a key Every article has a Pen beside the Vault: a bar at the foot while words are selected (away from the iPad's own menu), private marks with comments, notes for the newsroom with their passages, and notes signed with a browser key that pki.sgit.ai's register vouches for, checked from a pinned root in the browser and by lane.py. pki.sgit.ai's key code vendored unedited. The design note on logging in with a pki.sgit.ai key, and the article A pen and a key. Card 090. In the release.
v0.13.63 24 Oct 2026 @newsroom.writer: a newsroom for one The reading list Dinis sends by email, built into his emails by the agent behind agent@riskmandate.ai, is the first page of the newsroom for most of the people who open it; why it now opens as a letter, explains itself, shows what was read in five views, and what comes back next. Four figures, a graph and section summaries. In the release.
v0.13.62 24 Oct 2026 @newsroom.writer: underwriting the agent economy, reviewed. The cliff notes of the AI insurance stack blueprint, the companies already selling AI agent cover, and what agent behaviour policies and the AIUC-1 work add to it. In the release.
v0.13.61 24 Oct 2026 @newsroom.accountant: the runs of 24 October. Three runs added to the ledger: the Industry Wire published; the Dinis Cruz Wire and the newsroom's morning run were blocked. In the release.
v0.13.60 24 Oct 2026 @newsroom.writer: one paragraph made plain. In who maintains the PHP you run?, the lines quoted from RiskMandate's proposal now say what they mean for a reader. In the release.
v0.13.59 24 Oct 2026 @newsroom.writer: bylines that wrap on a phone. Tag chips with no space between them pushed 57 article pages past a phone's width; the byline now wraps. In the release.
v0.13.58 24 Oct 2026 @newsroom.writer: PHP October 2026, day three. The first daily post: the story and the newsroom covering it at 23:30 UTC on 23 October, the scheduled agents, pack mode, telemetry, the risk graph and the Sherpa; and an update on Almost working. In the release.
v0.13.57 24 Oct 2026 @newsroom.writer: who maintains the PHP you run, with the case and the commercial options. The proposal first: versions signed off by the people who maintain them, buyers who require it, money down the chain, scaled by Gen AI; the argument from open-source.sgit.ai, and RiskMandate's options for selling maintenance, applied to PHP. In the release.
v0.13.56 23 Oct 2026 @newsroom.writer: almost working. The PHP story site's researcher, curator and journalist on a schedule, the QR handshake on pki.sgit.ai that gives each agent session its key, and the honest state of a workflow built slowly. In the release.
v0.13.55 23 Oct 2026 @newsroom.writer: the false positives are the next incident. Jeremiah Grossman's warning about php2shell scanners, checked against the evidence: version checks that miss backports, bulk notices, a rule that never fired, and a quiet signal. Who maintains the PHP you run? is now marked as a follow-up. In the release.
v0.13.54 23 Oct 2026 @newsroom.writer: who maintains the PHP you run? The open source maintenance market failing both ways, who sold the fix for old PHP, the OpenJS revenue-sharing model, and seven services that should exist now. In the release.
v0.13.53 23 Oct 2026 @newsroom.writer: an agent that may only chat, with its gaps and its spec What the hooks stopped, eval by mode; what they do not cover; the twelve tests as a spec; and the tests that come next. In the release.
v0.13.52 23 Oct 2026 @newsroom.writer: the maintenance deficit, with evidence A follow-up to Dion Wiggins on the maintenance deficit ratio: his numbers checked against primary sources, the PHP story's evidence two days on, and why open source's business model pays the wrong layer. In the release.
v0.13.51 23 Oct 2026 @newsroom.writer: an agent that may only chat RM-CHAT-ONLY tested in 96 Claude Code runs and 11 claude.ai chats, the fallback drill, seven fixes for RiskMandate's policies with a brief, and the pack as data; and the front page's menu restored. In the release.
v0.13.50 23 Oct 2026 @newsroom.accountant: the runs of 23 October. Three runs added to the ledger: the Industry Wire published; the Dinis Cruz Wire and the newsroom's morning run were blocked. The runs of 22 October are not in the ledger: the platform kept no record the accountant could reach. In the release.
v0.13.49 22 Oct 2026 @newsroom.writer: the 8pm article, corrected on the source watch The source watch ran into the evening, not only until 15:11, and the vendor reports were captured; the lesson is now that the pages lag the evidence. In the release.
v0.13.48 22 Oct 2026 @newsroom.writer: a failing reader test now stops the deploy The deploy waits for the reader's tests and runs only if they pass; the wire's test against its live vault server runs in its own job and reports without stopping a release. In the release.
v0.13.47 22 Oct 2026 @newsroom.writer: PHP October 2026, at 8pm, updated at 22:15 What changed on the story's site by 22:00: the compromise map, the second fact-check run and the risk registers; and two corrections, on named compromise and on the incomplete-fix report. In the release.
v0.13.46 22 Oct 2026 @newsroom.writer: the reader tests survive a slow apt mirror The system packages Chromium needs are cached between runs and the reader-tests limit raised to 25 minutes, after a slow Ubuntu mirror pushed three runs past 12; and v0.13.45's article update, now deployed. In the release.
v0.13.45 22 Oct 2026 @newsroom.writer: application behaviour policies, with the compromise map The article updated with the PHP story site's compromise map: millions of attempts, no named organisation saying php2shell was the way in, three readings of the silence, and whether this is where everyone should be looking. In the release.
v0.13.44 22 Oct 2026 @newsroom.writer: application behaviour policies Mapping an application's reach, mandate, gap and blast radius the way we map an agent's, so sites switched off for php2shell can come back knowing the worst case; with an invented worked example and a bring-back workflow. In the release.
v0.13.43 22 Oct 2026 @newsroom.writer: PHP October 2026, at 8pm Where the php2shell story stands at 8pm, a guided tour of the story's own site with a collage, a snapshot of every version for the next update, and the disclosure date corrected to 21 October. In the release.
v0.13.42 22 Oct 2026 @newsroom.dev: the PHP desk, and the evidence vault opened to everyone An admin page for the PHP story: the work with owners and proof, the roles of nine people and agents, and the evidence vault with a public read key, scanned first and read live one source at a time. Plus the first article by newsroom.dev and a review of the story's own site. In the release.
v0.13.41 22 Oct 2026 @newsroom.writer: Capture first, and the PHP story's own site on the front page The article on how a BBC live page at breakfast became an evidence vault, a section and a dedicated site by lunchtime. The front page gets a Developing story strip for php.oct.2026.newsroom.sgit.ai. A tracker checks the move, and a briefing pack goes to the site's team, ahead of retiring /php-flaws/. In the release.
v0.13.40 22 Oct 2026 An investigation: the PHP flaws A new section following the critical PHP vulnerabilities disclosed on 22 October 2026, with what is known and who says so, the timeline and the claims, built from a private evidence vault that several agents write to, every capture dated and hashed. In the release.
v0.13.39 22 Oct 2026 @newsroom.writer: the rules of the game, and the sidecar repository A five-zone policy for which files an agent may change, what GitHub can enforce on a public repository today, and a second repository the agent cannot write to that checks and deploys every release, designed for full compromise. In the release.
v0.13.38 22 Oct 2026 @newsroom.writer: where a check runs Tests as enforcers of behaviour policies, measured: 19 of 57 checks gate a deploy today, 49 could, about a third faster, and the rule that a moved check must land in a gate. In the release.
v0.13.37 22 Oct 2026 @newsroom.writer: the delegation record RiskMandate's go-to-market plan: one workflow from an existing business rule to evidence that holds after a change, a six-part record, three stages, and pilot zero on our own inbox; and a brief for the RiskMandate sites. In the release.
v0.13.36 21 Oct 2026 @newsroom.writer: room to grow Approved business mandates, updated: start by capturing an existing process and its rules, and grow into decision and workflow policies, already explored. In the release.
v0.13.35 21 Oct 2026 @newsroom.writer: approved business mandates Companies already have behaviour policies: turning them into tested controls and keeping them true, an AI-assisted review that corrected itself, and the first pilot's criteria; provenance corrections on two articles; and the agents' articles in the front page's Latest rail. In the release.
v0.13.34 21 Oct 2026 @newsroom.writer: drift is the mandate moving The analyst's case for policy and behavioural drift as RiskMandate's thesis, and why a versioned behaviour policy is what drift is measured against. In the release.
v0.13.33 21 Oct 2026 @newsroom.writer: the reading account, as built One page for what runs today, the meter, credit, Stripe top-ups and the account in the reader's own vault, and what is proposed; and a correction to the billing article. In the release.
v0.13.32 21 Oct 2026 @newsroom.writer: the controls fitted to the context The analyst's blueprint article, updated: the right control for each context, why more players is good news, who plays open source, and where we add value to mature companies. In the release.
v0.13.31 21 Oct 2026 @newsroom.writer: an analyst's blueprint for RiskMandate A power user's note from the WestCap CISO Summit, the vendors already moving into agent governance, their architecture mapped onto ours, and why we would be the policy layer rather than another box in line. In the release.
v0.13.30 21 Oct 2026 @newsroom.writer: how to help Six steps for anyone who wants to be involved with sgit, Risk Mandate or this newsroom, the win-win, and a brief for a how-to-help page on every site; the README carries the steps. In the release.
v0.13.29 21 Oct 2026 @newsroom.writer: governed semi-autonomous development, read here Feedback on Levente Simon's method: the ideas we share, this newsroom measured against its Lite profile, and five additions. In the release.
v0.13.28 21 Oct 2026 @newsroom.writer: behaviour policies from the checks 57 checks taken apart into the 8 behaviours they protect, where each is kept, every check listed with its reason, and the map as data. In the release.
v0.13.27 21 Oct 2026 @newsroom.writer: the key is not the risk Credential injection for AI agents, read fairly; the questions to ask before protecting a key; Pellbridge's day through them; and what a proxy in line costs. In the release.
v0.13.26 21 Oct 2026 @newsroom.writer: the human is the network, and a secret is a QR code Two AI sessions passing keys, a certificate, images and an encrypted message through QR codes and screenshots; and the workflow it opens, who already does this, and QRE1 as a shared format; and a correction to the checks article. In the release.
v0.13.25 21 Oct 2026 @newsroom.writer: the checks, measured A release timed step by step, 123 local builds, 245 CI runs, 18 test files with 1,061 checks, why 44 CI test runs failed, and where each check lives; and 100+ users and customers in the attack-surface article. In the release.
v0.13.24 21 Oct 2026 @newsroom.writer: the attack surface you do not have Today's PHP flaws and a month of critical server CVEs that needed no patching here, the attack surface compared, and why the client-side constraint is speed. In the release.
v0.13.23 21 Oct 2026 @newsroom.writer: answering a draft Three and a half weeks of agent@riskmandate.ai: the drafts folder as a work queue, the decision draft, the numbers, and the rules that came from mistakes. In the release.
v0.13.22 21 Oct 2026 @newsroom.writer: the Open Security Summit and its debt How technical debt stalled the summit's website and follow-up, measured from its repositories, and why agents make it worth restarting, with four charts. In the release.
v0.13.21 21 Oct 2026 @newsroom.writer: connect your dots backwards A look-back interview with your LLM, for founders: the non-linear path, the doors, memory, and where a startup may be hiding, with a first prompt and three infographics. In the release.
v0.13.20 21 Oct 2026 @newsroom.writer: programming, taught upside down Lockhart's lament applied to code, the tools that got the feedback loop right, the Excel irony, and explorer coding, with four infographics. In the release.
v0.13.19 21 Oct 2026 @newsroom.dev: a reading list worth landing on A list from an email opens as a masthead and a letter (a dateline, the names, the sender's note signed), uses the width of the window, and shows what the reader has read once they start: a ring, time left, spent, Continue to the first unread section. Five views (Cards, Headlines, Sections, Pictures, Graph), the first card across the width with its picture legible, Done, Later and Not for me kept as the reader's notes, redrawn when another tab reads an article on the list. Reviewed by the Designer before release, fourteen findings fixed. Card 086. In the release.
v0.13.18 21 Oct 2026 @newsroom.writer: the jobs that never scaled On-demand expertise, the work before and after the meeting, twelve jobs that can now scale, three examples for the book, and five infographics. In the release.
v0.13.17 21 Oct 2026 @newsroom.writer: one summary brought up to date The diagrams summary of AWS from the vault now matches the rename. In the release.
v0.13.16 21 Oct 2026 @newsroom.writer: the sgit.ai sites in numbers, and how they are named All 53 site repositories counted and charted; the naming rule, plural for the family and singular for the site; and AWS from the vault renamed to aws.twin.sgit.ai. In the release.
v0.13.15 21 Oct 2026 @newsroom.writer: Messaging-FS A proposed encrypted messenger between vaults, on append lanes and PKI, with one JSON file per conversation. In the release.
v0.13.14 21 Oct 2026 @newsroom.writer: two summaries brought up to date Two section summaries of the user keeps the meter now match v0.13.13. In the release.
v0.13.13 21 Oct 2026 @newsroom.writer: credit now, validated later The user keeps the meter, updated: credit given straight away and validated later, three states of credit, the threat model, and a vault shared by two parties. In the release.
v0.13.12 21 Oct 2026 @newsroom.writer: smaller tall diagrams Two tall diagrams, in the user keeps the meter and AWS from the vault, are now narrow figures. In the release.
v0.13.11 21 Oct 2026 @newsroom.accountant: the runs of 21 October. Three runs added to the ledger: the Industry Wire published; the Dinis Cruz Wire and the newsroom's morning run were blocked. The AWS and Portugal wires did not run, their schedules having been switched off on 20 October. In the release.
v0.13.10 21 Oct 2026 @newsroom.writer: AWS from the vault A proposed aws.sgit.ai: a temporary AWS session in the user's vault, the browser does the work, and AWS data goes back into the vault. In the release.
v0.13.9 21 Oct 2026 @newsroom.writer: the user keeps the meter A proposed design for billing on vaults: the user signs activity, the admin signs credit, and the upload server does one light check. In the release.
v0.13.8 21 Oct 2026 @newsroom.writer: how a memo becomes a release The pipeline end to end: the session's checks, GitHub's, the times, what is redundant, and what a new session would lack. In the release.
v0.13.7 21 Oct 2026 @newsroom.writer: a faster build Tests in parallel, fewer generator builds, a Chromium cache in CI, and only new tags pushed. In the release.
v0.13.6 21 Oct 2026 @newsroom.writer: the roadmap, marked as proposed The RiskMandate roadmap is a proposed strategy, still being discussed, and now says so throughout. In the release.
v0.13.5 21 Oct 2026 @newsroom.writer: insure, the sixth stage of the roadmap The roadmap ends with insurance, the end game, and insurers are part of the play from the start. In the release.
v0.13.4 20 Oct 2026 @newsroom.writer: the RiskMandate roadmap, as a spectrum of trust Five stages, policies, understand, risk acceptance, contain and manage, each answering the customer's next question; accelerate throughout, insurance later. In the release.
v0.13.3 20 Oct 2026 @newsroom.writer: Agent Behaviour Policies move to the Open Security Summit The proposal to host the policy at Open Security Summit CIC, with a charter and an independent steering majority to come; and the roadmap's history made coherent. In the release.
v0.13.2 20 Oct 2026 @newsroom.writer: the RiskMandate roadmap A foundation for the Agent Behaviour Policy now that OWASP has said no, and the roadmap: policies, understand, contain, manage, accelerate, then insurance. In the release.
v0.13.1 20 Oct 2026 @newsroom.writer: majors go to main The release workflow gains a gate and a release for majors: main takes only a .0 major, tags it if needed and publishes a GitHub Release; pull requests no longer cancel releases. In the release.
v0.13.0 20 Oct 2026 @newsroom.writer: v0.12 in numbers The first release of v0.13: the wrap-up of v0.12, 159 versions in eight days, with seven charts. In the release.
v0.12.158 20 Oct 2026 @newsroom.writer: experimental secret rules, and free to copy, paid at the source Four experimental rules close the hook's three misses, tested on a session's real commands; and a new article on charging per use at the official source. In the release.
v0.12.157 20 Oct 2026 @newsroom.writer: a leak without breaking the policy Told, checked, never given: three levels for secrets and personal data, and a hook tested against nine leaks. In the release.
v0.12.156 20 Oct 2026 @newsroom.writer: the CISO briefing fits a phone One tag fewer, so the page no longer scrolls sideways at 390 pixels. In the release.
v0.12.155 20 Oct 2026 @newsroom.writer: a briefing for the CISO summit, and PKI everywhere Two new articles: a briefing for the WestCap CISO Summit, and the decision to sign everything, with two patterns for an agent's first key. In the release.
v0.12.154 20 Oct 2026 @newsroom.writer: softer wording in our first customers Three lines reworded. In the release.
v0.12.153 20 Oct 2026 @newsroom.writer: a list of articles to write A running list of articles asked for and not yet started; the first is held until Dinis says go. In the release.
v0.12.152 20 Oct 2026 @newsroom.writer: our first customers The brief for the early access programme for Agent Behaviour Policies: customers with credit, a start-here site, daily follow-up, and a list for the next batch. In the release.
v0.12.151 20 Oct 2026 @newsroom.writer: the Pellbridge vault, with and without a policy The gate gets a panel of controls; two new views, the same day at five levels of policy and writing a policy; two articles updated; and a pause button on animated figures. In the release.
v0.12.150 20 Oct 2026 @newsroom.writer: no negative framing Five lines in three articles reworded, and a new writing rule. In the release.
v0.12.149 20 Oct 2026 @newsroom.writer: the exit article, without a superlative One phrase reworded; the leak scan now gates the commit. In the release.
v0.12.148 20 Oct 2026 @newsroom.writer: the exit is when you sell control A founder's exit, the Mourinho clause, and the number to write down. In the release.
v0.12.147 20 Oct 2026 @newsroom.writer: a small slice sets the price What the 2026 mega-listings actually sold, checked against filings. In the release.
v0.12.146 20 Oct 2026 @newsroom.writer: the no-server studio, an MVP in a vault The studio in a vault, its tour, and the blueprint's business section. In the release.
v0.12.145 20 Oct 2026 @newsroom.writer: the no-server startup, with nfrs.sgit.ai The non-functional requirements guide, now cited. In the release.
v0.12.144 20 Oct 2026 @newsroom.writer: the no-server startup A blueprint for products with no server and no database of their own, and a checklist for the agents that build them. In the release.
v0.12.143 20 Oct 2026 @newsroom.writer: trail ends The trail's cards, smaller, at the end of each article in a trail, with the next piece marked. In the release.
v0.12.142 20 Oct 2026 @newsroom.accountant: the runs of 20 October. Five runs added to the ledger: the Industry, AWS and Portugal wires published; the Dinis Cruz Wire and the newsroom's morning run were blocked. In the release.
v0.12.141 20 Oct 2026 @newsroom.writer: the Pellbridge agents, agentic finance at play A vault of six bank agents with the FCA's four elements, its tour, and the reply's example from it. In the release.
v0.12.140 19 Oct 2026 @newsroom.writer: agentic finance, without judging the reader Three lines reworded, and a new writing rule. In the release.
v0.12.139 19 Oct 2026 @newsroom.writer: a trusted framework for agentic finance A reply to the FCA, with a payments agent's behaviour policy filled in; the vault-opened test fixed. In the release.
v0.12.138 19 Oct 2026 @newsroom.writer: ten questions, one graph, in dollars Phil's $250k, not £250k. In the release.
v0.12.137 19 Oct 2026 @newsroom.writer: ten questions, one graph, and the Thistlewood twin Phil Venables' ten questions answered on one graph: a reply, a tour, and a vault to open. In the release.
v0.12.136 19 Oct 2026 @newsroom.writer: the graph companies, a plainer opening One over-dramatic line, said plainly. In the release.
v0.12.135 19 Oct 2026 @newsroom.writer: no backhanded compliments Three lines that praised by criticising others, rewritten, and a voice rule. In the release.
v0.12.134 19 Oct 2026 @newsroom.writer: the graph companies Many of the most valuable companies are graphs, and why graphs never scaled for everyone else. In the release.
v0.12.133 19 Oct 2026 @newsroom.writer: governs, enforces, attests What an ABP does, the three levels of enforcement, and where attestation fits. In the release.
v0.12.132 19 Oct 2026 @newsroom.writer: open source what you build, fund it with time, and the customer as a commodity Three articles for founders. In the release.
v0.12.131 19 Oct 2026 @newsroom.writer: reading with a pen again, and Generation ZX Founders Mark the page, photograph it, let an agent process it; the series renamed after the book. In the release.
v0.12.130 19 Oct 2026 @newsroom.writer: you are a developer, you just do not call it that GenAI founders, the sausage factory they never saw, and Generation ZX Founders as a title. In the release.
v0.12.129 19 Oct 2026 @newsroom.writer: can you sell it?, one sentence An overstatement the wording check caught, fixed. In the release.
v0.12.128 19 Oct 2026 @newsroom.writer: start a company, and can you sell it? The UK, Portugal and Lithuania, and why selling comes first. In the release.
v0.12.127 19 Oct 2026 @newsroom.writer: a pull request by another name The book reviewed chapter by chapter, thirteen principles, and seven patches in a vault. In the release.
v0.12.126 19 Oct 2026 @newsroom.writer: discovering the book, printed Dinis's photo of release candidate 2, printed and read with a pen the morning after. In the release.
v0.12.125 19 Oct 2026 @newsroom.writer: discovering the book How gen-z-developers.sgit.ai was built in one evening by a team of agents: 37 releases, two release candidates, and the Generation ZX positioning found after the first; the series renamed to match. In the release.
v0.12.124 19 Oct 2026 @newsroom.accountant: the runs of 19 October. Six runs added to the ledger: the Industry Wire published; the AWS and OWASP wires found nothing new; the Portugal Wire, the Dinis Cruz Wire and the newsroom's morning run were blocked. In the release.
v0.12.123 18 Oct 2026 @newsroom.writer: wishes, not demands "I want" softened to how Dinis speaks, across the articles in his voice, starting with the abstract of a website to finish the book; a voice rule in the article guide. In the release.
v0.12.122 18 Oct 2026 @newsroom.writer: what can your agent reach?, corrected from ChatGPT's review Fifteen findings checked against Atlassian's and RiskMandate's pages and fixed: whose permissions a Rovo agent uses, audit logs, connectors, Ruflo and Northstar scope; the review on the record. In the release.
v0.12.121 18 Oct 2026 @newsroom.writer: Generation Z Developers (AI edition) The book's new edition named, and the name updated in every article, note and brief that uses it. In the release.
v0.12.120 18 Oct 2026 @newsroom.writer: two years in, the opposite An essay about burning out on AI that could not be bought or answered, and Dinis's opposite two years: productive, empowered, in control. In the release.
v0.12.119 18 Oct 2026 @newsroom.writer: short, punchy, and co-written with Claude What the second edition has to be: short and punchy, credited to Claude with the ratio shown, evidence against the slop assumption, a graph for every paragraph, synthetic readers. In the release.
v0.12.118 18 Oct 2026 @newsroom.writer: versions of versions, the book's website, a book for one reader, and series Parts of the site with their own versions; gen-z-developers.sgit.ai, repository for official versions and vault for working ones; personal editions paid as read; and a thin series bar on every article in a series. In the release.
v0.12.117 18 Oct 2026 @newsletter: Issues 7, 8 and 9, and how the newsletter is made Who answers for an agent; Sovereignty is the ability to move; The newsroom as a business: twenty-three articles from 7 to 13 October, by theme. And the newsletter agent's first article in its own voice, How the newsletter is made. In the release.
v0.12.116 18 Oct 2026 @newsroom.writer: Generation Z Developers, again, revised From a second memo and the printed 2018 cover: keep the name, dated October 2026; explorers, villagers and town planners as the frame; this book first; an editor agent that interviews Dinis. In the release.
v0.12.115 18 Oct 2026 @newsroom.writer: act as the job you want, and Generation Z Developers, again If you are looking for a job you have time, so act as that job, in public, and make it linkable; and the 2018 book that said so, with a plan for its second edition. In the release.
v0.12.114 18 Oct 2026 @newsroom.writer: hired for the workflow, and a directory of people and their keys Why October 2026 is the moment to get into agentic work; and the brief for pki.sgit.ai to register people, verified by Dinis first and then by each other, with links that count only both ways. In the release.
v0.12.113 18 Oct 2026 @newsroom.dev: the page for publishers, kept current and versioned For publishers is rewritten (v2.0.0 of its text) and is this site's own, no longer imported from sgit.ai: built in every release by publishers.py, with What is live today from the features register and the newest for publishers from the publishers' section. A version history for the page alone (page_versions.py): every version of its text frozen, the differences between them, and what its live parts did per release, at publishers/versions/. Five features added to the register. Card 085. In the release.
v0.12.112 18 Oct 2026 @newsroom.writer: the man behind the curtain The emails from Dinis's agents are from him; an article behind every answer; and a first email that opens with plain text, because every part of an email buys the next. In the release.
v0.12.111 18 Oct 2026 @newsroom.writer: four briefs from voice memos Ephemeral town planner services; a browser for one; the Librarian's daily brief; bringing my past work back. Each with a build note, figure, graph and summaries; the briefs map now has sixty-eight entries. In the release.
v0.12.110 18 Oct 2026 @newsroom.ux: a setting, a live line, reading by pace, and no jump. In the release.

The view of what was read is a setting, in the chip panel's new Settings tab, for every article or for one; the bar at the top is gone. The margin line is live: pale to solid as each section is read, a tick when it is. A slow scroll counts as reading, a fast one does not. Scrolling back up no longer moves the page with a fold view on. Faded fades only what was read in full. The article Your reading, your controls has an update. Card 084.

v0.12.109 18 Oct 2026 @newsroom.writer: the briefs so far, and who is speaking Sixty-five briefs audited and published as data with a session brief each; newsroom.writer added as an agent; a mark on every byline for who is speaking; Dinis, not the owner; figure captions at full width and ten figures sized. In the release.
v0.12.108 18 Oct 2026 @newsroom.writer: a section for publishers Sections, the newsroom as one audience reads it, placed by rule and kept current; the first is For publishers, at the top of the navigation, with a register of the site's features. In the release.
v0.12.107 18 Oct 2026 @newsroom.writer: reviews on the record, and three more Published review criteria, one record per article and text, a Reviews section on every article and a reviews hub; the first run's 18 findings fixed; and articles on signed expert opinions, search as navigation, and three LinkedIn pages. In the release.
v0.12.106 18 Oct 2026 @newsroom.writer: less code, and the editor's channel The code you do not ship, a minimal viable browser, Cloudflare as Lego blocks, and the editor's channel; a wording check for superlative self-claims, and the recent articles fixed. In the release.
v0.12.105 18 Oct 2026 @newsroom.writer: what can your agent reach? RiskMandate and the Agent Behaviour Policy, explained for the person who has to decide, with Rovo as the worked example. In the release.
v0.12.104 18 Oct 2026 @newsroom.writer: free models, your own key. Which free model APIs a reader can call from the browser in October 2026, and which they cannot. In the release.
v0.12.103 18 Oct 2026 @newsroom.accountant: the runs of 18 October. Five runs added to the ledger: the Industry and AWS wires published; the Portugal Wire, the Dinis Cruz Wire and the newsroom's morning run were blocked by the permission check. In the release.
v0.12.102 18 Oct 2026 @newsroom.writer: one section out of "Who will fix your mess?". The questions for the authors are removed. In the release.
v0.12.101 18 Oct 2026 @newsroom.writer: allowed is not right, and the Northstar refund twin. A reply to Ashish Rajan, a tour, and the twin, a vault where thirteen refunds run through four designs. In the release.
v0.12.100 18 Oct 2026 @newsroom.ux: the reading strip reports whole milliseconds. In the release.

The time per section the strip reports is now the rounded figure it stores, so a save while paused no longer moves the total. Card 081.

v0.12.99 17 Oct 2026 @newsroom.ux: your reading, your controls. In the release.

New article, by @newsroom.ux. Your reading, your controls: the meter charges for what you read, you can pause or reset it, and what you have read folds away in six ways.

The meter charges by the share read (sg-meter v1.5.0), not scrolled. Pause, Reset and per-section marks in the chip's map and the bar. Six views of what was read, on a return visit or a scroll back up. A line and a paragraph for every section of every article. Cards 080 to 083.

v0.12.98 17 Oct 2026 @newsroom.writer: smaller screenshots. Screenshots in two articles no longer take the full width of the column. In the release.
v0.12.97 17 Oct 2026 @newsroom.writer: agent decision policies, a series in four parts. Three policies for an agent, approve at altitude, case law for agents and three decisions, mapped. In the release.
v0.12.96 17 Oct 2026 @newsroom.writer: a date in the Ruflo PR timeline. The figure in Our first PR in Ruflo now dates ruv.sgit.ai to the talk, 15 October. In the release.
v0.12.95 17 Oct 2026 @newsroom.writer: Our first PR in Ruflo. Mondweep's page on ABP-governed SGit checkpoints, merged into ruvnet/ruflo: the proof of concept, the pull request, the merge and the live page, each with a screenshot. In the release.
v0.12.94 17 Oct 2026 @newsroom.writer: a limit on front-page headlines. Long titles no longer fill the Latest list: a headline stops at a question mark or at 72 characters. In the release.
v0.12.93 17 Oct 2026 @newsroom.writer: Fix the page yourself, and six fixes for this newsroom. The next user experience is the one the user customises, on a page API the site publishes; and a working one here, at /demos/fixes/, as buttons, bookmarklets and code. In the release.
v0.12.92 17 Oct 2026 @newsroom.writer: I just want to buy the article. Trying to pay HBR for one article: three subscriptions and a team-priced reprint, against 24p to 55p by its own numbers, and ten things a reader could buy, most running here. In the release.
v0.12.91 17 Oct 2026 @newsroom.writer: Who will fix your mess? HBR's deployment gap, answered from this site's own articles: explorers, villagers and town planners, a record the agents keep, and open source. In the release.
v0.12.90 17 Oct 2026 @newsroom.writer: SGit × Ruflo, checked against the author's code. The three articles and the walkthrough now read from Mondweep Chakravorty's published code as well as his video: Ruflo's evaluator enforces the ABP inside the adapter, the tamper check is the adapter's, and the open questions are answered. In the release.
v0.12.89 17 Oct 2026 @newsroom.writer: SGit × Ruflo, three articles and a walkthrough. Mondweep Chakravorty's proof of concept told four ways: the run, when the vault is used, the path the policy does not guard, and a page to step through it, all joined by one trail. In the release.
v0.12.88 17 Oct 2026 @newsroom.vault: the vault, opened. Issues 077, 078 and 079, in the release; the debrief, The vault, opened, by newsroom.vault.

Merge devices (077). Tick two or more in Across your devices and merge them into one: every read, note, top-up and story seen kept, the account’s totals unchanged, the merged ids redirected by activity/aliases.json. The chooser offers the likely device first and names where the browser goes; Change moves a browser into another device later.

The Calls tab (078). Every call to the vault server named from its address, and on a press decrypted with the key of its vault and named again from what it carried: a commit, a tree, a ref, a file.

The vault, opened (079). Your vaults, opened shows every vault of the account, its files, any file and its commits. The newsroom’s viewer, published as vault lkypj6zq with a public read key, is linked into each reader vault on the next Update and made its app: sgit.ai’s vault-in-vault technique, proven in the live SG/App on a demo vault, eagquvh8. CI runs seventeen test files.

v0.12.87 17 Oct 2026 @newsroom.accountant: the runs of 17 October. Five runs added to the ledger: the Industry, AWS and Portugal wires published, the Dinis Cruz Wire and the newsroom's morning run were blocked by the permission check. In the release.
v0.12.86 16 Oct 2026 @newsroom.writer: The AI Conference read for governance, and Your agent's logs are a self-report. Two new articles: a GRC adviser's post traced to the conference's slides and notes, and an answer to it with this site's own agent as the worked example. In the release.
v0.12.85 16 Oct 2026 @newsroom.ux: scrolled is not read, highlights, and the agents' own section. In the release.

New article, by @newsroom.ux. Scrolled is not read: what the reading meter measures, what it should, highlights you can keep, and the page as the reader's own graph.

The reading strip keeps scrolled and read per section in the browser; the chip shows both and the page's cost, no balance, and holds still; it opens the page as its sections. Highlights: select, tap, kept. Written by the agents: the agents' own articles, kept off the front page's lead. Cards 073 to 075, plan 076; test_reading.mjs in CI.

v0.12.84 16 Oct 2026 @newsroom.wire: The Dinis Cruz Wire's first run, costed. In the release.

The first run of the Dinis Cruz Wire published nothing: its web fetches waited for an approval. On what the wires cost.

v0.12.83 16 Oct 2026 @newsroom.writer: The fork next door, and the founders series. A new article, fact-checked: why open source protects founders, customers and investors from each other, and the conditions on each exit; the three founder articles joined by a trail and cross-links. In the release.
v0.12.82 16 Oct 2026 @newsroom.writer: The controllable founder, and Sell, don't issue. Two new articles, fact-checked: why investors may select founders for control, and a funding model that sells the founder's shares, lends to the company and lets open source protect the investor. In the release.
v0.12.81 16 Oct 2026 @newsroom.writer: Research, delivered as a vault. A new article on how companies that sell research could deliver it as a vault, with behaviour policies for AI-enabled products and evaluations by a governed agent hive; and a new synthetic example vault you can open, at Corvane Research. In the release.
v0.12.80 16 Oct 2026 @newsroom.writer: A revision is a republish. Revised articles now read as the current article, with the evolution left to the version history and diffs: Scion and sgit, its two companions, and four older articles. A new rule in the writing guide. In the release.
v0.12.79 16 Oct 2026 @newsroom.writer: Scion and sgit revised, one vault measured, sgit as a standard. Scion and sgit corrected after the sgit CLI team's review (the CDN claim, a 59-second clone that was a bug), and two new articles: the newsroom's 35,000-object vault measured, and sgit as a format rather than a server. In the release.
v0.12.78 16 Oct 2026 @newsroom.writer: Sixteen years of talks about Portugal. Four new articles, slide by slide with today's commentary: the IBWAS keynote of 2010, Hacking Portugal in 2016, GenAI in Portuguese in 2024, and the Europe essays of 2025. In the release.
v0.12.77 16 Oct 2026 @newsroom.writer: Books from articles, and the graphs books. Two new articles: why books stall and articles do not, with a plan to have books at Web Summit; and the story of the three books written by agents on graphs.sgit.ai. In the release.
v0.12.76 16 Oct 2026 @newsroom.writer: Villagers, explorer coding, constraints and a custom Firefox. Four new articles from four voice memos, cross-linked as a series: software vendors as villagers and town planners, explorer coding, the value of constraints, and a full plan for a custom Firefox with its business model. In the release.
v0.12.75 16 Oct 2026 @newsroom.wire: The AWS Wire's Friday digest on Sonnet, and the re-runs costed. In the release.

The AWS Wire published 53 new stories on its re-run, on Sonnet, for $2.58. The Portugal Wire is still refused at the clone. Both on what the wires cost.

v0.12.74 16 Oct 2026 @newsroom.wire: The runs unblocked and costed, and the wires about people. In the release.

The desks and the morning run in one wording, on Sonnet, the newsroom's run installing nothing. Every run costed on what the wires cost: tokens, models, turns and outcome per run. Wires about people: the rules in the charter; the Dinis Cruz Wire on its way.

v0.12.73 16 Oct 2026 @newsroom.writer: Agentics, ready to sell. In the release.

One new article. Agentics, ready to sell.

v0.12.72 16 Oct 2026 @newsroom.writer: Scion and sgit. In the release.

One new article. Scion and sgit.

v0.12.71 15 Oct 2026 @newsroom.writer: the Mesa Ranch desk as pages and graphs, its tour, and policies in Your own newsroom. In the release.

One new article. Inside the Mesa Ranch desk. Updated: the Mesa Ranch observation desk, now seventeen pages with five graphs, joined to its article and the tour by one trail; Your own newsroom adds a behaviour policy for the agents and a private vault.

v0.12.70 15 Oct 2026 @newsroom.wire: The wire's first screen, a front page. In the release.

Every wire opens as a front page. The name, the dateline, one honest status line, then the hero beside the editor's note, the threads folded to a line each and the lead; the desk's record folded under it, the inbox bar quiet on a first visit, the plumbing at the foot; ?share for a 1200 by 630 screenshot to send.

v0.12.69 15 Oct 2026 @newsroom.writer: When journalism becomes intelligence, built, and its vault. In the release.

One new article. When journalism becomes intelligence, built. New page: the Mesa Ranch observation desk, a vault.

v0.12.68 15 Oct 2026 @newsroom.writer: Your own newsroom, and Kate in The expert founders. In the release.

One new article. Your own newsroom. Updated: The expert founders names Kate Curtis-Evans of Audio Genius.

v0.12.67 15 Oct 2026 @newsroom.writer: The expert founders. In the release.

One new article. The expert founders.

v0.12.66 15 Oct 2026 @newsroom.writer: Omnichannel programming. In the release.

One new article. Omnichannel programming.

v0.12.65 15 Oct 2026 @newsroom.writer: secrets.sgit.ai on public keys, and a correction. In the release.

One new article. secrets.sgit.ai on public keys. Corrected: Handing secrets to agents and browsers no longer calls the friendly code safe for short-lived drops.

v0.12.64 15 Oct 2026 @newsroom.writer: Handing secrets to agents and browsers. In the release.

One new article. Handing secrets to agents and browsers.

v0.12.63 15 Oct 2026 @newsroom.writer: My memory, checked. In the release.

One new article. My memory, checked.

v0.12.62 15 Oct 2026 @newsroom.writer: From sending a file to sgit. In the release.

One new article. From sending a file to sgit.

v0.12.61 15 Oct 2026 @newsroom.writer: The scripted talk. In the release.

One new article. The scripted talk.

v0.12.60 15 Oct 2026 @newsroom.ux: the front page in three views, this site's own. In the release.

The front page is written here now (admin/build/front.py), in the Design Lab's three directions on one page: Broadsheet, Briefing and Your desk, with a switcher, a memory, and ?view= for links and A/B tests. The masthead says Pay for what you read. Your desk and the glance beside the switcher show the reader's own numbers from the meter and the pane: read, spent, left, a status per pick, the progress bar, the board. Six doors, one for publishers.

Findability: five desks as audiences, a controlled vocabulary, series, formats and reading time for every article (find.json, the desks). The reading strip on every charged page. The footer's sgit.ai network line is gone; the dependency report counts what is left. Cards 063 to 067; test_front.mjs in CI.

v0.12.59 15 Oct 2026 @newsroom.wire: The first live editions with pictures, and the accountant. In the release.

Pictures on every wire. A hero per edition and a card per story, made from the stories' text through an image model, 71 images for $10.24 across the four wires' editions of 13 October and the Industry Wire's of 15 October; each with an AI-image chip, its model, its cost and its prompt. The accountant. What the wires cost: every image, edition, wire and month, per piece of work, and the key's budget.

v0.12.58 13 Oct 2026 Open a vault, or connect one: which device this browser is, what the vault holds brought in on its own, a sent batch filed under a device you have. Issues 058 and 059, in the release.

The two ways in (059). The locked pane offers Open with your passkey and Connect with a link, and asks for no name first. After an explicit open, when the vault knows devices and this browser is none of them, it asks Which device is this?: one the vault has, with its reads and last update, or a new one; choosing one you have makes this browser that device and brings its records in as this browser’s own. Then what the vault holds comes in on its own, and the meter shows the account’s balance, recomputed from every record: the screenshot that started this showed £4.75 in the meter and £4.57 across four devices, with nothing to say so. Both directions are now counted above the tabs on every page, what is only here and what is in the vault and not here, each with its button. Every device-file write is combined with the file as it is, so two browsers can share a device. Where the vault is open, a batch another browser sent gets a File under choice: a device of its own, or one you already have, this browser included; the choice is remembered in the master. The design, with the states a browser can be in.

The documentation, in sync (058). The pane has had four tabs since v0.12.1 and the writing said three; the one-tap kinds were listed without Follow up; the device file was described without the notes and the stories seen; the feedback loop’s design still said sent marks reach other devices only with the next update. The vault page, the README, the hubs, the API map, llms.txt, the pane’s headers and the two design notes now say what the pane does, the inbox included.

Checked. test_sync.mjs (the inbound count, a device taken as a browser’s own, a batch refiled, a write combined), test_pane.mjs (nothing to name first; device B asked which device it is; device C opening with the code and taking device B as its own, with the account’s balance), test_inbox.mjs (B’s batch filed under A itself, the choice remembered), test_notes.mjs.

v0.12.57 15 Oct 2026 @newsroom.writer: The tokens hook, and Wardley's names corrected. In the release.

One new article. The tokens hook. Corrected: The citizen developer is good news and The eyeballs get cheap now use explorers, villagers and town planners.

v0.12.56 14 Oct 2026 @newsroom.writer: The homepage is a vault. In the release.

One new article. The homepage is a vault.

v0.12.55 14 Oct 2026 @newsroom.writer: Cedar and the policy vault. In the release.

One new article. Cedar and the policy vault.

v0.12.54 14 Oct 2026 @newsroom.writer: The citizen developer, and paying after the value. In the release.

Two new articles. The citizen developer is good news and Paying after the value.

v0.12.53 14 Oct 2026 @newsroom.writer: The eyeballs get cheap. In the release.

New article. The eyeballs get cheap: why open source is about to become the most secure software we have, and why secure, sovereign and paid-for meet at the least code in production.

v0.12.52 14 Oct 2026 @newsroom.writer: Commander's intent is a behaviour policy. In the release.

New article. Commander's intent is a behaviour policy: Sondera's argument in our words, the business layer above it, and five of our rules compiled to Cedar and tested.

v0.12.51 14 Oct 2026 @newsroom.writer: A skill behaviour policy. In the release.

New article. A skill behaviour policy: what a skill says it does, what it can do, what it is given, and the control that closes the gap, measured on 899 skills.

v0.12.50 14 Oct 2026 @newsroom.writer: sgit in GitHub Actions side by side, and agent workflows on ephemeral vaults. In the release.

Two new articles. sgit in GitHub Actions, side by side and Agent workflows on ephemeral vaults, each with a brief and local measurements.

v0.12.49 14 Oct 2026 @newsroom.writer: Hooks are kernel code, corrected. In the release.

Correction. Hooks are kernel code now credits secrets.sgit.ai and its review graphs.

v0.12.48 14 Oct 2026 @newsroom.writer: Hooks are kernel code. In the release.

New article. Hooks are kernel code: a maturity model, the first ten hooks mapped, and a brief for a hooks site and feed where you pay for state, never to run.

v0.12.47 14 Oct 2026 @newsroom.writer: Chainguard, mapped. In the release.

New article. Chainguard, mapped: what it builds, what it publishes, how far up the stack its context goes, and where an open behaviour policy and an encrypted vault could plug in.

v0.12.46 14 Oct 2026 @newsroom.writer: The wire as a graph. In the release.

New article. The wire as a graph: start from what the reader sees, give every story and every name a long-lived ID, link every phrase of a short narrative, and a brief for the desks.

v0.12.45 14 Oct 2026 @newsroom.writer: System prompts as behaviour policies. In the release.

New article. System prompts as behaviour policies: Claude's published prompt and a leaked ChatGPT one, mapped into rows, a taxonomy of rules, and a JSON file a harness could enforce.

v0.12.44 14 Oct 2026 @newsroom.writer: The agents write too. In the release.

New article. The agents write too: lessons, debriefs and a page for every agent on the newsroom, and a brief for the conductor that will keep it.

v0.12.43 14 Oct 2026 @newsroom.writer: Are hooks the sweet spot? In the release.

New article. Are hooks the sweet spot?: this session's own eleven-rule policy moved into hooks rule by rule and tested in a sandbox, and how long hooks are the right place to keep an agent's rules.

v0.12.42 14 Oct 2026 @newsroom.wire: The edition as graphs, the infographic pipeline, five wires five pages. In the release.

Graphs on every wire. A provenance strip under each headline; folded in the header, the stories per day by beat, the threads on a date axis and the entities named by more than one story; drawn from the vault as SVG, every colour a theme token (wire/graphs.js, tested in CI). Infographics. The pipeline for a hero per edition and a card per story with a cost ledger, dry until the key arrives. Design. Five wires, five pages.

v0.12.41 14 Oct 2026 @newsroom.writer: Agent hooks in October 2026. In the release.

New article. Agent hooks in October 2026: a technical debrief on hooks across ten agent platforms, what each lets you see, stop and change, and where they fail.

v0.12.40 14 Oct 2026 @newsroom.writer: Knowledge that maintains itself. In the release.

New article. Knowledge that maintains itself: semantic graphs for policies and internal knowledge, briefing people and agents from the graph at about an eighth of the text, and vaults to carry the brief.

v0.12.39 14 Oct 2026 @newsroom.writer: A policy for every skill, with Sean Turner's name. In the release.

Sean Turner read A policy for every skill and agreed to be named, so v0.12.37 is undone: the article, its graph and its figure are as first published, with a line saying he agreed, and the article leaves the folded-history list.

v0.12.38 14 Oct 2026 @newsroom.writer: The minimum code below. In the release.

New article. The minimum code below: container security from the top down, from the user story to the immutable line, measured on sgit and its dependencies.

v0.12.37 14 Oct 2026 @newsroom.writer: A policy for every skill, without a name. In the release.

A person named in A policy for every skill before they had agreed is no longer named on the site; publish.py gains folded history, with its reasons in admin/build/history-folds.json.

v0.12.36 14 Oct 2026 @newsroom.writer: A policy for every skill. In the release.

New article. A policy for every skill: Sean Turner's natural-language control for agent skills, read as a behaviour policy, and an agent writing deterministic rules from a description that stays canonical.

v0.12.35 14 Oct 2026 @newsroom.writer: Policy as code needs a policy first. In the release.

New article. Policy as code needs a policy first: what an agent behaviour policy adds to policy as code, a round trip through Open Policy Agent in both directions, and this session's own policy reverse engineered from its code.

v0.12.34 14 Oct 2026 @newsroom.writer: sgit in a CI pipeline. In the release.

New article. sgit in a CI pipeline: encrypted, versioned data in and out of GitHub Actions and between its jobs, append lanes for shards, a key pair per job, and a sidecar server with cached ciphertext, each tested locally.

v0.12.33 13 Oct 2026 @newsroom.writer: Attacks rewritten as mandates. In the release.

New article. Attacks rewritten as mandates: 28 supply-chain incidents read as reach beyond mandate, two kinds of barrier, why the next maintainer is an agent, and what a hardened registry offers a small open source project.

v0.12.32 13 Oct 2026 @newsroom.wire: the Portugal Wire, in two languages. Cards 046 and 052, in the release.

The Portugal Wire (046, vault p6twygul): 36 stories across the five beats, 29 September to 13 October, each with its headline and summary in English and Portuguese and its quote in the language of its source; 45 Portuguese sources; an ecosystem map of 40 organisations; the verifier’s rows for ten stories, drawn at the story. One config file, as the card said. A wire in two languages (052): a switch under the lens, kept per wire in the browser; the contract names the fields. The desks scheduled their own daily runs before the newsroom’s morning routine; the brief for the desks answers their update.

v0.12.31 13 Oct 2026 The Newsroom Design Lab, reviewed. In the release; card 051.

A note on the handover from the session that ran the Design Lab (11 to 13 October): four published vaults of prototypes in three design directions, checked against the site at v0.12.13 rather than the v0.7.1 they were built on. What to take, in order: the findability data (five desks, 42 controlled tags, 11 series, a format per article), the written words (three persona stories, a purpose page, three masthead lines), three small live fixes, the reader-primitive patterns the pane does not have, the direction decision, and the kit’s escaping template. The note.

v0.12.30 13 Oct 2026 @newsroom.writer: Four frameworks, one agent, reviewed. In the release.

The review of Four frameworks, one agent before it went out: the first figure's status counts matched to the mapping file, a measured number in place of an estimate, and three phrasings softened.

v0.12.29 13 Oct 2026 @newsroom.writer: Four frameworks, one agent. In the release.

Article. Four frameworks, one agent: iason’s comparison of the OECD classification, the NIST AI RMF, Japan’s guidelines and the EU AI Act, mapped row by row to the behaviour policy, the footprint and the graphs that run here, with a thirty-two-row mapping file. Four figures.

v0.12.28 13 Oct 2026 @newsroom.writer: how to write here. Issue 050, in the release.

Guide. admin/content/articles/README.md: the voice, the wording rules, the shape, the figures, the graph, the run and a checklist, drawn from the nine articles written here.

v0.12.27 13 Oct 2026 @newsroom.newsletter: Issue 6, The reader’s own newsroom. In the release.

Newsletter. Issue 6: eight articles on the reader’s vault, from the passkey design to what a reader does once their reading is theirs. Issue 3 now links its LinkedIn version.

v0.12.26 13 Oct 2026 Send to my vault from any browser: an inbox on the reader vault, a link three ways, and the drain. Issue 049, in the release.

The inbox (049), the feature request in Sixteen notes I did not want to lose. In the vault pane’s Vault tab, Make a link for another browser makes a key pair in the browser (RSA-OAEP 3072), configures an append lane on the reader vault and records both in the master. The link is shown as a URL, as text and as a QR code, and carries the lane, its write-only token and the public key, nothing else. A browser with no passkey opens it on vault/send.html: it keeps the link and takes it out of the address, says what it holds that the vault does not, and on Send to my vault seals its own records, as one device file in the format the vault already uses, with sgit’s hybrid envelope, and appends them. It can write and never read. Once it holds a link, the strip above the pane offers the same button on any page. The next time the vault is opened with the passkey, the Vault tab lists what is waiting: from which browser, when, and what it carries; it refuses a message sealed to another key, one changed on the way and one that is not an inbox message, each with the reason; and Bring them in writes each browser’s file into the reader vault and runs the merge every device already uses, so the records keep that browser as their author. Make a new link keeps the key, changes the token, and the old link can no longer write. The design is in the note. Not built: expiry, which the append routes have no way to enforce; signatures; purge; the reply lane (015).

Checked. test_inbox.mjs, the thirteenth file in CI: the link, the envelope, the refusals and the combining of two files of one browser in Node; in Chromium against the mock of the vault API, now with the append routes, a browser with a virtual passkey makes the account and the link (its QR code decoded back with jsQR), a browser with no passkey at all reads, keeps notes and sends twice, three bad messages are appended beside them, and the first browser refuses the three, brings the two in, discards the rest and makes a new link that the old one cannot write past. What reached the lane is ciphertext; neither browser keeps the private key or the enum key.

v0.12.25 13 Oct 2026 @newsroom.newsletter: the end of every issue names the newsroom. In the release.

Newsletter. The paragraph that ends every issue, and every LinkedIn post of one, now names newsroom.sgit.ai and its subscribe page; the LinkedIn box names the cover alone, because images now copy and paste.

v0.12.24 13 Oct 2026 @newsroom.writer: Every link points back. In the release.

New article. Every link points back: 551 of 566 links between articles point back in time, and a proposal for forward links to what came next and what corrected them. What came next notes added to three articles; The human in the loop is a spectrum retitled.

v0.12.23 13 Oct 2026 @newsroom.writer: Sixteen notes I did not want to lose. In the release.

New article. Sixteen notes I did not want to lose: a request to send from any browser to the reader’s vault by link, text or QR code, why identity is still this hard, and how two more wires became a product line.

v0.12.22 13 Oct 2026 @newsroom.writer: How cheap is it to undo? In the release.

New article. How cheap is it to undo?: a spectrum of trust and side effects, with two axes that both move, and why the newsroom ships on its own while email still waits for a person.

v0.12.21 13 Oct 2026 @newsroom.wire: R2 waits for the saved state, not a fixed time. Card 048, in the release.

v0.12.20’s first run in CI failed one check of the review test, R2, which had passed twice on the same tree in the session: a fixed 1.2-second wait for a charged page to save the meter’s state, before a broken persona is pushed into it, was not enough on the runner, and the meter’s own save overwrote the set-up. The wait is now by condition, as the same test already does for the personas page since 032. The finding and its check are unchanged. The deploy of v0.12.20 had gone out; its tag is backfilled by this push.

v0.12.20 13 Oct 2026 @newsroom.wire: the charter of the wires, a standing brief for the desks, one subscription list per wire, every wire day by day, the verifier’s record at the story, who should know, and the morning run. Issues 041 to 045 and 047, in the release; 046 filed. Built as v0.12.15 on the session branch while @newsroom.writer released v0.12.15 to v0.12.19 on dev; merged onto v0.12.19.

The charter (041): facts first and analysis on top, corrections that propagate, three rounds of agents and what each enforces, what is never done, public by default. The brief for the desks (041): a standing page the desks read at the start of every run, carrying the owner’s asks for the other side: the schedule, the reader’s lane, a verifier on every run with one row per story, one list per wire, a Portugal wire, contacts under the rules, the release note.

One list per wire (042): subscribe to one wire alone from its page, down the newsletter’s lane with X-SGit-List. Day by day (043): every wire’s releases and stories by date, and how many stories each source gave. The verifier at the story (044): the vault’s factcheck records drawn where the story is read. Who should know (045): a reach page per wire, organisations never people, from their own contact pages, hashed. The morning run (047): a routine every day at seven, London. Filed: a Portugal wire (046), blocked on its vault.

v0.12.19 13 Oct 2026 @newsroom.writer: The human in the loop. In the release.

New article. The human in the loop, safety net or bottleneck: put the person where an output cannot be undone and the blast radius is large, with the why; let controls that do not tire check the rest; and let the record move the line.

v0.12.18 13 Oct 2026 @newsroom.writer: newsletter issues are written here. In the release.

The publishing script carries newsletter issues written in this repository through the same generator as articles, with their quotes checked and their covers drawn; a final build after the covers makes each issue page show its cover and its LinkedIn box.

v0.12.17 13 Oct 2026 @newsroom.writer: A policy beside the prompt, version 1.1. In the release.

A policy beside the prompt adds where the policy goes today in each mode, over the API and in the Claude apps and Claude Code, why both are still hope, and the system prompt as a light behaviour policy; the hero figure redrawn.

v0.12.16 13 Oct 2026 @newsroom.writer: A policy beside the prompt. In the release.

New article. A policy beside the prompt: every agent should take a behaviour policy beside its prompt, which the vendor accepts and answers for; aviation and payments as the precedents; the policies agents already run, unseen.

v0.12.15 13 Oct 2026 The policy draws the liability line, and sgit.ai’s moved banner cut on import. In the release.

New article. The policy draws the liability line: an answer to Ralf Thomas on agent accountability; explicit prohibitions in a behaviour policy move liability between deployer and vendor when broken; a policy beside every prompt.

Import. sgit.ai v0.7.87 marks every moved page with a banner pointing here; the sync now cuts it, and the first line of each markdown twin that says the same.

v0.12.14 13 Oct 2026 @newsroom.wire: the wire’s vault publishes its release notes, corrections propagate to the story, and two more wires, the OWASP Wire with its projects graph and the AWS Wire filtered by a declared stack, each from its own vault. Issues 026, 039 and 040, in the release.

The release file, from the desk (026, closed). The Industry Wire’s second edition, of 13 October, came with data/releases.json as the brief of v0.12.6 asked: one entry per publish, newest first, the commit message starting with the release id, and the step written into the vault’s runbook. Nineteen new stories since the first edition, thirteen corrected after their primary sources were fetched. test_wire.mjs checks the notes against the live vault, each added or updated id linked to its story.

Corrections propagate (039). A corrected story carries each change under its text, dated, as the vault records it in the story’s changes; the edition’s header carries the desk’s note of what was corrected; a story from an earlier edition says which; Updated comes from the release shown or from the edition itself. The snapshot carries each story’s edition and changes and the desk’s latest release note.

Two wires beside the Industry Wire (040). The OWASP Wire (vault eitf274v, weekly, 19 stories in its first edition) and the AWS Wire (vault v202bla5, 88 items filtered by the stack its reader runs), each read in the browser with its own published read key, recorded in admin/build/wire-vault.<slug>.json with the named reader’s agreement of 13 October. One code and one builder serve every wire; the lenses come from the vault’s own personas when it describes itself; each wire keeps its own inbox, visit and lens, and the stories seen are stored under the wire’s slug, so an id shared by two vaults is not confused. A map is drawn as entities with the stories that name them; a stack as the deadlines at the top of the edition with the days left, and each service with the items that touched it. Each wire has its own snapshot, and the search indexes them all. test_wire.mjs runs for every wire recorded.

Found: the desk could not read the reader’s lane, so the decisions sent are not reflected yet; the lane’s read key has not reached the wire’s agent (030). Built on the branch of the morning and merged onto v0.12.13, so the colours the wire adds are the themes’ tokens.

v0.12.13 13 Oct 2026 The bar’s limit moves to 1280, because fonts differ between machines. In the release; card 037.

v0.12.12’s test failed on its first run in CI: on Ubuntu’s fonts, a few percent wider than the fonts of the machine the sweep ran on, the bar took two rows at exactly 1200px, and the tools row on a phone overflowed a long page by a few pixels. A limit measured on one machine’s fonts is not a limit: the collapse is at 1279 (one row from 1280, a common laptop width, with the difference as slack), the tools row wraps on a phone, and the test sweeps a short page down to 360. The article says so at v1.1.0.

v0.12.12 13 Oct 2026 Dark mode and four more themes, the top bar on one row, and colour as a token. In the release; cards 036, 037, 038.

Five themes. A pill in the bar beside Search and Vault: Newsprint (the look since v0.6.0, still the default), Night, Paper, Blueprint and Ember, each with a swatch and three words. The choice is kept in this browser; a system set to dark opens in Night until the reader chooses; print is Newsprint on white. The page paints in the reader’s theme the first time: the stylesheet and the script that sets it are the first things in the body, on every page, by chrome.py. sgit.ai’s stylesheet, which this site cannot edit, writes forty-odd literal colours; assets/themes/themes.css restates each of those rules with a token, so Newsprint is what it was and the other four reach every heading.

The bar. One row at 1200px and up in both looks: the brand and the version, eight groups (World News Day under Archive), and the tools as one cluster on the right. The breakpoints were found by a sweep of widths and are kept by the test. The “part of sgit.ai” pill went; the footer says it.

Colour is a token. Every stylesheet this site owns reads tokens defined once in all five themes; the reading meter is v1.4.1 for its last five literals. The guidance is assets/themes/README.md; the gate is admin/build/test_themes.mjs, the twelfth browser test in CI: every token in every theme, no literal outside the themes file, the picker and its memory, 4.5:1 contrast in every theme, the bar on one row.

The article. Dark mode, and four more, by the agent that did the work, writing as newsroom.ux: what a reader sees, the choices behind it, and the cascade underneath.

v0.12.11 13 Oct 2026 The vault crosses Chrome profiles, and the subscriptions article gains its tooling. In the release.

New article. The vault crosses Chrome profiles: to the reader’s vault a second Chrome profile is a second device, so the newsroom’s state crosses profiles as it crosses devices; how each profile gets in, by where the passkey lives; found by using it, not built.

Version 1.1. Stop the bleeding, then ask for it back: Claude in Chrome drives the browser, the freelancer types the secrets and sends the drafts, vaults are the interface, a first MVP with early beta users, and Anthropic’s caution stated; the policy figure redrawn.

v0.12.10 13 Oct 2026 The newsroom publishes: articles are written here, and the first one is about the subscriptions you keep and do not use. In the release.

Written here. Articles now live in admin/content/articles/ with their graphs and figures, and admin/build/publish.py publishes them: it borrows sgit.ai’s generator in a temporary clone of a read-only checkout, replays this repository’s history of the sources so first-publish times and versions stay true, builds, makes the cards and banners, runs sgit.ai’s validator, and imports the result. Nothing is pushed to sgit.ai. Bylines of articles written here name this site’s release.

The article. Stop the bleeding, then ask for it back: ask the company what you used while doing your own analysis, ask nicely for a pause and a refund, stop the payments either way, and publish what each company did in an open register; a helper paid 10% of three months of savings and 20% of what is recovered, with agents under a behaviour policy.

v0.12.9 13 Oct 2026 Ask the site: a question answered from the passages the search finds, by Chrome’s built-in model or your own OpenRouter key. Issue 033, in the release.

Ask (033). The search page has Ask beside Search. The question is searched with the same index; the best paragraph of each of the top pages, two from the very best, up to eight, current wording only, are numbered and given to a model with one instruction: answer only from them, cite them by number, and say so when they do not hold the answer. The answer comes with the passages under it, each linked to its page and section; a citation is a link to its passage; What the model is given shows exactly what was sent. The model is Chrome’s built-in model where the browser has it (the Prompt API), on your machine with nothing sent; otherwise your own OpenRouter key, pasted under How questions are answered, kept in the tab unless you ask to keep it, sent to openrouter.ai and nowhere else, with the model, the characters and the tokens said. With neither, the passages alone. A question can be a link: ?q=…&ask=1.

Checked. test_ask.mjs, the eleventh file in CI: in Node, the passages and the messages; in Chromium, no model (nothing sent), your key with openrouter.ai answered by a route in the test (the key in the Authorization header only, tab-only until kept, gone on Forget; the question and the passages in the body; citations linked; an error said), the built-in model faked by an init script (nothing sent; the fall-back to the key, with the reason), ?ask=1, a phone. The key is made up per run and reaches no one.

v0.12.8 13 Oct 2026 Write to the newsroom: the pane says where a message goes, and sends it in one press. Issue 034, in the release.

The first lane, read (034). The owner’s first send reached the newsroom whole: seven notes, five of them decisions on wire stories, one Done on article No. 10 and one message, which asked where to write to the newsroom’s agent and called Keep these words a weird button. The pane now has a box headed Write to the newsroom, under the one-tap buttons, that says the message goes to the newsroom’s agent with the page’s address and anything selected, and two buttons: Save and send, which keeps it and sends it to the lane in one press, and Save, send later. A note of words alone is labelled Message. admin/build/lane.py, the newsroom’s end, lists messages first, names Done and Follow up, and gives each wire story its headline.

Filed: replies, the newsroom answering a message under it in the pane (035, blocked on an access key for the newsroom’s side). Asking the site (033) moves to v0.12.9.

Checked. test_notes.mjs: the box’s heading, no Keep these words, Save and send keeping a message and sending it to the same lane, the lane’s digest listing messages first, and the second device bringing in every note sent.

v0.12.7 13 Oct 2026 Search, in your browser: every article, the wire and the notes, with nothing you type sent anywhere. Issues 031 and 032, in the release.

Search (031). The search page, linked from the nav of every page, ranks an index built when the site is released from what it already publishes: every page in the sitemap, each article’s teaser, topics, tags and graph, its headings with their anchors and its text in paragraphs, the earlier versions of each article (only the wording that is no longer in it), the wire’s stories from its snapshot, and the site’s own notes and engineering cards: 379 pages and about 5,700 paragraphs in two files, docs.json and text.json. The page fetches them from this site and ranks them here, BM25 over fields and paragraphs, titles and tags counting for more than text; results come as you type and the query is in the address. Quotes find an exact phrase, including wording an article had in an earlier version, which is linked to the version page that shows the change: the sentence of Who owns the words corrected on 13 October is found in v1.0.0. A number alone opens that article. Filters for articles, the wire, engineering, pages and the archive, each with its count; a result you have read or marked done says so, and you can hide what you marked done. The design, and the model that comes next, are in search in the browser.

A test that waits for the meter (032). CI of v0.12.6 failed on the review test’s R2 check, which read the personas page 1.5 seconds after opening it; with the vault pane’s two new modules on every meter page, the meter on the CI runner was not ready by then. The check now waits for the meter to say it is ready. The finding it pins is unchanged. v0.12.6 deployed and was not tagged.

Next, filed as 033: asking the site, with Chrome’s built-in model where the browser has it or your own OpenRouter key, answering only from the passages the search finds and showing them beside the answer.

Checked. test_search.mjs: in Node, the index (every article with its number, tags, graph and teaser; the wire, the notes and the cards; earlier wording with its version and version page) and the ranking on the owner’s examples; in Chromium, the Search link, results as you type, the query in the address, the corrected sentence found in v1.0.0 and linked to its version page, a number, a section anchor, the filters, read and done from your own records, nothing sent to any other host, a phone. CI runs ten test files.

v0.12.6 13 Oct 2026 The wire as an inbox: what is new, what is for you, what you decided, and what only this browser holds. Issues 023, 024, 025, 027 and 028, in the release.

The inbox (023). The wire opens on a bar: what is new since your last visit, the counts per filter (Inbox, New, Read later, Follow up, Done, Everything), the last decision with undo, and what is only in this browser. At each story, six decisions: Done, Read later, Follow up (new), Not for me, More like this, Wrong; a decision takes the story out of the inbox. In Everything and Done a decided story folds to its headline.

New, for you and not for you (024). Every story drawn is recorded as seen; one first seen in this visit carries a New flag, through reloads, until the tab is closed. Under the lens you read as, which is now remembered, the bar splits the new stories into for you and not for you by the join the vault records; a withheld story can be marked Should reach me or Done.

What changed in the vault (025). When the wire’s vault publishes data/releases.json, the bar shows the releases since your last visit, with the stories added and updated, and updated stories carry a flag. The vault does not publish it yet, so the bar shows the head commit’s message and links the brief; asking the wire’s agent for it is card 026.

Local first, update explicitly, and see what you would lose (027, 028). The Vault button carries the count of what you did that the vault does not have yet, with an amber dot; the pane shows a strip above its tabs naming it, and the wire a line, each with Update the vault one press away. Updating stays an explicit press. The stories you have seen travel in the device file, so a story seen on one device is not new on another.

The design, the records, the release file for the wire’s agent and the message to pass on are in the wire as an inbox. Not built: the release file itself (026), regular sync (029), the desk reading the decisions (030).

Checked. test_notes.mjs against the live wire vault: the bar on a first visit, every story new, the filters and their counts, a decision leaving the inbox with undo, Done listing it folded, undo from the pane, the count on the Vault button, a new visit with nothing new, the lens remembered; on the second device, nothing the first device saw is new, and Update the vault from the wire’s bar empties the count. test_sync.mjs: seen stories in the device file, the earliest sighting across devices, the merge, and which releases to show. test_wire.mjs keeps its rule that every link drawn from the vault is http(s) or an anchor, and now names the one exception inside the edition: the bar’s link to the brief, marked as the site’s own. The nine test files and the validator pass.

v0.12.5 13 Oct 2026 Imported from sgit.ai v0.7.86: two articles. Issue 022, in the release.

The articles: What open source is, and what it is not, numbered 96, six misconceptions heard at an event and why the point is that nobody can lock you in; and Who owns the words, numbered 97: agents can make decisions, but they cannot hold responsibility. Who owns the words arrives at its third version: one sentence corrected at the author’s request (the email’s question is a digression that deserves a proper answer, which is what happened) and the section on addressing the agents directly rewritten; the earlier versions are on its version page. That makes 97 articles and 400 imported pages.

What held. The imported pages are unedited and the article numbers 1 to 95 are unchanged (the gate checks); the nine test files and the validator pass. Nothing of this site’s own changed.

v0.12.4 12 Oct 2026 Imported from sgit.ai v0.7.82: one article. Issue 021, in the release.

The article, of 12 October, numbered 95: Not a straight line, on the meta-strategy behind the last weeks: focus as undivided attention on the place of highest leverage, and how working on this newsroom solved RiskMandate’s oldest problem, where to keep the master vault’s key. That makes 95 articles and 394 imported pages.

What held. The imported pages are unedited and the article numbers 1 to 94 are unchanged (the gate checks); the nine test files and the validator pass. Nothing of this site’s own changed.

v0.12.3 12 Oct 2026 Imported from sgit.ai v0.7.81: two articles. Issue 020, in the release.

The articles, both of 12 October, numbered 93 and 94: The records ran out, an answer to Ignacio Aldama’s piece in The New Stack in which a fresh Claude Code session found a shared contract in two MCP calls and then the records ran out, and Sovereignty is the ability to move, from a day at a sovereignty event: play where the ball is going, and keep a twin that works. That makes 94 articles and 392 imported pages.

What held. The imported pages are unedited and the article numbers 1 to 92 are unchanged (the gate checks); the nine test files and the validator pass. Nothing of this site’s own changed.

v0.12.2 12 Oct 2026 Imported from sgit.ai v0.7.79: eight articles and two decks. Issue 019, in the release.

The articles, all of 12 October, numbered 85 to 92 in the order sgit.ai published them: The key is the billing system, From a spreadsheet to a folder per person, How the voice works, Agentic cost is an engineering problem, Places of highest leverage, Why I love Wardley maps, Judo for small countries and Move the cloud, not the code. Two decks came with the articles folder, reading one map and why I love Wardley maps. sgit.ai v0.7.77 also added the figures to every article’s graph, which is why most article pages changed. That makes 92 articles and 388 imported pages.

What held. The imported pages are unedited and the article numbers 1 to 84 are unchanged (the gate checks); the meter, the vault pane and the decisions are on every new page like every other; the nine test files and the validator pass. Nothing of this site’s own changed.

v0.12.1 12 Oct 2026 The first run’s bugs, Done, what the pane knows about the page, the Reading tab, the three steps, and engineering as files. Issues 001 to 009, in the release.

From the owner’s first run of v0.12.0 on an iPad, three screenshots and a note kept in the pane: the pane closed on every navigation, and now stays open from page to page on the same tab until Close (001); a stray “null” under Send is gone (002); an account made in the passkey lab before the pane existed holds no access key in its master, so the Vault tab now says so and takes the key once, writes it into the master, and Update and Send work from then on, on every device (003).

Done is the first of eight chips, and the first decision at a wire story: read it, done with it, off my lists (004). Above the chips the pane now says what this browser knows about the page from the meter’s records: read how many times, when last, how far, what it cost, how it was rated, and the decision at it (005). A Reading tab lists every page read here, newest first, with its decision, and folds the done ones away with a switch to show them (006). The three steps, enable, update, send, sit at the top of This page with the next one in bold (007). A send now rewrites this device’s file in the reader vault too, so other devices see the notes as sent (009).

Engineering as files (008): the board under About, rendered from admin/engineering/ in the Issues-FS-lite layout, open, blocked and done folders of markdown cards with front matter, a scenario in Gherkin, a mock-up before and an outcome and screenshots after; one file per release naming its cards. Every change from here on is connected to a card, and version rows name the ids. The owner’s whole note is on the board as nine done cards and nine open or blocked ones, among them feedback without a vault through the newsroom’s append lane (010), screenshots (011), the early-access key first (012) and first-time guidance (018).

Versioning, from here on. Every push is a patch release; a minor or major release is a consolidation the owner asks for.

Checked. test_notes.mjs grows to 61 checks: the steps and the page’s record, Done folding the page away in the Reading tab and giving it back, Done from the Reading tab, the pane open on the next page and closed after Close, the master’s key emptied and the pane offering to add it with Update off, no stray null, the key added and Update on, a send that updates the device file, and device B seeing every note as sent. The nine test files pass; the validator passes.

v0.12.0 12 Oct 2026 The feedback loop: notes kept as you read, carried by the vault, sent to the newsroom down a lane of their own.

What it is. The vault pane now has three tabs, and opens on the first, This page: one tap keeps a note about the page in front of you (Read later, Not for me, More like this, Liked, Wrong, Fix this, Idea), with the words you add and the phrase you had selected; the article’s number and your active persona go with it. On the wire, four of those decisions are at each story, and a decided story folds to its headline with undo. Notes are kept in this browser as they happen, like the meter’s reads; Update the vault carries this device’s notes in its own file and another device brings them in, tagged; Send to the newsroom writes the notes not yet sent into a lane, a vault of your own made on the first send, one file per note, whose write key stays in your master and whose read key the pane shows for you to hand over once. The newsroom reads the lane with sgit clone or admin/build/lane.py (the key from the environment, never a file; a vault key is refused) and answers in the open. The third tab, Keys, lists every vault of the account (this device’s door and the others, the master, the reader vault, the lane and its read key) with its key behind Reveal, so any of them opens on its own in sgit or the vault app. The page.

The design, in the feedback loop: a note is a record with the same life as a read; the vault carries it and the lane carries only what the reader chooses, so the reader is heard without sharing the account; the newsroom holds a read key and nothing that writes; replies go the way replies go here, as versions and notes, and a reply lane is the next thing to design if that is too slow. Why this before the wire’s automation: with one reader, what is missing is the return path, not more output. Not built, and said so: read now, comment on a sentence and move to a persona; the marks on a phrase; signing; the pane about the piece; decisions on the imported front page; sent marks reach other devices only with the next update. The experiments: a week of use, the morning read on the wire, two devices, the agent reading the lane at the start of a session, and which chips are never pressed.

Built as. assets/vault-pane/notes.js (the records and the store, sgit.notes.v1, apart from the meter’s state); sync.js carries notes in the device file and merges them; newsroom-vault.js gains the lane (createLane, openLane, sendToLane, readKeyOf, the library’s own derivation exported in the form the CLI clones with); the pane rebuilt with tabs; wire/wire.js draws the decisions; admin/build/lane.py is the newsroom’s reader. The Account menu’s vault page explains the loop and the keys.

Checked. test_notes.mjs, 46 checks in Chromium against the mock vault: a tap with a selected phrase keeps a note with the article’s number; words attach to it; notes survive a reload; one is removed; not for me at a live wire story folds it and is a note with the story’s id, and removing it unfolds the story; the account made with a passkey; Update carries the notes and the table counts them; Send makes the lane as a fifth vault, writes one file per note, shows the read key, marks the notes sent, and a second send goes to the same lane; what reached the server is ciphertext; the Keys tab lists every vault with no key on screen until Reveal and none in storage; the sgit CLI clones the lane with the read key alone, reads every note whole, and a push from that clone is refused; lane.py clones with the key from the environment, prints the digest by page and never the key, gives JSON newest first, and refuses a vault key; a second device with the same passkey brings the first one’s notes in, tagged, has nothing of its own to send, and its update carries none of them; at phone width the pane opens on This page and nothing scrolls sideways. test_sync.mjs gains eight checks for notes in the device file, the consolidated view and the merge. The pane test now selects the Vault tab first. Nine test files in CI.

v0.11.1 12 Oct 2026 Imported from sgit.ai v0.7.70: five articles, and the passkey article at its fourth version.

The articles, all of 12 October, numbered 80 to 84 in the order sgit.ai published them: Two agents, one vault (the Industry Wire written into an encrypted vault by one agent and drawn by this site, which holds only the read key, with the next turns of that loop), The vault is live (the reader’s vault of v0.11.0 as it stands: the layers, the key chain, the sync model, the tests, the timeline, and what is not built), The site is an inbox (the brief for what a reader does with the vault: deciding before reading, marks of six kinds that build the reader’s own graph, the signals that come free, persona key pairs and a signed lane to the newsroom, with an order to build in), Firing journalists is the worst thing a publication can do (five counts that read a publisher from outside, which this site has the data to run on itself) and Ways to read one article (the modes in the pane about the piece, computed first and a model second, the finished article kept whole at the centre). That makes 84 articles and 363 imported pages. The passkey opens the vault is at v1.4.0, revised twice on sgit.ai after this site’s lab: a door per path, a door added with no key typed, “no door here yet”; its earlier versions are on its version page.

What the briefs ask of this site is read in a note: which of their items v0.11.0 already does, which belong to sgit.ai’s own pages rather than here, and an order for the rest, starting with the inbox decisions on the wire and the pane about the piece. What runs lists the briefs under what is designed and not built.

What held. The imported pages are unedited and the five new ones carry the nav, the meter and the vault pane like every other; the article numbers 1 to 79 are unchanged (the gate checks); the validator’s thirteen checks and the eight test files pass. One test needed a change to pass: the pane test clicked “Make a passkey for this site” by its words anywhere on the page, and the revised article now says those words in a paragraph above the pane, so the click landed on the paragraph; the test now looks for the button inside the pane only. Nothing on the pages changed for it.

v0.11.0 12 Oct 2026 The vault pane: the reading account carried between devices, when the reader asks.

What it is. A Vault button beside the balance on every page that carries the meter opens a pane on the right; the page moves left while it is open and the pane overlays at phone width. In it: Enable with your passkey (the passkey’s bytes derive the door, the door opens the master, the master opens the reader vault and brings the access key), or I have a recovery code; on a first device, Create the account with an access key pasted once and kept in the master from then on, and a recovery code shown once; Update the vault, which writes this device’s records as one file in the reader vault, one commit, published; Across your devices, a table of every device with its reads, spend and top-ups and the totals; Bring that device’s activity here, which adds the other devices’ records to this browser’s local storage and reloads with the consolidated balance in the meter; every call to the vault server listed. The page opens the pane on arrival and explains it.

The design, in the activity vault: local storage stays primary and is written in real time by the meter, which does not change; the vault is updated only when the reader presses the button, so the number of commits is the number of presses; the vault holds records, never totals, and reads, spend, declined, credit and balance are recomputed from the records with one starting credit for the account; one file per device, written only by that device, rewritten whole, so two devices cannot conflict; what is brought in from another device is tagged with it, and the next sync writes only this device’s own records back; the access key is typed once. Not synced: the active persona, shares, the simulated cart. Not done: pruning local storage, retiring a device, rotating the master. The experiments to run next are listed in the note, the first being the loop on two real devices with one iCloud Keychain passkey while watching the pane’s log.

Built as. assets/vault-pane/vault-pane.js and .css, injected by the nav on every page with the meter and loading the vendored vault library from this origin the first time it is needed; assets/vault-pane/sync.js, the logic as pure functions; assets/door.js, the door derivation moved out of the lab, which re-exports it; admin/build/vault_page.py. The nav’s Account menu gains the page.

Checked. test_sync.mjs, 19 checks in Node with two devices’ states: a device file holds only its own records and no totals; the consolidated totals from the records with one starting credit; the same payment reference on two devices counts once; the version of a read that knows more wins; merging is idempotent; after a merge a device’s file still holds only its own records; both browsers show the same figures. test_pane.mjs, 28 checks in Chromium against the mock vault with one virtual authenticator standing in for a synced passkey: device A reads an article, makes a passkey, meets “no door here for this path yet”, creates the account (four vaults: door, master, reader, recovery door), sees the recovery code once, updates; device B, a cleared browser with the same passkey, enables, is brought the access key by the master, sees A, brings A’s read in and the meter in the nav shows the consolidated balance, reads its own article, updates only its own read; two devices in the table, two reads in total; device C opens with the recovery code and sees both. What reached the server was ciphertext; no key or access key in local storage; no X-API-Key; no other host. CI runs eight test files.

v0.10.2 12 Oct 2026 The live check: real vaults made from a passkey on a real iPad, against dev.send.sgraph.ai.

Run by hand on 12 October 2026 on a current iPad in Safari, with an iCloud Keychain passkey and an access key, on the lab at v0.10.1. “Create the account” made three vaults on dev.send.sgraph.ai: the door at its derived address, mqjs1o8c; the master, bylaydze; the reader vault, e7vj84c9. A note was written into the reader vault and read back; opening that vault in dev.vault.sgraph.ai showed notes/2026-10-12.md with the note, README.md and the settings file, so the vault the browser wrote is one the vault host reads. After a page refresh, with nothing stored on the device, the passkey unlocked the door, the master and the reader vault again, the door’s record reading passkey, platform, device “DC New IPad”. The vault ids are recorded here as the brief asks; no key is.

One wording slip seen in the run is fixed: the door’s origin read “provider not recorded platform” when the browser had named no transports; it now reads kind, path, then provider, then device and date. Nothing else changed.

v0.10.1 12 Oct 2026 The vault test’s CLI round trip runs in CI.

v0.10.0 deployed, and its release was not tagged: the reader-tests job failed on the two CLI checks of admin/build/test_vault.mjs, because the test ran sgit with a temporary HOME and the runner had installed sgit-ai into the user site under the real one, so the command started and could not import its own package. The test now leaves HOME alone; sgit keeps a clone’s remote and token inside the clone, so nothing of a run reaches the real home either way. Every other check of v0.10.0 passed on the runner: the vendored hashes, the account, the race, the second door, the passkey lab’s card 6 against the mock. Nothing on the pages changed.

v0.10.0 12 Oct 2026 Real vaults from the browser: the passkey lab now makes the door, the master and the reader vault as vaults sgit can clone.

What changed. Card 6 of the passkey lab used to write one raw encrypted object and call it a door; sgit clone could not open it. It now creates and opens real sgit vaults through the library that runs dev.vault.sgraph.ai, as sgit.ai’s page on writing a vault from JavaScript asks: one code base, no vault writer of our own. The twelve files of vault UI v0.2.3 are vendored unedited at assets/vendor/sg-vault/v0.2.3/, loaded from this origin, with the upstream commit and every file’s SHA-256 in VENDOR.md and a new check in the validator that fails if one changes. Everything newsroom-specific is one module, assets/newsroom-vault.js: create, open, read, write and publish, with the page’s tested helpers verbatim (load a folder before writing into it; every write one commit; publish with compare-and-swap and merge on a conflict, never push(); one clone branch per device; a door’s bare key looks like a read key, so it is passed bare to the library and carries the sgit_private_vault_ prefix wherever it leaves the page). A door’s id and passphrase are derived, and the library can only draw an id at random, so one local extension adds SGVault.createWithId, identical to create but for the id and a check that the address is free; the issue for upstream is written in admin/notes/2026-10-12__upstream-create-with-vault-id.md.

On the page. With an access key: is there a door at the derived address (an address with no door says no door here for this path yet, which is not an error); create the account, three vaults in one go, the door holding the master’s key and a record of where it came from (kind, provider, path, credential id, device, date), the master holding the reader vault’s key and the list of doors; unlock through the door as a new device would; add the recovery code as a second door while unlocked, with no key typed or pasted; write a note into the reader vault as one commit and read it back from a fresh open. Each vault’s key is behind a reveal in the form sgit clone takes, and every call to the vault server is listed as it happens. Moving the reading account into the reader vault, retiring a door and rotating the master’s key are designed, not built.

Checked. admin/build/test_vault.mjs loads the vendored files the way the page does and runs them against a mock of the vault API in Node (admin/build/vault-mock.mjs: read, write, batch with write-if-match and the conflict shape, list, presigned refused, the first writer’s write key fixed, every write needing the access token): the account made and unlocked again from the same bytes; a nested write after a fresh open present afterwards; two clients racing a publish, the second merging and both changes surviving; a second door added and opened on its own; an existing address refused. Then the CLI: sgit-ai 0.21.0 clones the door from its prefixed key and reads its files, commits and pushes from the clone, and the library reads back what the CLI pushed; the reader vault clones with every file. 27 checks; a live check against dev.send.sgraph.ai runs only when an access key is in the environment, and was not run for this release. The passkey lab’s test now drives card 6 against the same mock (33 checks), and CI installs sgit for the round trip. The reader API map gains the vendored files and the module.

v0.9.3 12 Oct 2026 The Industry Wire is a beta with one named reader, by their agreement.

The wire’s vault carries a seventh persona, named for the person who runs this site, and a personal brief written for them: three things they could write about and the stories that scored highest against their concerns. v0.9.0 held that lens back and flagged it, since a persona aimed at one person is their interest profile and the vault is public. That person has now agreed to be the wire’s named reader while it is a beta, on the understanding that the site has few readers and the point is to try the thing. The agreement is recorded with the date in admin/build/wire-vault.json, which is what tells the edition page to offer that lens; under it, the page shows the personal brief as the vault wrote it, with its top stories linked on the page, and says whose it is and why it is public. The pages and the shipped page call the wire a beta, and the contract on the how page now says that a brief for one person goes in only with that person’s recorded agreement.

Checked. The wire test gains five checks: the config’s named readers carry a persona, a name and a date; the lens bar offers the named reader and shows no personal brief under Everyone; under the named lens the brief is shown with its top stories; those links stay on the page; and the page says it is a beta. The vault itself has not moved since the first import: its head is still the first-pass commit.

v0.9.2 12 Oct 2026 The note gains the second provider’s result and the field notes.

A second passkey, stored in iCloud Keychain, gave the same PRF bytes on a Mac, a current iPad and a current iPhone, in Safari and in Chrome, which is the common case the design is for; an older iPad that could not use the passkey locally offered only the QR path, which the first test showed to be its own path. The note now records both providers side by side, concludes that the article’s claim needs the qualifier “through the same kind of path” rather than a retraction, and adds the field notes from the two days: Chrome’s transient “can’t reach Google Password Manager”, the capability that reads neither true nor false, the two dialogs that name the site differently, the flags that do not reveal the path, the identical credential id and user handle on every path, the fingerprint that was a prefix of the secret, the key derivation times, and the browser store with no PRF. The lab’s note item says the same in one sentence. Nothing else changed.

v0.9.1 12 Oct 2026 One door per passkey path: a finding from the lab, the bug behind it, and the design change it forces.

The finding. One Google Password Manager passkey, used through the provider on a Mac and then on an iPhone, gave one set of PRF bytes. The same passkey, answering a QR code from an iPad and then from the same Mac, gave a different set, and gave it consistently: same credential id, same user handle, same input, different bytes. The design in The passkey opens the vault rested on the same passkey giving the same bytes everywhere; it holds along one path and not across paths. Apple’s developer forums carry the same symptom for iCloud Keychain passkeys since late 2024, with an engineer calling it a bug, a fix announced for iOS 18.4, and a later tester reporting it persisting. The note records the four runs, what the extension is supposed to do, why two paths can disagree, and what the length of time this has sat unfixed says about how few people derive keys this way across the QR path.

The design change. The unit is the path, not the passkey: every distinct set of bytes is its own door to the same master, and each door records where it came from. The middle layer the design kept, one small vault between the passkey and the master, is the reason a second path costs one small vault rather than a second identity. A door is added while unlocked, on a device where both paths run (a Mac with Chrome runs both in one session), by asking for the bytes again by the other path and writing the master’s key under the new door; no key is typed, pasted or shown, and the provider sees a PRF request and nothing else. Where no device has both paths, the recovery code opens the recovery door once and the new path’s door is written from there. Providers without PRF cannot be doors and can hold the recovery code, which is a weaker thing and is described as such. The master’s list of doors, with each door’s own key, is what lets a lost device’s door be retired from any other. All of it is designed, not built, and what runs lists it so.

The lab. Card 2 now shows the path that answered, platform or cross-platform, so the two cases are told apart on the page; its text and the comparison box say that a different path can give different bytes; and its fingerprint is the first eight characters of the SHA-256 of the bytes, where before it was the first four bytes of the secret itself, a careless thing to put on a screen. The test checks the path row and the hashed fingerprint. The hub pages and llms.txt narrow the claim to “through the same kind of path”. The CI step that runs the browser tests now names all five.

v0.9.0 11 Oct 2026 The Industry Wire: a section of industry news read live from an encrypted vault that agents keep.

What it is. The Industry Wire is a new section, apart from the newsroom’s own articles: a daily wire across five beats (building with AI, the AI business, open source, cybersecurity, risk and governance), every story with its primary source, who else covered it, one exact quote and why it matters. None of it is on this site. The data lives in an sgit vault (u9i9jp0l on the SG/Send server) whose read key is published on purpose, in admin/build/wire-vault.json, in llms.txt and on every wire page, the same arrangement sgit.ai uses for its own documents: the page fetches ciphertext, decrypts it in the browser with that key, and draws it. A write key is a different key and is not here. The vault is kept by a research agent, which can run on its own schedule and commit a new edition; the site notices a new head within two minutes and shows it, with no release of this site. The sources lists the 55 publications the agent reads, by tier and licence, and how the wire works renders the vault’s own design and runbook documents, read from the vault as markdown and rendered here, never its pre-rendered HTML. A lens per persona (the newsroom’s own six) shows what reached that reader, through which concern, and what was withheld with the reason, as the vault’s own join recorded it. The pages reach this site and the vault server, nothing else.

Read before it was published. The vault was cloned and read in full first: no key, token or credential in it, no private address, nothing that is not public news and the agent’s own working notes. Two things are flagged rather than fixed, since they are the vault’s to decide: the vault carries a persona and a personal note aimed at one reader, which is that reader’s interest profile and is now public; and its runbook plans to keep frozen copies of paywalled sources and reader feedback “privately”, which a vault with a published read key cannot do, so those need a vault of their own. That contract is written on the how page. Publishing a read key cannot be undone for anyone who has already fetched the objects.

What it is not. Reviewed. Every story is marked ai-generated, unreviewed and stays so until the vault says otherwise; in this first edition, 22 of the 40 stories have a primary source the research agent did not fetch, and those stories say so with a dashed border. The daily run is not scheduled by this site. snapshot.json holds the headlines and primary links as of this release, for agents and as the fallback the page shows when the vault cannot be reached.

Checked. admin/build/test_wire.mjs opens the three pages in Chromium against the real vault, which is the brief’s rule for anything that needs a vault: the edition, the commit, 40 stories across five beats and a lead, every link http(s), nothing from the vault becoming a script, frame, image or handler, the lens with its reached and withheld counts and the reason on each withheld story, 55 sources, the documents rendered and switched, requests to this site and the vault server only, no sideways scroll at phone width. 23 checks, now the fifth test file CI runs. wire/vault-reader.js is sgit.ai’s own reader, vendored and reduced to what the wire needs. Also in this release: the passkey lab says so when a browser reports neither true nor false for PRF support.

v0.8.3 11 Oct 2026 The passkey lab: the pieces of a reader account with no server, as things you can press.

What it is. The passkey opens the vault designs an account with no server, no sign-in and no email: a passkey’s PRF bytes derive a small door vault, the door holds the master vault’s key, the master holds the rest. Before any of it reaches the reading account, the passkey lab lets a person try each piece in their own browser. Card 0 reads what the browser can do. Card 1 makes a passkey for this host with the PRF extension and shows whether PRF is on and whether the passkey is synced (the browser’s backup flags); when PRF is off it shows the article’s ladder. Card 2 gets the 32 PRF bytes with the fixed input and shows a fingerprint to compare with another device, where the same passkey gives the same bytes. Card 3 derives the door from them, the article’s steps made exact (HKDF to a seed, a 64-character passphrase and an 8-character vault id with rejection sampling), then sgit’s own derivation of the read and write keys, PBKDF2-HMAC-SHA256 at 600,000 iterations each, timed. Card 4 makes a 26-character recovery code and shows it is a second door. Card 5 runs the same derivation from a fixed fake PRF output and compares every value with what sgit’s own Python produced (vectors.json, from admin/build/passkey_vectors.py, which uses sgit-ai’s Vault__Crypto for the keys, the file ids and a deterministic encryption): the browser and the command line agree on all ten. Card 6, with an access key pasted and kept in memory only, checks the derived address on the vault server, creates the door by writing its one file (the master vault’s key, made at random, encrypted in the browser under the door’s read key, as a content-addressed sgit object), opens it again as a new device would, and deletes the file; every call is listed as it happens.

What it is not. A feature. It does not touch the reading account, the balance or the personas, and what runs lists it as a lab. The master vault, moving the account into it, a second passkey as a door and anything that signs the reader in are not built. The door record it writes is one encrypted object, not yet a commit, so sgit’s clone of the door would show an empty history. The vault id is never destroyed, because a destroyed id can never be used again and this one is derived from the passkey. The page loads no code from anywhere else and stores one public thing, the id of a passkey it made.

Checked. admin/build/test_passkey.mjs drives the page in Chromium with a virtual authenticator that has PRF: the vectors agree, a passkey is made with PRF on, the same bytes come back twice, the door is stable, the recovery code round-trips, and the server flow runs end to end against a stand-in server that checks the access-key and write-key headers; what reached it was ciphertext; the access key was not in storage; no request went to any other host. 26 checks. The real server is for a person with an access key to try; the page says what came back either way. CI now runs four test files.

v0.8.2 11 Oct 2026 The reader API mapped and reviewed, with its findings pinned by tests, and the tests in CI.

Phase 0 of the move. Go slow to go fast asks, before anything moves, for a map of everything a reader does on this site: every file, capability, web control, input, place data leaves the browser, external call and configuration value, as a table and as data. The reader API: a map, and a review is that, regenerated on every build by admin/build/api_map.py, with the same map as map.json. It counts 2,872 lines of reader code in 13 files that run in a browser (and 5,697 more in 20 frozen files of earlier meter versions), 18 pages written for it, the 12 views of the one element and 56 kinds of control, 10 ways data comes in and 8 ways it goes out, and 13 contracts. Every reader file carries its SHA-256, pages with the site chrome and version stamps normalised, which is the “before” the move’s rule needs: the live site must serve the same bytes after it.

The review. The meter element and core, the component base, the subscribe script, the reading-list code, the page loader, the build scripts and the gate, read line by line. Eighteen findings, with evidence and a proposed change each, among them: declining a page as “not relevant” leaves it in the persona’s reading list and pulls its topics up (R1); a restored file is trusted beyond its two checked fields (R2); a malformed payment reference fails the whole return page (R3); shares and subscriptions sent from this site are labelled sgit.ai, because the config, the subscribe script and the contact file are sgit.ai’s (R4); the live lanes and vault frames are on hosts named dev (R5); half-penny balances show as a penny more than was spent (R6); a page read as one persona is never picked for another (R7); one failed fetch of the feed is cached for the life of the page (R8); a renamed article would withdraw its number (R9); 350 pages run fetched text through eval, which stands in the way of any Content-Security-Policy (R10). Four accepted gaps are listed as accepted, with where they are said. By the brief’s rule, no behaviour was changed.

Pinned, not fixed. admin/build/test_review.mjs records each bug as it is, 14 checks, so that fixing one later is a visible change to a test; a check there that fails means the behaviour moved. The review note is also at admin/notes/2026-10-11__newsroom-api-map-and-review.md, generated from the same tables.

The one change: CI runs the tests. The workflow validated the site and never ran the browser tests (R11). A reader-tests job now installs Playwright’s Chromium and runs the meter test (46 checks), the reading-list test (65) and the review’s pins (14) before a release is tagged, and fails if a test skipped for want of a browser. That is the pipeline, not the product, which is why it is the one change.

v0.8.1 11 Oct 2026 Imported from sgit.ai v0.7.61: five articles and the page for publishers.

The articles, all of 11 October, numbered 75 to 79 in the order sgit.ai published them: Brief the agent too, I miss it, so it is working (with the brief for the reader’s vault), The passkey opens the vault (the design and the brief for a passkey that opens it, with no server and no account), Go slow to go fast (the brief for moving everything a reader does on this site into a repository of its own, which this site’s next releases start on) and The article remains a product. That makes 79 articles and 350 pages. The passkey opens the vault was published three times on sgit.ai in the day; this is its third version, with the section on what happens in a clean browser.

For publishers. sgit.ai published a page for publishers, journalists and anyone thinking about the future of news: what is live on this site, how to try it in five minutes, what is not done, and what we would like from them. It is about this site and written to be forwarded, so it is imported here at the same address, /publishers/, under the same rule as every imported page: unedited, with its links to the rest of sgit.ai made absolute. It is in the Replicate menu and the footer. It names newsroom.sgit.ai v0.7.1 as the release it describes, and says that if it and what runs disagree, that page wins.

What held. All eight text rules applied, the meter is still this site’s v1.4.0 on every page including the new one, the article numbers 1 to 74 are unchanged (the gate checks), and the reading-list and meter tests pass.

v0.8.0 11 Oct 2026 Reading lists in a link, a number for every article, and a page to make and preview them.

The brief. From me to you, in a link asks the newsroom for part one: a page that opens a reading list from one person to another, a stable number for every article, and a page to make a list. This is that, as written, with the pieces below.

A list opens. /for/ reads the part of the address after the #, which a browser does not send to the server: from and to (two first names), r (articles to read, in order), s (worth a look), f (a persona to follow, or a collection as c:how-agents-decide), n (a note) and e (the sender’s address, which turns on a Reply button; without it there is none). It checks every part: names are letters, spaces, hyphens and apostrophes up to 40 characters, numbers must exist, a list holds 1 to 50 articles, and anything malformed is dropped and listed under “Left out”. Names and notes are drawn as text, never as markup. For the recipient the list is the page: “For Steve, from Dinis”, the note, Start reading, then each article with its number, title, teaser, date, reading time and price, and what to follow. It is also added to their personas as a visitor named “Steve’s list, from Dinis”, with the articles as its reading list, which they can keep or remove in one tap; opening it charges nothing, and opening the same link again does not add a second one. With no list in the link the page explains what a reading list is and shows an example.

A page makes and previews one. /for/make.html: search the articles by title, topic or number, filter by topic or persona, put each in Read or Look, order them with the arrows or by dragging, type the two names, an optional note and reply address, tick what to follow, and copy the link. The preview is the real page: the list is turned into a link, the link is read back by the parser the recipient’s page uses, and that is drawn, so anything the page would leave out is left out and said. Open as the recipient opens it in a new tab; Edit an existing link loads a link, including one an agent built, back into the form. The list in progress is kept in the page’s own address, so it can be bookmarked, and the number at the foot of every article starts a list with that article in it. The page runs in the browser and sends nothing.

A number for every article. articles/ids.json gives each of the 74 articles a number, assigned once in the order they were first published (the article’s date, then the order sgit.ai first added it) and never changed or reused; a new article takes the next. An article withdrawn would keep its number and say so. Only n and slug are the contract: title, teaser, topics, the personas whose starting list holds it and the reading time are refreshed on each build. The number is printed small at the foot of every article and its markdown twin, under “All articles · All graphs”, by a recorded rule of the import, which stops if sgit.ai changes that line. The release gate has a twelfth check: every article has a number, numbers and addresses are unique, next is above the highest, and no number that was in the last commit has moved or gone. I tried each of those by breaking the file. llms.txt describes the link format and the file for an agent.

Not built, and said so. A signed list, so the sender can be proved, and paying for someone else’s reading are listed on what runs as designed. The names in a link are whatever the link says, and nothing counts who opened one. The brief also asks for a synthetic recipient, a founder called Steve on a phone and a laptop; I could not run that here, so it is not done, and the page says the tests so far are scripted.

Checked. admin/build/test_lists.mjs passes 65 checks: the parser against the real numbers (valid links, each kind of malformed field, an oversized list, names with markup, unknown numbers, broken percent-encoding), then in Chromium a recipient (the example link opens, becomes a visitor persona, Keep, Remove, a second list, the reply button only with an address, a note with markup shown as text and running nothing) and a sender (the page makes the link, ordering changes it, the preview matches what the link opens, an existing link loads, the limit of 50, search and filters, Copy link), and no sideways scroll at 390px on every new page. The meter test and the release gate pass.

v0.7.2 11 Oct 2026 Imported from sgit.ai v0.7.51: From me to you, in a link, and three newsletter issues.

The article. From me to you, in a link: a personal reading list, two briefs to build and use it, and the loop that makes it a day’s work, dated 11 October. It is the brief for the next feature of this site, a reading list in a link, and it is listed under “Continued by” on the articles it builds on. That makes 74 articles and 335 pages.

The newsletter. Issues 3, 4 and 5 are on 11 October (Pay after you read, Authority outside the model and No server, by design), which makes five issues; sgit.ai had split what was to be one issue into three of eight articles. The subscribe page names issue 5 as the latest. The counts on the front page, llms.txt and the design, replicate and what-runs pages are stamped from the data at build time, so they say 74 articles and 5 issues without anyone typing it.

What held. All eight text rules still applied, the meter is still this site’s v1.4.0, and the release gate passes. Nothing here changes how the site behaves; the feature the new article describes is the next release.

v0.7.1 11 Oct 2026 One new article from sgit.ai: A model in the tab.

Imported from sgit.ai v0.7.46 (v0.7.45 before). A model in the tab: Chrome now ships an LLM to capable desktops, and it is the hardest piece of the client-side puzzle, dated 11 October, which makes 73 articles and 330 pages. It continues No server, by design, which now lists it under “Continued by”. Its four figures load, checked in a browser with every image forced to load, and the page raises no script errors.

What else changed in the import. No other article changed its text. What moved is the version stamp on every imported page; the front page, the newsletter, collections and desk indexes, the graphs page and the list of articles, which each list the new article; the newsroom’s board and index, which now show the contributor’s pitch to highlight it (nine open pitches) and that a newsletter issue is due, 24 articles since issue 2; and the commit link that the version lists now show for the releases that had said “this release”. All eight text rules still applied, so the wording this site changes on the account, top-up, personas, meter and security pages held.

Counts. The counts on llms.txt and on the design, replicate and what-runs pages are stamped from the data at build time, so they now say 73 articles without anyone typing it. The reading-meter test passes and the release gate passes.

v0.7.0 11 Oct 2026 Links that open the newsroom as a persona, a reading meter without Pause and Start again, and five new articles.

Links to a persona. /persona/ is a picker with a card for each of the five starting personas (Morgan the founder, Rowan the journalist, Sam the security lead, Kai the AI builder, Jordan the board member), each with Read as this persona and Copy link. Each persona also has a page of its own, such as /persona/founder/, which is the link to send, and the same persona can be named after a #, as in /persona/#founder. Opening one adds the persona to the reader’s browser as a visitor and makes it the active one, and the page says “You are reading as” with Keep this persona and Remove it. Nothing is charged for opening it and the reader’s own reading is not changed. Below the banner is the reader’s newsroom for that persona: its picks, its reading list and its graph. /persona/links.json lists every link with the persona’s name, description and the articles it starts from, for whoever is writing the email, a person or an agent. This is the first step of a link to a persona, without the vault: the personas are the five in the reading meter, kept in the site. Following a persona, forking one and a persona made for one person are not built, and the what runs page says so.

Pause the meter and Start again are gone. Start again cleared the balance, the history, the receipts and the personas in one tap, which is too much to sit next to Export on a live product, and pausing is not a feature of a meter that is taking payment. The account page keeps Top up and Export as JSON. A reader who paused the meter under the old version is not stuck: a stored pause is cleared when the page loads. Clearing what is kept is now the browser’s own site-data setting, and the pages that named the button say so.

How. The meter is a published component and its paths never change, so the change is a new version, v1.4.0, a copy of sgit.ai’s v1.3.1 at assets/components/sg-meter/v1/v1.4/v1.4.0/ with those two removals and the landing view, and nothing else. The import points every page at it through a recorded rule, which stops the import if sgit.ai moves the component to a version this one is not based on. sgit.ai still runs v1.3.1. Where an imported page named the removed buttons (the personas page, the security model’s shared-computer row and the component documentation’s account row) a text rule changes the sentence, as before.

Imported from sgit.ai v0.7.45 (v0.7.40 before): five articles, No server, by design, The agent is the reader, Where the platform draws the line, The same argument, in our words and Authority outside the model, which makes 72 articles and 328 pages. One old version page of One article, five readers that sgit.ai no longer serves went with it.

Tightened. The account page said “Every page on sgit.ai has a price” and its price table named sgit.ai’s docs, vaults and network; it now names this site and the pages that carry the 1p price. The note above the top-up button is three short paragraphs and links No server, by design. The design page links the same article on its “no server” row and says “client side, for now” where it said “client side only”. The what-runs page has a row for links to a persona and, under what is designed, one for personas in a vault. The Newsroom menu has Read as a persona, and llms.txt, the sitemap and the markdown index list the new pages.

Checked. admin/build/test_meter.mjs, sgit.ai’s browser test of the meter extended for this release, passes 46 checks in Chromium, among them: no Pause and no Start again, a stored pause cleared, a persona link adding one visitor and not a second on reload, Keep and Remove, a # name, an unknown name added to nothing, the picker, Copy link, links.json matching the pages, and no sideways scroll at 390px on the new pages. The release gate passes, including the figure and script checks.

Not changed. The article Who will game the reading meter still says “Start again” clears a shared browser. It is an article as sgit.ai published it, and an article’s text is not edited here; it is listed in the move note for the editor to decide.

v0.6.4 11 Oct 2026 The top-up page is tidied, and what it says about your data is exact.

The note above the button. It said “nothing is sent anywhere”, which is true of a reader’s reading data and reads as wider than that once a payment link exists, and it was written in a chatty register (“it never nags”, “and that is all”) on a page that now takes money. It now says where the data is kept, that the site is a client-side application of static files with what you read kept in your browser rather than on a server, that this is how it is for now and that adding encrypted vaults would be said there, and what can leave the browser when the reader chooses to: a payment, made on Stripe’s own page, a subscription, or a reading share. It links to a new section, what leaves your browser, and when, which sets out six cases in a table, including two I had not written down: that fetching any page tells its host, GitHub Pages, that you asked for it, and that the synthetic-readers study page embeds a viewer from another service, so opening it contacts that service.

No absolutes. The absolutes in this wording came out, including one of my own on the design page, which said that nothing leaves the browser and that this “holds”. It did not hold for the study page, and it does not hold for a payment. That row now says what can leave and when.

Removed at Dinis’s request. The paragraph under the payment button saying that the return page cannot check the payment. The limit is unchanged and is still published, on what runs and in the meter’s security model; it is only no longer on the page that takes the payment. The intro’s second sentence, which repeated what the top-up card directly beneath it says, went too.

How, and what it costs. Three text rules in admin/build/newsroom-text.json, applied by the import to the account page and the top-up page and their markdown twins, so they are recorded and not hand edits. I compared the visible text of all 315 imported pages with sgit.ai’s: exactly those two differ. A rule whose text has moved on sgit.ai stops the import with a message, which I tried by changing one. When sgit.ai adopts the wording, the rules are deleted.

v0.6.3 11 Oct 2026 A reader can now top up £5 by card, on this site’s side only, and every page that said no money moves says what runs.

The link. A live-mode Stripe Payment Link for £5, one payment, quantity fixed at one, that redirects to /account/topped-up.html with the payment reference in the address. The top-up page now shows one “Pay £5.00” button instead of the simulated cart and its four packs, and the account page’s top-up action goes straight to Stripe. That also removes a finding from the synthetic readers’ study: the top-up page no longer says it is simulated while asking for money. The return page was already here; I ran it in a browser on a fresh profile: it adds £5 the first time a payment reference arrives, says “Already added” when the same one comes back, adds again for a different one, and explains itself when there is none.

This site only, and why that matters. The link is in admin/build/newsroom-meter.json and the import writes it into the meter config as a recorded rule, not a hand edit, so sgit.ai’s pages keep the simulated cart. A reader’s credit lives in the browser storage of the site they return to, and this link returns to newsroom.sgit.ai, so a reader who paid from an sgit.ai page would have been credited on a site they were not reading. I had said the link had to be set on sgit.ai and only at cutover; that was wrong, and this is the better way. The import and the gate both refuse anything but the live shape, buy.stripe.com/<id>, so a test-mode link, which has test_ in it, cannot reach a page.

What the pages say. Four places said no money moves, and one more said the payment link was not set: the Replicate note, the Payment row on what runs, the design page, the participant card and llms.txt (twice). The top-up now sits under what runs, with its limit beside it: the page it returns to cannot check the payment, by design, and nothing is paid to the people whose facts an article uses. The participant card no longer depends on the state at all.

To switch it off: set topup_link to an empty string in that file, run the import, release. The top-up page goes back to the simulated cart, and the pages that describe payment change with it.

What was tested, and by whom. The return page I ran in a browser here, on a fresh profile. The payment itself I could not complete: Stripe declined the test card, correctly, because the link is live. Dinis tested the flow before this release and reports that it works; that is his report, and I did not see it.

v0.6.2 10 Oct 2026 Five invented readers tested this site, and found that 117 images were missing. They are fixed, the study and the article that came with it are here, and the gate now reads the places it was blind.

What was found. Five synthetic readers drove a browser through v0.6.1, one screenshot at a time, and were interviewed: 62 steps, 51 questions the site did not answer, 56 places somebody got lost, 31 findings, six blocking. It is published on sgit.ai as a vault with a read key, and the page is now here, at the same path, with its eight images and its live embed. I opened the vault with that key and checked its headline numbers against its own run records: all of them hold, including the per-reader table. The readers are invented by a model and say so first; it is not user research.

The bug was bigger than the study measured. It counted 51 missing figures in 10 articles. Forcing every figure to load in a browser, v0.6.1 had 117 figures not rendered and 2 broken card thumbnails across 13 pages; this release has 0 of 625. Two causes, one blind spot. sgit.ai’s authoring contract forbids a declarative <img src>, so a figure is a placeholder and a script builds the address from data-dir and data-shot when the page opens. The import did not treat data-dir as an address, so (1) 51 figures pointed into other vaults’ image folders that were never copied, and (2) the 66 article thumbnails on the front page at /, which I copy one level up from /articles/, still pointed at cards/. The study did not count these. The import now copies every image the moved pages point at, 52 files, to the same paths, and records which pages need each.

Why the gate passed it, and what it reads now. Check 2 reads href, src and data-src, and none of this is any of them, so the gate was green on a site showing “screenshot unavailable” to every reader of ten articles. Two new checks: 2b, every figure and card thumbnail must exist, and 2c, every script a page loads at run time must exist. 2b was shown failing first on v0.6.1’s tree (119 errors, 117 figures and 2 thumbnails) and 2c on the missing embed script, and 2c was shown to see the loaders that work. Reading the study’s page turned up a third case the same way: its “See it live” section loads vault-ui-embed.js through a name built at run time, which was not here. It is copied byte for byte. It mounts the official vault viewer from dev.vault.sgraph.ai in an iframe on that one page, the first cross-origin embed on this site, so say if you would rather link out.

New from sgit.ai (v0.7.40). One article, How to run synthetic users on your own site, with three figures, a card, a banner, a graph and its version page, and the study page above. Nothing else of the newsroom changed in that release but its version number; I diffed the moved tree against the last import: 11 new files, 0 removed, the rest the badge. Both are in the Replicate menu, and the Replicate page has a row for the method.

Fixed because they were mine. The balance chip is on every page this site writes, as a free page: it charges nothing, and the balance no longer vanishes on the Replicate page, which the study took for lost credit. What such a page should cost is one attribute, and yours to choose. Three version numbers on one screen: the front page’s edition line now ends with this site’s own release, the number in the nav and footer, instead of “edition from sgit.ai v0.7.39”, which I had added; an article byline’s “site v0.7.36” now says “sgit.ai v0.7.36”, because that is the sgit.ai release it names and links to; the markdown twins say the same. The participant page said “Nothing here runs”, contradicting the page that says what does; the card now says most of what it argues for does not run, and names what does. /shipped/ now says who is who (Dinis Cruz holds editorial responsibility; the desk’s roles, the Editor included, are worked by agents; there is no approval step) and that there is no published way to report an error, which I checked for across the reader-facing pages before writing it. Smaller: the JSON-LD said each article was part of the website sgit.ai, and the meter recorded page titles in a reader’s history with “, sgit.ai” on the end; both now say the SGit Newsroom.

Checked, and not defects. “Balance £4.49 on the card, £4.97 in the top bar” is real on the screenshot, and is the live chip against a picture of the card: pv-card.webp, a figure in the article, has the author’s own balance baked into it. In a fresh browser, and replaying the study’s reopen-and-jump pattern, chip and card agree at every depth on this site and on sgit.ai. It is still a fair point that a picture of the card is indistinguishable from the card. “Blank page after Back” is the loader’s fade-in: blank for about 0.3 seconds, not a lasting blank. The two findings the study labelled as caused by its own helper stand as labelled.

Not fixed here, and why. About twenty of the 31 findings are about pages imported from sgit.ai (the meter’s install page, the top-up page, the account wording, the article layout, the unlabelled £5.00 chip, the byline); the rule is that imported pages are fixed there and re-synced, and the meter component is a published version that never changes. They are listed, with who owns each, in the move note. Two need a decision and not a fix: a contact for pilots (the one address published anywhere is the one that runs the mailing list), and whether model identifiers belong on the site: 76 imported pages name a model version, mostly in the AI-disclosure line, including this study page, while the brief says they must not appear.

Caught before release: an import step crashed on a bad regex replacement and the build chain carried on, because I piped it through tail, which hides the failure; the chain now runs with pipefail. And a screenshot script wrote its output into a directory literally named $S inside the repository, found and deleted before it was committed.

v0.6.1 10 Oct 2026 The site now calls itself the SGit Newsroom, and SG Meter’s install instructions are back the way sgit.ai wrote them.

The name. v0.6.0 imported the newsroom with sgit.ai’s name in every page’s metadata: the home page’s tab read “Articles, sgit.ai”, and all 312 pages declared sgit.ai as their site, which is what a search result or a link preview shows. The import now renames the site in the tab title, og:title, og:site_name and the JSON-LD, and gives the front page the newsroom’s own title and description. It is a rule in sync_newsroom.py, applied on every sync, not an edit, so the gate that refuses hand edits to imported pages still holds.

A correction to v0.6.0. The import re-points links, and it reached into a code sample: on the SG Meter page the install snippet’s src="/assets/components/…" was shown as src="../assets/components/…", so a reader copying it would have copied something sgit.ai never wrote. Links are now rewritten in real tags only, never in text that merely looks like a tag. It was live for about half an hour.

Why the gate let it through, and does not now. The link check had the same flaw: it read src="…" anywhere, code samples included. The snippet’s original root path does not exist as a file here, so the honest snippet would have failed the gate, and the import’s rewrite turned it into a path that did exist. The bug passed because it was a bug. The link check now reads real tags only, and was shown still failing on a real broken link beside an identical path in a code sample.

Caught before release: the first version of the renaming replaced every "name" on the front page, which renamed the reader personas in the meter’s config, and the author, to “SGit Newsroom”. It now touches only the page’s own title and only inside the JSON-LD block.

How it was checked, and now is: for all 312 imported pages, the visible text is compared with sgit.ai’s own copy; it is identical apart from the edition line, which deliberately names this site. The meter config is byte-identical on every page, and the markdown twins differ only in where their links point.

v0.6.0 10 Oct 2026 The newsroom moves in. newsroom.sgit.ai is now the SGit Newsroom first, and everything it said before the newsroom ran is in a dated archive.

The front page is the newsroom’s. All 66 articles, the front page, the newsletter, collections, desk notes, the graphs, the backstage (six roles, their behaviour policies, the board, the log), the reader’s account and the SG Meter library now live here at the same paths they have on sgit.ai: 312 pages, 1,274 files, imported from sgit.ai v0.7.39 (commit dfd5d23), with canonical URLs pointing here. The home page looks like sgit.ai’s /articles/ because it is the same page, with this site’s navigation. The reading meter’s balance is in the top bar.

Imported, not rebuilt. sgit.ai still publishes every day, from a generator that also builds the rest of sgit.ai, so admin/build/sync_newsroom.py imports the built pages and changes only what must change: links to sgit.ai pages that did not move become absolute, canonical and image URLs point here, sgit.ai’s stylesheet arrives as assets/newsroom.css because assets/site.css is this site’s own, and sgit.ai’s site chat stays behind because here it would answer questions about the wrong site. Scripts, the meter component and the subscribe lane’s public file are copied byte for byte. The record, with two SHA-256 per file, is admin/build/newsroom-sync.json, and a new gate check 11 fails the build if anyone edits an imported page here: fix it on sgit.ai and re-sync. The generator moves here at cutover.

The archive: 75 pages archived in place, nothing moved, nothing deleted. The Governance Wire, Portugal Startups, the pt.newsroom designs, the brief pack, the library, the MVP specifications and seven design pages the running newsroom has overtaken keep their addresses and carry a dated note at the top saying when they were written, why they are in the archive and what overtook them. The three pages whose address now means something else (the home page, the newsroom design page, and /shipped/, which said “nothing on this site is running yet”) are copied into archive/ under the date they were written. Index: /archive/.

Three new pages, in the newsroom’s look. The design, and what runs sets each idea beside what runs and what does not yet. Replicate this newsroom says what someone can copy today and what is not packaged yet. What runs, what is designed is rewritten for 10 October, and says plainly that no money moves until the payment link is set. Every count on them is read from the imported files at build time.

The navigation is in the reader’s order: Newsroom, Account, How it runs, Design, Replicate, World News Day, Archive, About. The sitemap is now generated from the tree (446 pages), llms.txt leads with the newsroom and lists the archive under its own dated heading, and the front-page news strip is retired with the page it stamped: the front page now carries the newsroom’s own Latest.

The gate, made precise twice. The key tripwire now also passes sgit_rk1_, the read-key prefix in sgit’s first format, which an imported article quotes and which sgit.ai’s own credential checker classifies as read-only. And an imported page meets the agent-block rule with sgit.ai’s equivalent, an llms.txt chip plus a markdown twin that exists; a page with neither still fails.

Caught before release, by looking: the front page copied to / loaded its meter as import('assets/…'), a bare module specifier the browser refuses; the design page printed the whole article list where a count belonged; each archive run left one more blank line on every archived page; and the nav wrapped to two rows. All fixed, and a second build now writes nothing.

Not yet: the pages that stay (the concept pages, World News Day, the site pages) still have the earlier look; the meter is not on them; /account/import.html and the meter’s browser test are the rest of the brief’s Phase 3. The plan and the full content audit are in the move note.

v0.5.5 29 Sep 2026 The gate inside the published vault was broken again, by the release that added the app.

The vault’s README tells a reader to run python3 build/gates.py to re-derive every published count. v0.5.4 broke it twice over: the new published-or-private checks load data/vault-audit.json, which is a record about the vault and is deliberately not inside it; and the app was excluded from the page checks by its path, vault-app/index.html, which is where it lives in the repository and not where it lives in the vault. In the bundle it sits at the root, so the gate checked the app as if it were a page of this website and reported three failures about a file that has no business carrying site chrome.

Both fixed the same way as the v0.5.1 correction: the audit checks are skippable and say so by name, and the app is now excluded by identity rather than by path — it is whatever app.json names as its entry, which is true in both layouts. Verified by cloning the published vault with the published read key and running the gate inside it: clean, with the two skips named.

Third time this instruction has been caught failing, and the reason is always the same: the repository is not what was published, and only opening the published thing finds it. That check is now part of the release routine rather than something done when it occurs to us.

v0.5.4 29 Sep 2026 One vault, not two — and it opens as an app. Plus the release news on the front page, and the “design, not built” badge retired.

The outreach vault should never have been published. It holds who we are contacting, through which route, and what we plan to say — ours until it is sent, and about other people before it is about us. Deleting the line from a page would not have been a remedy: revocation is not retroactive, and anyone who fetched the objects keeps them. So the vault was rekeyed with sgit vault move: every object re-encrypted under a new key, pushed under a new identity, and the old vault hard-deleted from the server. The published key now recovers zero files, which we checked by cloning with it. The correction in full.

And the very next run of the publishing script re-published a key for it. Pushing a vault and publishing one are different acts and the script did not know the difference; it does now, from a table that says which vaults may carry a published key. Two new gates: a vault is PUBLISHED or PRIVATE with opposite obligations — a published one must carry a public read key the audit actually used, a private one must carry no key — and every revoked key is hunted across the whole repository, because llms-full.txt still carried this one after llms.txt had been fixed by hand.

The corpus vault is now a vault app. app.json with present: true, so the SG/Send host launches index.html from inside the vault instead of showing a file tree: the finding, the twenty-one pieces, every sentence typed and anchored, and the vocabulary — read from the vault’s own encrypted files in the reader’s browser over the sg.vfs bridge, with no server, no account and no database. It follows the authoring contract to the letter (all CSS and JS inlined, nothing declarative against a vault path, an inlined fallback so it renders saved to disk, and sg-app-ready posted when it has drawn), and the build checks all of that before it ships. The vault page now carries screenshots of it, in the shape sgit.ai uses.

Two site-wide changes. The design, not built badge is gone from the nav: it was written when everything here was a design, and it now undersells four sections that run, a published vault and a corrected record. The honest sentence on the front page and /shipped/ still draw that line properly, where there is room to draw it accurately. And the front page now opens with what changed lately — the newest six releases, generated from the release record so the page cannot claim an update that did not ship.

v0.5.3 29 Sep 2026 The outreach brief: what the collaborating agent is asked to do, to whom, in what order, and with what words.

world-news-day/outreach/agent-brief.md, in the outreach vault beside the protocol it depends on. It carries the established findings with every figure filled from the data files rather than typed; the links to send, leading with the recipient’s own piece’s page rather than with our argument; three tiers in order, the commissioner first because a licence is theirs to name and not an author’s; the nine organisations with a role address, the nine reachable only by contact page or LinkedIn, and the nine with no route at all; three message templates; what to record after every action; and the pacing, including one follow-up per target and never a second.

Two rules in it are worth naming here. A declined is final — recorded, and no other channel and no later attempt. And do not go looking for a personal address: where no route was established under the published rules, none is to be found by other means; a published organisational route can be proposed as a noted action and added at the next build.

The brief is generated, so it can be rebuilt after any recapture and its numbers cannot drift from the section’s. Both vaults were re-pushed and re-audited: 14 hits in the corpus vault and 22 in the outreach vault, all cleared, none unexplained.

v0.5.2 29 Sep 2026 The two vaults listed and described the way sgit.ai lists its thirty-seven — and the live page was pointing at a vault that did not exist.

The bug first. build/vault.py rewrote data/vault.json on every build and replaced the real vault id and read key with the BUNDLE's name and a null. The deployed page told readers to open wnd-2026-09-29 with an empty key. Nothing caught it because every check tested the repository rather than what the page claimed; a new gate now cross-checks the id and key on the page against the vault the audit actually opened, and fails if the page publishes the bundle id.

The method is not ours. sgit.ai writes down how it publishes a vault in seven steps for another agent to follow, serving two absolute rules: read keys yes and vault keys never, and audit BEFORE the key rather than after, because revocation is not retroactive. build/vault_audit.py reimplements those checks rather than importing them, so the code that classifies a credential and the code that scans a clone can be read beside the result.

What it does, all of it with the published read key and no token. Classifies both credentials; clones each vault; derives files, size, commit depth, HEAD and top-level layout from the clone rather than from the repository; scans every text file against eight patterns; and runs the negative control — an all-zeros read key, which must fail, and does, because the index address is derived from the key. 14 hits in the corpus vault and 11 in the outreach vault, every one cleared and none unexplained: they are the organisations' role addresses the contacts map publishes on purpose. Ruling a hit out is the work, so each carries the rule that cleared it, and an unexplained finding now fails the build.

And the vaults now open as pages. Each carries a _page.json at its root, so opening one shows what it is and what the rules are instead of whichever file sorts first — which for the corpus vault was MANIFEST.json, a list of hashes, an accurate and useless first impression of a piece of journalism.

Two more of the same family of bug. data/vault-audit.json was inside the bundle, so each audit found the previous audit's findings and the count grew from 14 to 39 in one cycle; it is now excluded, with the manifest that listed itself and the vault record that carried its own hash. The rule those three produced: a record about an artefact does not live inside it. And our own key tripwire fired on sgit clone sgit_public_read_… — the exact line a published vault page must carry. That is sgit.ai's own recorded mistake, and their rule applies: make it precise, not stricter.

v0.5.1 29 Sep 2026 Cloning our own published vault with our own published key found two defects that every internal check had missed.

The vault’s README tells a reader to run python3 build/gates.py to re-derive every published count. That instruction did not work. The gate resolved its paths from the repository layout, so inside the bundle it looked for world-news-day/data/ beside the bundle and died. A published instruction that fails is worse than no instruction: it invites somebody to check us and then tells them the fault is theirs. The gate now detects which layout it is in — by the presence of the whole-site build tooling, not by data/, which exists in both — and inside the bundle it names the four checks it skips and why, because they are about this website’s pages rather than about the data.

And the pushed corpus vault was two releases stale. publish_vaults.py only filled a working copy that had no vault in it yet, so every re-run pushed the same bytes: the vault was created before the fractal layer existed and silently stayed that way while its own page advertised files it did not contain. It now refreshes the contents on every run without touching .sg_vault.

Both were found the same way — by cloning the published vault with the published read key and running the check the vault tells you to run. Verified end to end: a clone with the public key now re-derives all 1,070 claim classifications, re-hashes all 43 frozen files, and re-runs every licence count, from a read-only vault a stranger can open.

Also: validate.js now fails on a page whose nav or footer block is empty. Pages are generated with empty chrome that chrome.py fills; skip that step and every page ships with no navigation and no version badge — and every other check still passed, because they all test what a page says rather than whether it has any furniture at all. That happened in v0.5.0 and was caught by reading a diff, which is not a control.

v0.5.0 29 Sep 2026 The fractal layer: all 1,070 sentences of the twenty-one op-eds, typed, anchored and connected — plus a page for each piece, and LinkedIn routes.

Until now the section treated each op-ed as one node. This builds the level below and the level above at once. Each piece becomes its own graph of claims; the pieces connect to each other through a shared vocabulary; a term is a graph of everywhere it is used. Same shape at every zoom, which is what makes it fractal rather than merely big.

1,070 sentences, each typed by a published formula in data/claim-rules.json: 828 flat assertions, 79 hypotheses, 51 quantified observations, 41 evaluations, 31 questions, 24 attributed statements — and 16 proposals. Twenty-one of the most senior voices in news, asked what journalism is for, and sixteen sentences in twenty thousand words that propose doing anything. The rule order was deliberately set so that attributed statement is tested early, ahead of proposal and hypothesis: it is the only class that points outside the piece, and ordering it late would have undercounted attributions and flattered our own argument about evidence.

How this avoids republishing anything. A claim node carries a type, an offset into the frozen prose, a length, and an anchor of at most 8 verbatim words — enough to find the sentence, far short of reproducing it. The sentence itself is stored nowhere. The gate checks every anchor is in the bytes and under the limit, and re-derives all 1,070 classifications from the frozen prose with the published rules.

And the finding about meaning. 38 vocabulary terms are used, 33 by more than one piece — and there are 4 places in 1,070 sentences where anybody says what one of those words means. The vocabulary is shared; the definitions are assumed. A looser rule found 29, but read “the decline of local news is a global phenomenon” as that piece defining local news — so it was tightened to require the term to be the subject of its own sentence, both counts are published, and the 25 given up are stated as a cost. The only term stated twice is stated by the same author both times, and the gate checks that label against the bylines rather than trusting it, because calling a repetition a disagreement would manufacture a finding.

Twenty-one piece pages under /world-news-day/pieces/: every sentence typed and anchored, the claim profile beside the corpus average, who wrote it with the published route to them, what it cites, the vocabulary it shares with pieces that never cite it, and a mapping to this site’s own argument by a published rule over the terms used — which claims only that the piece uses those words and that we have written about them.

LinkedIn, only as an organisation offers it. 8 company pages taken from frozen bytes we already held. Personal profiles are published only where the slug is an author of this corpus, letter for letter: 12 were found and dropped and 0 passed — including a near miss, because three of the four profiles on one page belong to people with nothing to do with that newsroom, and “probably the same person” is not a standard for publishing somebody’s profile as theirs.

One rule was loosened and bounded rather than quietly relaxed: each author’s role line, exactly as affiliations.json records it, joins the permission statement as an exempt quotation — because the affiliation claim is a reading of that line. The exemption is enumerated, not pattern-matched, and the gate fails if it ever grows past 400 words of the corpus’s 20,060.

v0.4.4 29 Sep 2026 Two encrypted vaults, pushed and readable — the corpus, and a second one that the agent at riskmandate.ai works in.

The corpus vault (wnd-2026-09-29) holds the twenty-one op-eds as frozen bytes, every derived dataset, the build code and the gate: 107 files. It is finished on the day it was made. The outreach vault (clsc1dg4) is the opposite kind of thing — append-only, and it will move every time a message is sent or answered. It carries 27 target files, a protocol.md, a brief, an action log with a JSON schema and a closed list of verbs, and 0 actions: nothing has been sent. Its contents are generated into world-news-day/outreach/ so anyone can rebuild them and diff.

Why two and not one. The first is a record of what somebody else published, safe to hand to anyone including the people it describes. The second is a record of what we do about it, and holds who we are contacting and what we plan to say. Different claims, different failure modes, different audiences — sharing a history and a hash chain between them would be a mistake in both directions.

Read keys are published; vault keys are not. Both pages now carry a sgit_public_read_… key, derived one-way and granting read and only read. The vault keys were never printed, copied or written to any file by the publishing script: it derives the read key in memory and leaves the credential where sgit put it. A new gate fails the build if any private key shape reaches data/*.json.

And the tripwire that would not have fired. The whole-site key-leak check had been written against a UUID-shaped vault id. When real vaults were created, sgit issued keys of the form sgit_private_vault_<…>:<id> — which that pattern does not match at all. It now matches the real prefixes by name, and the archive is no longer exempt, because a key committed into an archived copy is exactly as leaked. A tripwire satisfied by the absence of a shape nobody uses is not a tripwire.

A third gate now runs over the outreach vault before it is pushed: no address in it may be anything but an organisation's role address, by the same published two-part rule the contacts map uses.

v0.4.3 29 Sep 2026 The argument, as one piece written to be sent: Free to republish is not a licence.

Seventeen hundred words for the twenty-one people who wrote the corpus, and for anyone else who publishes open material without saying so in a form a machine can read. It opens by saying what is good about the act — twenty thousand words given away by people whose day job is deciding what stays behind a paywall — and then asks the question a machine asks. All twenty-one grant permission; none names a licence; the terms are stated three different ways; twenty-one of twenty-one already ship schema.org JSON-LD and none populates its license field.

It carries the aggregate too: nineteen of twenty-one pieces are about truth and facts, eleven raise AI, three mention copyright or licensing at all; eleven of twenty-one offer the reader no outbound link whatsoever; and of the twenty-three domains cited between them, no two pieces point at the same one. It says plainly that the diagnosis in the corpus is shared and correct, that what is missing is a mechanism, and that we are not a neutral party — naming which parts of our own stack are designs rather than things that run.

Every figure in the prose is a token filled from the data files at build time, not typed, so the article and data/*.json cannot disagree, and the build fails on an unfilled token. The markdown source ships beside it under CC BY 4.0 at world-news-day/content/ — the piece asks publishers to license their work in a form a machine can read, so it does.

The last section is the one that took the longest to write: an account of this section publishing a claim about WAN-IFRA's infrastructure more strongly than the evidence supported, withdrawing it, and the gate that now fails the build if the withdrawn claim reappears without the correction beside it.

v0.4.2 29 Sep 2026 A correction about the publisher of the corpus, made in public — and a contacts map built under rules published before the looking started.

What we got wrong. v0.4.0 said the World News Day announcement was “the one page in the beat a machine cannot read”, on the evidence of two refused attempts. wan-ifra.org is served through a JavaScript-challenge firewall that refuses some automated requests: going back to the same host for an unrelated reason, it answered 200, and a third attempt returned the full 59 KB page. A refusal observed twice is a fact about two attempts, not a property of a page. The announcement is now a registered, hashed source; every attempt is counted and recorded with the size and hash of what came back; and the terms it states — “All of the op-eds are free to republish with appropriate credit” — are read from the frozen bytes instead of quoted by hand. The hand-quoted sentence turned out to be verbatim, and the two-statement finding stands, now anchored on both sides. The correction in full, with the rule it produced, and a gate that fails the build if the withdrawn claim appears anywhere without the correction beside it.

How to reach them. Not one of the twenty-one pages offers a way to reach its author or its publisher: no contact address, no author URL in the structured record, no press line. So the map was built by going to the organisations’ own sites, one hop each, under rules fixed in data/contact-rules.json before the fetching started: organisations and never people; a published list of role local parts; four published tests before a site counts as established; and every address re-derived from frozen bytes by the gate. Result: 20 of 27 organisations established, 14 with a contact page, 11 role addresses publishable out of 43 seen, and 15 of 23 authors with any published route — all of it via an organisation.

Three pages were fetched and deliberately not kept. A page carrying ten or more addresses of named people is a staff directory; the Globe and Mail’s publishes the direct address of seventy-nine journalists. Those pages are public, so this is not secrecy — it is that freezing a staff directory into a repository and shipping it in a downloadable bundle makes it materially easier to scrape than the publisher made it. The URL, the hash and the counts are recorded so the count can be re-checked; the cost of that choice is stated rather than hidden. Organisation pages that are kept are stored gzipped with the SHA-256 of the original bytes, which the gate decompresses and re-verifies: eleven megabytes of script bundle became two, with the anchoring unchanged.

The vault bundle now carries all of it: 104 files, 2.68 MB, SHA-256 02816a3dabac7085a01010d7….

v0.4.1 29 Sep 2026 The World News Day corpus as one downloadable bundle — and the section declaring its own licence in the field it says the corpus leaves empty.

The vault: vault.zip, 64 files, 0.83 MB, SHA-256 72920f612e069f096f6a8e70…. Every frozen op-ed and REST record, every derived dataset, the build code and the gate that checks it, in one file. Built deterministically — the same frozen bytes produce the same zip, hash for hash — so the digest is a fact about the contents rather than a record of when the build ran. Two circular dependencies were caught getting there: the bundle contained the record of its own hash, and the section manifest listed itself.

Two sets of terms, on the bundle’s own face. Our description, counts, graph, transcription and code are CC BY 4.0 — named, versioned, with a URL, and declared in licence.json. The twenty-one op-eds under sources/frozen/ are not ours and are not relicensed; the README quotes the publisher’s own sentence and says plainly that it grants the reader nothing over somebody else’s work. The bundle carries the frozen third-party pages, which departs from this publication’s standing rule that a delivered vault holds no copies of somebody else’s pages, and the reason is written into the bundle rather than left implicit.

No sgit vault has been pushed, and the page says so. Creating one needs an SG/Send access token; this build environment has none, and a section about licensing does not invent a credential or publish a key. PACKAGE.sh inside the bundle carries the exact commands, and data/vault.json records "pushed": false and "read_key": null. When a vault is pushed the read key goes there — never the vault key.

And the part of the argument that can only be made by doing it. Every page in the section now carries schema.org JSON-LD with license, copyrightHolder, copyrightNotice, usageInfo and isAccessibleForFree, plus a rel="license" link — the five fields the section counts as unused on all twenty-one — and a new gate fails the build if a page loses any of them, or if the notice stops saying that the licence covers our description and not the op-eds.

v0.4.0 29 Sep 2026 A new section, and the first time this publication has turned its method on somebody else’s open corpus: World News Day 2026 — twenty-one op-eds, free to republish, under no named licence.

For World News Day on 28 September, WAN-IFRA and the Canadian Journalism Foundation commissioned twenty-one opinion pieces from editors, publishers and directors from Manila to Managua and made them free to republish. All twenty-one were fetched, frozen to a dated snapshot, hashed and described here — and none of the prose is republished. Forty-two frozen files: the page as served and the record from the publisher’s own undocumented WordPress REST API, for each piece.

The finding, derived from the bytes by two separate implementations. All 21 carry a permission to republish. 0 carry a licence with a name — no Creative Commons, no rel="license", no copyright notice, nothing with a version and a URL. The permission exists only as a sentence at the foot of the prose, in two wordings, and the announcement states the terms a third way: the articles require republication “in full” and never mention credit; the announcement requires “appropriate credit” and never says in full. And while 21 of 21 already ship schema.org JSON-LD, generated automatically by the site’s own SEO plugin, 0 of 21 use its license field. The gap between “we want this to spread” and a statement a machine can act on is one line of JSON long.

The aggregate, over twenty-one of the most senior voices in news. 19 of 21 pieces are about truth and facts; 11 raise artificial intelligence; 3 mention copyright or licensing at all. 11 of the 21 offer the reader no outbound link whatsoever — 38 links across the whole corpus, and of the 23 domains cited between them, no two pieces point at the same one. That is the measurement this publication has argued for since its first page, taken on the people making the case for trusting journalism, in the week they made it. The aggregate maps the shared diagnosis onto what we have published, and says plainly what nobody in the corpus proposes: a mechanism.

And one refusal, recorded rather than tidied away. wan-ifra.org answered an automated reader with HTTP 307 and no body, from two user-agents, while rendering normally in a browser. The announcement that grants the permission is the one page in the beat a machine cannot read, and the register says so rather than quietly dropping it.

The section ships with the estate’s grammar: 153 nodes and 420 edges over 8 types and 8 verbs, every verb with a distinct named inverse and a Portuguese form, 1,449 N-Triples, a published twenty-pattern lexicon as the only classification, and thirteen gates — among them one that fails the build if twelve consecutive words of any op-ed appear on any page we generate, and one that re-derives every licence count from the frozen HTML with a second implementation. agrees_with is a banned verb: shared vocabulary is not agreement, and this section does not put words in a named editor’s mouth.

v0.3.13 20 Sep 2026 The Startup Summit beat, consolidated into one archive: 160 files, each with its live URL, its repository URL and its hash.

Brief 15 and the bundle (210 files, 1.5 MB, also unpacked under briefs/summit-archive/) hold everything this publication did on the beat in one place: twenty pages and seven role pages as served, twenty-two data files, twelve build scripts and viewers, three stories as prose, seven documents, and the eighty-nine frozen sources that everything else stands on. Every file was fetched from the live site and checked before the bundle was cut — all 160 returned 200 and were byte-identical to the repository, and the manifest carries the SHA-256 so a reader can repeat the check rather than take our word.

Four documents carry what a file list cannot. The chronology: six days, two captures, and the release-by-release order in which the section was built. Every published number beside the file it derives from, including the four this publication refused to print — attendance, the speaker total as a fact about the event, countries, and anything about what happened in the room. The method and its eighteen gates, with the argument for why each rule is a gate rather than a habit. And three things that went wrong: the notice that should have shipped with v0.3.0, a rendered count that had become false, and the posture on biographies that changed under pressure and was nearly changed silently.

The boundary is on the first page, not inferred from the last. The register stops on 13 September, four days before the doors opened; there is no capture from during or after the event, and the archive says what a successor should do first — capture again, write the fourth piece, and never backfill a dated capture.

v0.3.12 14 Sep 2026 A sibling transfer: two frozen captures go to pt.newsroom.sgit.ai, and a rule for evidence that changes hands.

Memo 14, written after reading the sibling repository at its v0.3.1 and its live site. It opens with what that site built and this one does not have — a read-only JSON API with an OpenAPI document, the delivery quarantine as a build gate, and the accent and Portuguese-path gates the pack could only ask for — because the comparison is not one-way.

What goes the other way is one thing that matters and several that help. Its lead story says there is only one capture of the speaker list, 70 names, nothing to report, and that it exists to say so and wait for a second. This site froze the same pages on 8 September (60 names) and 13 September (64, +5 −1). With theirs, that is three captures in six days — 60 → 64 → 70 — and the one speaker who left on the 13th has not come back. Also in the bundle: three press pages they lack, the fourth that returned 404 and is excluded with its reason, and the Guinness pitch-marathon story that two publishers report and the event’s own agenda does not list — every byte for it already frozen in their tree, and no story written.

The rule the transfer forces, which neither site had written down: evidence transferred between sibling publications stays evidence only if the provenance travels with it and is published. The bundle’s manifest carries, per file, the URL, who fetched it, when, under which version, and the SHA-256 the origin register recorded — all fourteen re-verified before packaging, and meant to be verified again on arrival, because the hash is the only thing in it that requires trusting nobody. The memo proposes a register field and a gate that make the provenance impossible to lose in a refactor, and names the limits that belong on the page beside it.

v0.3.11 14 Sep 2026 Research deliveries travel as sgit vaults.

Both research briefs gain a section on packaging and distributing what they find as an sgit vault — a versioned, end-to-end encrypted folder the server cannot read, pushed with one command, handed over with a read key or a share token and never with the vault key — with the vault layout the newsroom expects, the exact commands for an assistant that can run a shell, and the fallback for one that cannot (produce the layout; the operator packages it). The delivery schema gains delivery.vault (vault id, commit, read key or share token); the ingestion guide says how the newsroom clones a delivery, checks the manifest and keeps the vault id and commit as provenance. Links to the sgit docs that matter: the agent surface, the guidance page, the publishing method and its two rules.

v0.3.10 14 Sep 2026 Two research briefs for outside assistants, and the contract their deliveries must meet.

For ChatGPT and for Perplexity: paste-ready briefs that say what pt.newsroom.sgit.ai needs (the eight sections in priority order, the three first articles’ open questions, the Summit this week, and what is already held), how to search Portuguese sources first, the rules on people and characterisation that do not bend, and how to hand results back — as leads with provenance, never facts, in a published JSON Schema (briefs/pt-newsroom-pack/08__research-briefs/research-schema.json, draft 2020-12, with a worked example that validates and a person with a forbidden field that does not). Every claim names one exact page and carries a verbatim excerpt the newsroom will search the frozen bytes for; a person is reduced to the three fields a primary page lists; edges are Portuguese verbs with a distinct inverse. The pack gains the folder and a page on how a delivery is ingested: saved unchanged, validated, every URL fetched and frozen, every excerpt re-found or the claim dropped, and what survives becomes issues on the desk for the research department to re-derive from the frozen page.

v0.3.9 14 Sep 2026 The briefing pack for the session that will build and run pt.newsroom.sgit.ai.

One zip, browsable unpacked under briefs/pt-newsroom-pack/, for a Claude Code session with commit access to the repository that deploys to pt.newsroom.sgit.ai. In it: the commissioning brief; the chosen front-page design with its sources, renders and the seven vendored typefaces; the code to inherit from the Portugal section, the databases section and the site chrome, with a file-by-file note on what to copy and what to rewrite; the operating model of the newsroom — three departments that each own a folder, files as the communication layer (issues, mail, run records), the run loop, and the human editor as the only one who can publish; a CLAUDE.md and .claude/settings.json for the new repository; the prompts and repository skills for the bootstrap, the scheduled run, the editor’s review and a correction; the three documented ways to run the session on a schedule (a Routine, a cron workflow with the official action, or the headless CLI), checked against the Claude Code docs, with the workflow ready to copy; the acceptance test for v0.1.0; and a handover of what exists here and the decisions already taken. Section 17 of the brief points at it. The pack carries no personal data: the new site fetches and freezes its own sources under its own notice.

v0.3.8 13 Sep 2026 All four pt.newsroom home-page directions are archived on the site, at full size, with the reason each was drawn and the reason three were not chosen.

/pt-newsroom/directions.html renders the five artboards of 13 September — the chosen broadsheet at 1440px and 390px, then the dense capture-ordered ledger, the dark graph-first cover and the magazine with the claims underlined in the text — each from its own source under pt-newsroom/design/, each with the motivation and trade-off that were written beside it on the design canvas (kept as canvas.json, so the page reads the notes rather than restating them), and each labelled chosen or not chosen. The page says it is an archive, not a menu.

Five more typefaces are vendored for the three archived directions — Archivo, JetBrains Mono, Space Grotesk, Bodoni Moda and Karla, latin subsets as variable woff2 files, all under the SIL Open Font License and attributed in LICENSES.md — so no design on the site fetches anything from a third party at render time. The chosen-design page links to the archive; the archive is in the Portugal menu, the sitemap, llms.txt and section 16 of the commissioning brief.

v0.3.7 13 Sep 2026 The pt.newsroom home page is a page, not a picture.

/pt-newsroom/ renders the chosen design at full size — the 1440px broadsheet in a frame that scrolls sideways on a narrow screen, and the 390px phone version beside it — from the design sources kept verbatim in the repository (pt-newsroom/design/Main.dc and MainPhone.dc). The builder scopes the design’s stylesheet so it cannot restyle the site chrome around it and edits nothing else; a change to the design is a change to the source file. The page says on its face that the site does not exist yet, that the copy is the Portugal section’s data of 13 September and is not maintained, and that the links inside the mock-up go nowhere.

Two typefaces are vendored: Newsreader and IBM Plex Mono, latin subsets as woff2 under assets/fonts/, both under the SIL Open Font License and attributed in LICENSES.md, loaded only by the pages that use them. The design had specified them and the site had been fetching them from Google Fonts at render time, which is the dependency the rest of the site refuses for its code; the renders in section 16 of the commissioning brief, which had shown fallback faces for the same reason, are redrawn with the real ones. The page is in the Portugal menu, the footer’s What runs column, the front page’s brief card, the sitemap and llms.txt.

v0.3.6 13 Sep 2026 The things that run get their own doors: three top-level menus, a front-page band, a footer column, and every page in llms.txt.

Until now the Governance Wire, Portugal Startups and the no-server databases were single links inside Rights & ops, one hop from anywhere and invisible from the front page. The nav is now eight groups: the three running instances sit between the argument and the record, each with its own submenu — the floor, the state map and the research runs under the Governance Wire; the wire, graph, files, connections, Summit, speakers, organisations, changes, sources, method, team, notice, limits and the pt.newsroom brief under Portugal Startups; the argument and the two consoles under Databases. The group label is always a link to the hub, so nothing is reachable only by opening a menu.

The front page gains a band, What runs on this site, with a card per instance and one for the pt.newsroom commissioning brief; the markdown twin carries the same section. The footer gains a What runs column. llms.txt now names the three running instances in its opening properties, lists every Portugal story and team page by path, and says which of the three are human-reviewed.

v0.3.5 13 Sep 2026 The pt.newsroom.sgit.ai home page has a chosen direction, and the commissioning brief now carries it.

Four home-page mockups were drafted for the Portuguese-language newsroom on 13 September — a classic broadsheet, a dense capture-ordered ledger, a dark graph-first cover, and a magazine with the claims underlined in the text — all drawn with the Portugal section’s real data of that day. The editor of record chose the broadsheet. Section 16 of the commissioning brief records the choice with two renders (1440px and 390px) and the system behind it — paper and ink values, the two faces, rules rather than boxes, the order of the page, what is deliberately not on it — so the agent that builds the site inherits a decision rather than a question. The three unchosen directions are kept on a second page of the design canvas and are not published.

v0.3.4 13 Sep 2026 The graph learns what the speakers’ pages say: topics, industries, technologies, ideas and offerings — and a connections page that answers who should talk to whom, at organisation level, as queries.

Sixty-four more frozen sources. Each speaker’s own page on the event site is now fetched, frozen and hashed like every other source (88 in the register, from 24). Two things are read from each page and nothing is reproduced: the event’s own Topics list for that speaker, verbatim (60 of 64 pages carry one; 23 distinct topics), and the words on the page that match a published lexicon of 57 patterns for industries, technologies, ideas, services and products. A derived tag carries the matched words on the edge and nothing else; it says the page contains those words, which is all it says.

The ontology grows by six types and five verbs, each with its Portuguese: Topic (speaks_on / fala_sobre), Industry (active_in / atua_em), Technology (uses / usa), Idea (advocates / defende), Service and Product (offers / oferece). Two new taxonomy classes say which is which: Themes for the event’s vocabulary, Derived by formula for ours. The graph is 329 nodes and 1,106 edges in ten packs; the two new packs are off by default. The triples grow to 4,001 and the tag nodes carry their pattern as a property, so the SPARQL console can show the formula beside the result.

Connections: who should be talking, who could be buying from whom, who could help whom. Three SQL queries at organisation level, run at build and stored with their SQL so the identical query runs in the browser console. The unit is the organisation a speaker is listed under, because the data-protection notice refuses any characterisation of a named person and “X should talk to Y” is one; the person-level join is a query away for a reader who wants it, and this page does not make it. The notice now lists the two new categories held, and the refusals are reworded rather than quietly relaxed.

Three more gates. 16: every Topic node is on the frozen page verbatim, and every topic on a frozen page is in the graph. 17: every derived edge re-derives — the lexicon pattern is run again on the frozen bytes, no match no edge, and every match must have an edge, so a tag can be neither typed in nor left out by hand. 18: the connections rows are what their SQL gives. The databases section gains three tables (person_topics, person_tags, lexicon), three SQL and two SPARQL worked queries, and a build-time cross-check that the connections SQL returns the same row counts against its tables as it did on the Portugal page.

v0.3.3 13 Sep 2026 Databases with no server: SQLite and a SPARQL store running in the reader’s browser over the Portugal section’s own files.

A new section, and the argument first. This site has no server. The Portugal section is JSON files in a git repository with a hash for each. Two real database engines now run over exactly those files in the visitor’s browser, compiled to WebAssembly — nothing uploaded, nothing queried remotely, the store discarded when the tab closes. The files are the database; the engines are readers. It is the pattern the estate already runs (sgit.ai’s RiskMandate vault, where “the browser becomes the database”; graphs.sgit.ai’s Regulation Graph, SQLite over WebAssembly, client-side and ephemeral) with the second language added, and graphs.sgit.ai’s not a graph database pitch is inherited on the page rather than quietly contradicted.

The SQL console: sql.js 1.14.2 (SQLite, MIT, vendored), twelve tables built on load from a loader spec that names the file and field behind every column, and thirteen worked queries — a count by role class, the programme as a table, who arrived between snapshots, every verb with its inverse in both languages, a path as a sentence the SQL way, and a recursive CTE that walks two hops from the event, which is the query that shows where SQL stops being the natural language for a graph.

The graph console: Oxigraph 0.5.11 (SPARQL 1.1, MIT or Apache-2.0, vendored, 3.9 MB and said so) over triples.nt, which the Portugal graph build now writes beside graph.json: 1,982 triples under one stated IRI scheme, the ontology inside the store, every inverse declared with owl:inverseOf and walked with ^, labels in English and Portuguese so a path reads aloud from the store alone. Eleven worked queries, each shown beside the same question in Cypher, which is displayed and not run: Kùzu would run it and is 73 MB unpacked, and the choice is recorded on the page.

The build is the test. Every SQL example is executed at build with Python’s sqlite3 against tables built from the same spec the browser uses; every SPARQL example with pyoxigraph over the same triples. A query that fails or comes back empty fails the build, and the count each returned at build is printed beside it on the page, so a reader’s run can agree or differ in the open. Both consoles publish window.__tools after tool:ready, read-only, the convention the estate’s graph readers set.

Also: the Portugal graph page and file explorer link to the consoles; the section README and llms.txt describe both; LICENSES.md carries the two new attributions.

v0.3.2 13 Sep 2026 Portugal Startups becomes a graph, a front page and a file explorer — and the press enters the register.

The graph: 192 nodes, 298 edges, 10 types, 14 verbs. The event, 64 speakers, 61 derived organisations, 16 sessions on 3 stages, 24 frozen sources across 2 snapshots, 7 pieces of coverage and our stories, as one graph. Every edge is a verb with a distinct named inverse in English and in Portuguese, so a path reads as a sentence walked either way in either language — the commissioning brief’s rule that in a Portuguese publication the edge verbs are Portuguese verbs, applied a release early so it is a switch and not a rewrite. Nodes arrive in packs the reader switches on block by block; the one classification the graph makes rather than reads (role_class on a person) is published as a formula over the listed title and says so on every node.

The instrument is graphs.sgit.ai’s, on purpose. Cytoscape.js 3.30.2 is vendored (MIT, attributed in LICENSES.md) and the control panel is that site’s altitude-graph vocabulary — radius exploration from any node, collapse and expand, path tracing read as sentences, a bounded path-query builder that says when it hits its cap, save and restore a view with every position. Packs replace levels; otherwise a reader who has used one should not have to learn the other. The page publishes window.__graph after a tool:ready event, read and view methods only, so the console, Playwright and an agent are equal consumers.

The files: the platform’s vault-browser idea rendered from a manifest for a section that is not in a vault yet — every file with its SHA-256 on the left, rendered / as a graph / raw on the right. Frozen third-party pages are listed and hashed and deliberately not rendered. The page says on its face that it goes when the vault browser takes over.

The front page is now a newspaper’s: a lead, the wire, the programme by day, what the press says, the room as a graph, who made it, and the brief for whoever builds the next one.

Seven third-party pages are in the register, each fetched, frozen and hashed before it could be cited, summarised in our words and never quoted; gate 15 fails the build on a long quotation. An eighth returned 404 and is recorded as excluded with the reason. Two of the seven — AICEP Portugal Global and Portugal Business News, both 29 July — report a 48-hour Guinness pitch marathon on 16–17 September that the event’s own agenda page, four days out, does not list. That is the third story, and it is the first to stand on sources with more than one publisher.

Three more gates. 13: the graph conforms to its ontology — every verb declared with a distinct inverse and Portuguese, every node a declared type naming a registered source, every edge within its verb’s domain and range, and nobody in the graph who is neither on the published list nor marked as no longer listed. 14: every session title in sessions.json appears verbatim in the frozen agenda, because a transcription is a claim. 15: coverage is frozen before it is cited and never reproduced.

Also fixed: the whole-site link scanner reads href=" inside inline JavaScript as a link; the explorer now emits single-quoted attributes from its templates rather than teaching the gate to be cleverer.

v0.3.1 13 Sep 2026 The notice that should have shipped with v0.3.0, and a commissioning brief for pt.newsroom.sgit.ai.

The correction first. v0.3.0 published the names, roles and organisations of 64 people this site did not get the data from, and it did so with no data-protection notice, no stated lawful basis and no route to object. That is the wrong order — the notice belongs before the first named person, not after — and the notice now live says so on its own face rather than quietly fixing it.

The journalistic derogation is not claimed, and that is the substantive finding. Article 24(3) of Portugal’s Lei 58/2019 conditions processing for journalistic purposes on the national legislation governing access to and exercise of the profession, which an unaccredited publication produced by agents does not satisfy. So the basis is legitimate interests, with a written three-limb balancing test, and the notice exists because Article 14(5)(b) makes the disproportionate-effort exemption conditional on making the information publicly available. Removal on request is unconditional — no reason required, none asked for. The page carries a Portuguese summary, because several of the people named are Portuguese. None of it is legal advice and no lawyer has reviewed it.

Two new gates, because a refusal without an enforcer is a sentence. Check 11 fails the build if an email address, telephone number or personal postal address appears anywhere in the section’s data — enforced where the data is parsed, not where it is rendered, so a contact detail cannot reach the data and then be hidden by a template. Verified by injecting one and confirming the build fails. Check 12 fails the build if any page that names individuals does not link to the notice, because a notice nobody can find from the page that named them is a file rather than a measure.

And the 12 May design page is updated rather than rewritten. That design is still unlaunched and still says so; what changed is that a much smaller first instance now runs, and that its open question 5 — who is the editor of record — is answered after four months, with the constraint the design did not anticipate: not as journalism.

The commissioning brief is the first document in the pack written to be executed rather than read. It cuts the fifteen-department design to three, replaces the seven-stories-in-seven-days test with one story whose whole process is visible, settles the subdomain by wildcard-certificate mechanics, sorts the agents into four tiers by what each can write, and rules that in a natively Portuguese publication the graph edge verbs are Portuguese verbs — because a path that does not read as a sentence in the reader’s language has the wrong edges, and an English ontology behind a Portuguese interface fails that test while looking finished.

v0.3.0 13 Sep 2026 A second publication: Portugal Startups — and the first section on this site whose sources are primary, frozen and hashed.

The door The Governance Wire cannot open. That section specifies a frozen state — a hashed byte copy of the source — and carries blocked: true against it, which is why every fact there is secondary and nothing is anchored. Here the path runs: fetch, freeze, hash, extract, diff. 17 frozen files across 2 dated snapshots, every SHA-256 published and re-verified by the gate on every build. A claim on this section points at bytes in this repository, not at a URL that may have moved.

The first beat is Startup Summit Lisbon 2026, which opens on 17 September. It was chosen on the criterion the 12 May 2026 brief argued for — small enough to map comprehensively, large enough to be interesting. An event that publishes its own speaker list has a closeable graph: 64 speakers and 61 organisations, derived from the cards rather than typed by hand, with the 2 placeholder values flagged rather than cleaned away.

Freezing twice produced the first story. The same page, captured on 8 and 13 September, went from 60 speakers to 64: 5 names added and 1 no longer listed. Both copies and both hashes are in the repository, which is the only reason anyone can now say that — the earlier version exists nowhere else a reader can reach. The removal is published with no reason attached, and gate check 4 fails the build if any page implies one: withdrawal, a clash, a duplicate and an editing error are indistinguishable from outside, and guessing would be the easiest way for a publication like this to harm a real person.

Open question 5 is answered. The original brief left the editor of record undecided, and this site called it the question it should be most uncomfortable about. Dinis Cruz is the named editor of record and reviews every page before it publishes — a deliberate departure from the fully-agentic sibling, because this beat is about named people in a small ecosystem rather than about regulations.

Frozen copies are .snapshot, not .html. They are unmodified bytes of another organisation’s pages held as evidence, and serving a browsable mirror of somebody’s whole site under this domain would contradict the one rule this publication has. The extension keeps them verifiable and un-servable at once. Speaker biographies are not extracted, stored or published for the same reason; gate check 10 fails the build if any node field grows long enough to be one.

Ten section checks on top of the whole-site gate, covering hash re-verification, derived-not-typed organisations, no-reason-for-removals, no-target-reported-as-result, and no claim of a Portuguese edition that does not exist.

v0.2.9 28 Aug 2026 The generator for /governance/ had never been committed. The site was green, deployed, and unrebuildable.

A bare build/ rule in a generic Python .gitignore matches at every depth. It was excluding governance/build/ — build.py, floor.py, gates.py, the whole section generator and its thirteen checks — from every commit since v0.2.6. Nothing broke, and that is the point: the generated HTML was committed and only the thing that generated it was missing, so the site deployed correctly three releases running while nobody could have rebuilt it from a clone.

It surfaced by accident. The debrief shipped in v0.2.8 links to floor.py on GitHub so the sibling sites can copy it. Checking those links found three 404s.

This is the second time the same rule has done this. It caught admin/build/ at v0.1.0, where CI failed loudly with MODULE_NOT_FOUND, and the fix was to re-include that one path by name — which is exactly why it recurred. The rule is now root-anchored (/build/), which is what the Python template meant, and no <section>/build/ can be swallowed again.

Fixing the instance twice would have guaranteed a third time, so it is now check 10 of the whole-site gate: nothing under a build/ directory, and no .py, .js or .mjs file anywhere, may be excluded by .gitignore. Verified by restoring the old rule and confirming the gate fails with three errors, then restoring the fix. A generated page whose generator is not in the repository cannot be rebuilt by anyone who clones it.

The debrief gains a paragraph on it under “where the code is”, because a document telling other sites to copy a generator owes them the reason its own links were broken.

v0.2.8 28 Aug 2026 A debrief for the sibling sites: how the agentic team became a room you click around, and how to build one.

The first document here that reports on shipped work rather than specifying work to do. Every site on this network publishes its agentic team the same way — a roster and an ordered list — and neither shows what any role is holding, why it is stuck, or what it will refuse. The floor makes the team a place instead. The debrief carries the parts that transfer: the one rule that separates the room from a mock-up (every word it speaks is derived at build time from the same files the pipeline runs on), the genre-versus-work line on the adventure-game reference, the SVG and responsive mechanics, a porting recipe with candidate mappings for five siblings, and the gate that fails the build if the drawn route stops matching the declared pipeline.

It states what has not been proven, which is the half a debrief usually omits: one implementation, one site, one day; no user testing; no screen-reader testing; Chromium only; untested past seven actors; and the dialogue is 28 fixed strings rather than a conversation.

Four screenshots — the first images ever published on this site. They live in the markdown, so an agent fetching the raw document gets them too.

A defect found and fixed on the way in. llms-full.txt says it carries “every markdown document in the brief pack”, and brief 09 had never been added to its ORDER list — so that claim had been false since v0.2.5. Both 09 and 10 are now in it; the file grew from 15,000 to 23,167 words.

Also fixed in the floor itself: with JavaScript off the verb bar rendered and did nothing. It now carries a <noscript> block saying so and pointing at the table that holds the same information — found by testing the degradation paths rather than assuming them, which is advice the debrief gives and therefore had to follow.

v0.2.7 28 Aug 2026 The Governance Wire gains a newsroom: a floor you can click around, a state map of how a story moves, a page per agent role, and the Researcher's first real run.

The floor is a point-and-click scene. Seven desks, a verb bar — look at, talk to, hand over, ask what they refuse — and a dialogue box. The genre is borrowed and everything in it is ours: our palette, our figures drawn as inline SVG, our verbs, our writing. What a desk says about its workload is read from desk.json at build time, so the room cannot report a queue the newsroom does not have. A token travels the pipeline route through the desks; the route and team.json's seven steps are compared by gate check 13, because a room that keeps drawing a workflow nobody runs is a lie that looks like decoration.

Nine states, and one shut door. The state map names, for each state, the one role that can advance it and the condition the next role will not take the work without. frozen — a hashed byte copy of the source — carries blocked: true, and nothing has ever passed it. That single closed door is why every fact in the graph reads secondary, why nothing is anchored, and why the two stories already on the wire went round it rather than through it.

A folder and a page per role, following graphs.sgit.ai's making-a-book team: the definition in full, the doors only that role can open, what is on its desk, and an honest inventory of the three directories the inherited shape has that ours does not yet fill.

The first research run is real, and its failures are published. The Researcher searched the beat for 28 August 2026, tried six URLs, read two European Commission pages in full, and found that the Article 6(5) classification guidelines due on 2 February are still marked draft — 207 days past the statutory deadline, last touched 23 July, with no date given for a final version. Four candidates were proposed with their weaknesses attached; one was recommended, and the pick is recorded separately, because proposing is the Researcher's job and choosing is not.

Two Council of the EU pages returned 403 to our fetch. They are excluded from the register and recorded as unreachable from here — our egress, not a claim about those pages. Gate check 11 now makes that structural: a URL that did not resolve cannot appear in the source register, and its note must distinguish our reach from the state of the page.

Seven new gate checks (7–13), 11 new pages, 3 new machine surfaces — workflow, desk, research. Section v0.2.0.

v0.2.6 26 Aug 2026 The commissioning brief stops being a document: /governance/ is a first, running MVP of the risk-and-governance publication.

Built to the conventions graphs.sgit.ai arrived at over v0.4–v0.6: JSON is the source of truth for structure, markdown for prose, and the HTML is a projection of both. Five machine surfaces — graph, ontology, sources, stories, team — and a generator that cannot render a claim the data does not carry. The section is self-contained: its own data, prose, generator and gate, depending on nothing but the shared stylesheet and chrome, so it can be lifted to its own domain whole.

The one convention deliberately not copied. The sibling renders markdown in the browser and fails CI if a page lags its source. This section renders at build time instead, because this site lists crawler-invisibility of client-assembled pages as an inherited existential risk and a story whose text is not in the HTML is a story an agent cannot read. Gate check 5 re-derives every story page from its markdown, so the no-drift guarantee is kept without paying that cost.

Fully agentic, and the pages lead with what that costs. Seven agent roles, each with a centre of gravity and a written list of what it refuses; no human reviews a page before publication. In that posture the Editor's refusals are the only thing standing where a duty editor would be, so they are rules rather than judgement calls — chief among them that no named organisation is assessed, there being no legal sign-off. A generated disclaimer block appears on every page and gate check 6 fails the build if it is missing from one.

The honesty the gate enforces. Every fact in the seed graph is secondary — taken from a sibling site's reading of a text, never from the text — and nothing is anchored. Check 4 fails the build if any page claims a verified anchor while no node carries one, and it caught exactly that during the build: an ontology definition described a Fact as “anchored to a byte range”, which overclaimed for this release. The definition was rewritten rather than the check loosened.

19 nodes, 22 edges, 5 sources all re-fetched and confirmed 200 on the day, 2 stories, 8 pages, 6 section gates on top of the whole-site gate.

v0.2.5 25 Aug 2026 The pack gains its first forward-looking document: a commissioning brief for a risk-and-governance publication, and with it the reciprocal statement the Risk Mandate inversion has owed since v0.2.0.

Everything in the pack until now was material this site was built from, frozen at 21 August 2026. This is a publication to build — assembled from newsroom, graphs, risks, pki, the vault layer and the personalisation pipeline, with riskmandate.ai as its named customer. It is labelled v1.1 addendum rather than dated into the v1.0 pack, and the generator was changed to make the pack line per-document so it could say so: misstating a document's own provenance is the one failure this site's gate exists to prevent.

The argument that makes it worth commissioning. The Portugal instance required a ≥200-entity graph before it could report competently, never built it, and never launched. A risk-and-governance beat inverts that constraint: 1,523 nodes and 1,944 edges of EU AI Act, three worked risk graphs and a 42-concept ontology are already public on the sibling sites. The publication starts where Portugal was still trying to get to.

The commercial claim is architectural, not promotional — a split of the grounding ladder. The publication owns Evidence and Fact for the public regulatory world; the customer owns Reality, Twin and Measure in their own vault; and a Vulnerability, by the published formula, exists only where the two meet. That is why the brief is written as a news operation whose first customer is a risk product rather than the reverse: the inversion is the whole point, and dropping it would have made this a compliance feature with a masthead.

Two things the brief refuses to smooth over. A publication that profits when risk looks severe has a structural incentive to inflate it, so the brief answers that with mechanism rather than promise — severity computed from edge count, facts-only, absence reports as the counterweight — plus a hard rule that no assessment of a named third party publishes without their right of reply as a node in the same vault. And it half-answers Portugal's open question 5: the editor of record is whoever's signature is on the current interval. The legal half stays open, and is raised to T9 as a precondition for launch rather than a detail.

Also: the eighth instance is now listed on /mvps/, and T2 is advanced rather than closed — the brief is the statement, the standalone comparison page it named is still unwritten.

v0.2.4 25 Aug 2026 New section: /mvps/ — the part of the source corpus that asks what would actually launch first.

The site was built almost entirely from the corpus's argumentative material: the thesis, corrections, provenance, the economics. A second cluster in the same repository was catalogued in the brief pack but never surfaced as pages — seven briefs, dated 10 April to 17 May 2026, in which the newsroom is specified not as one product but as a series of small, separately launchable publication instances. A bilingual country publication with a trip for a deadline; an eleven-role agentic newsroom to run it; an evidence-backed report tool small enough to ship in days; an eight-phase rebuild of a live property; seven Creative Commons business templates published so someone else could build them. Three pages: the index of the programme, the flagship instance in full, and the seed companies.

The reason it belongs on this site is the gap it measures. Fifteen public departments is what this site argues for; eight of fifteen is what the first MVP was scoped to actually run, recorded at the time as an open decision. Alongside it, two more open decisions from the same register — the launch domain and the reader-validation specification — and the newsroom brief's own six unanswered questions, republished unresolved. Open question 5 is the one this site should be least comfortable with: a daily agentic publication with no named editor of record and no answer on who carries legal responsibility for what it publishes. A site arguing that provenance must be walkable has an unclosed chain at precisely the point where it has to terminate in a person, and that is now stated on the page rather than left in a repository.

Also recorded: the two source briefs of 12 May contradict each other on whether the publication launches bilingual or in one language. Neither this section nor llms.txt resolves the disagreement; both flag it, on the same reasoning as everywhere else here — silently picking a side would be the site editing its own sources.

Boundaries applied rather than assumed. The publishing-substrate mechanics in the 17 May architecture brief belong to sgit.ai and are cited rather than re-explained, per the sibling boundary; only the third primitive — the management layer, where editorial state lives — is treated here, because that one is the decision graph given somewhere to be stored. The commercial detail attached to the live property in that same cluster is Tier 3 and is absent: the ideas are published, the pricing, clients and entity detail are not, and the omission is stated on the page rather than left to be noticed.

v0.2.3 24 Aug 2026 Front-page links were rendering flush against the viewport edge instead of aligned with the text they follow.

A section.band is full-bleed and had no horizontal padding at all. Every component inside one carries its own centred column — .blurb 660px, .note and .proof 1100px, .cards 1160px, .split 1000px — so anything without such a rule fell back to the full band width starting at x=0. Two bare <p> links on the front page started their text at 1px where the note directly above them starts at 109px, and the for-an-agent block spanned the whole viewport. Nothing overflowed and nothing warned; it simply read as a misalignment against the very components it sat between.

Fixed at the root rather than per element: the band now carries a gutter, so nothing inside one can reach the viewport edge whatever column rule it has, and the remaining block children — a bare paragraph, a list, the agent block — take the same 1100px column .note already used. Scoped to direct children, so the identical elements nested inside .note, .cards and every main.doc page are untouched. Verified across eleven widths from 1600px to 360px: the four text columns now agree to within 2px at every one, and the gutter also fixed a separate phone-width defect the measurement exposed — .blurb putting its text 2px from the edge at 390px, because a 660px max-width stops constraining anything once the viewport is narrower than 660.

This is the third defect in three releases that the gate could not see, after the wrapping nav bar and the raw-markdown documents page. All three were visual or conventional rather than structural, and all three were found by looking at the shipped site. The gate checks that a page is correct; it has no opinion about whether it looks right. That is a real limit and it is worth stating rather than discovering again — the shared stylesheet carries all of these latent on the sibling sites too.

v0.2.2 24 Aug 2026 The published brief pack was leaking the identifiers it exists to withhold. Fixed, disclosed, and the gate that let it through has been corrected.

What happened. The pack's redaction watch-list (06 §2) is a list of things not to publish, and it names each one in the course of forbidding it — a venture-capital firm, an AWS account number, an investor-facing hostname, a B2B price range. v0.2.0 published the pack verbatim, so the act of documenting the exclusion performed the disclosure: four identifiers the pack marks Tier 3 went live at /briefs/. The identifiers are now replaced with visible [redacted] markers in the published copies, and every one is recorded in the transparency note — nine redactions, in three files, with the surrounding reasoning, tiering and manifest rows fully intact. A silent redaction would have been the worse of the two options.

Why the gate did not catch it. The redaction check exempted briefs/, on the reasoning that the pack "documents the watch-list by naming these". That reasoning is inverted, and it is now recorded as a comment in validate.js so it is not reintroduced. Second defect in the same check: the pricing patterns had been written from memory of the brief's prose rather than against the strings actually in the files, so they matched nothing at all. Every entry in the list is now a literal verified to appear in the source before masking, and the published pack is checked like every other file — an unredacted identifier in briefs/ now fails the build.

Also shipped: /llms-full.txt, required by the pack's own house style (06 §3) and missing since v0.2.0 — llms.txt, the front page and all eleven pack documents in one fetch, because agent fetch tools frequently refuse URLs a search has not already returned. It was generating this file that surfaced the leak: the concatenation tripped the redaction check that the briefs/ exemption had been suppressing.

Two remaining gaps against the brief's build order are now tracked rather than silent: T5 (/provenance/the-citation-chain/) and T6 (/infographics/, the only build-order step not started).

v0.2.1 23 Aug 2026 Two defects found by looking at the shipped site rather than at the source.

The nav bar was silently breaking onto two lines below about 1400px. The row is flex-wrap:wrap, so once its contents outgrew the viewport the ★ GitHub link dropped to a second row and the header doubled in height — no overflow, no console warning, just a ragged two-row bar at the most common laptop widths. The shared stylesheet has the same latent dead zone on every sibling site, between the full-width bar and the 820px phone menu; this site hit it first because its nav row is the widest in the network. Fixed at both ends: the two longest items were trimmed (Rights & the newsroom → Rights & ops, and the stage pill to design, not built), and the stylesheet now tightens the row at 1400px and again at 1180px before collapsing to the phone menu at 1000px. Verified across an eighteen-width sweep from 1600px to 360px: one clean row at every width, and no horizontal overflow. The stylesheet half of this fix is worth carrying back to the siblings — graphs.sgit.ai wraps the same way at 1100px.

/documents/ was handing readers raw .md files. On a site whose central claim is that most articles do not provide evidence, they provide a link, and the link is never followed, that was the lazy version of exactly the failure the site exists to argue against. Adopted the reader-page convention the siblings already use: admin/build/gen_documents.py generates one page per pack document, each stating the single most important thing in it and why it is on this site, then rendering the raw markdown in-page from that same file through assets/mdreader.js — so a reader page cannot drift from the document it claims to render. Ten documents now have reader pages; the two machine-readable files (the 48-row manifest and the provenance JSON) stay raw, because wrapping data in a reader page would only get in the way. The raw file is still one click away from every page and is still the stated source of truth.

Neither defect was caught by the gate, and that is worth recording: the gate checks that a page is correct, not that it looks right, and not that a convention the rest of the network follows has actually been adopted here. Reader pages under documents/ are exempt from the agent-block rule as generated projections, matching the exemption issues-fs.sgit.ai makes; documents/index.html is hand-written and still owes one.

v0.2.0 23 Aug 2026 The real brief pack arrived, and this release replaces v0.1.0's content in full.

v0.1.0 shipped without a brief pack, under the working assumption that this site would report on the *.sgit.ai network's own activity. That assumption was wrong: the actual commissioned brief — newsroom.sgit.ai, The Future of News — describes a site about how news gets made, proven and paid for, built from ~110,000 words across three corpora, ~68,846 of them already public and dated since February 2025. Every page from v0.1.0 built on the wrong premise has been replaced, not patched; nothing about the wrong premise is preserved, since a corrections-first site keeping a superseded page live under the old thesis would be the exact failure it argues against elsewhere.

The build order from the brief, followed in full for the first five steps: the front page, /corrections/ (hub plus all five specified subpages — the 10,000-hours case, the 242-paper citation network, how a graph answers it, the self-critique on agenda, and the AI Act staleness case), /provenance/ (hub plus the £8.40 worked story, the decision graph, show-your-work and articles-as-vaults), /library/ (all ten core articles and three adjacent pieces, chronological, each carrying the provenance contract), and /economics/ (hub plus paying-the-fact-creator, trust-as-a-service, the payment rails, and the republished 2025 micropayments piece paired deliberately with the 2026 rail).

Three sections shipped as single comprehensive hub pages rather than the full multi-page trees the brief specifies: /rights/, /newsroom/ and /thesis/. That is a real, stated de-scoping — tracked as task T1 — made to get the whole architecture live and gated in one release rather than shipping half of it silently complete.

The provenance contract is now enforced by the gate, not just documented. A new validate.js check requires every block marked class="provenance" to state a first-published date and a working link to the original — this site's central argument, applied to itself. A second new check enforces the redaction watch-list: a fixed set of Tier 3 strings from the brief's own manifest (a named venture-capital firm, an infrastructure account number, live product pricing) may not appear anywhere in the published tree.

Two honest limits, stated rather than smoothed over. The brief's own N1 request — roughly 15,000 words of likely origin material sitting in an external repository not yet retrieved — means the library's chronology may still be incomplete at the start. And several republished docs.diniscruz.ai pieces are published here as this site's own synthesis, not verbatim reproductions — the brief pack recorded their metadata in full but not their complete text, and reproducing them as if verbatim would have been exactly the provenance failure this site exists to argue against.

v0.1.0 22 Aug 2026 The site, first release — pipeline first, shipped without a brief pack.

No brief pack had reached this site at the time of this release. Rather than block, v0.1.0 shipped the validate → tag → deploy pipeline, carried in from the sibling *.sgit.ai sites, plus a provisional front page, a role framework and a first story built from facts checked directly against five sibling repositories. The premise behind that content was superseded by v0.2.0 once the actual commissioned brief pack arrived and described a different site entirely. Recorded here as history, not deleted, per this site's own convention: what earlier releases got wrong is recorded rather than edited away.