Home / Articles / When journalism becomes intelligence, built / Versions / v1.1.0
When journalism becomes intelligence, built: what changed in v1.1.0
From v1.0.0 (2026-10-15, 421049b07) to v1.1.0 (2026-10-15, b4d872995), paragraph by paragraph.
8 paragraphs added, 0 removed, 1 changed in place, 43 unchanged. About 446 words added and 0 removed. Insertions are marked like this, deletions like this; unchanged runs are folded to one line; figures appear as their file names.
1 unchanged paragraph
Summary: Francesco Marconi shared the slides of an INMA talk with a line I agree with completely: "As AI makes content production easier, journalism's competitive advantage will increasingly come from how we observe the world." His deck makes it concrete with a data-center buildout in the desert: journalism as delta science, a map of who wants what, an atlas of events each with a definition and an observable signal, a domain ontology, observations that add to what the newsroom knows, a workflow of observe, compare, check, report and update, and a community impact tracker. I wanted to see it working, so this site's agent built it: a vault on this site, an app of seventeen pages that follows the eleven slides andslides, runs each one on real data structures.structures, and adds five graphs of the same knowledge. The atlas has all 12 families and 78 event types; the ontology has six entities and six relationships; sixteen source records produce seventeen observations, each a subject, a change, a time and its evidence; and the tracker turns them into an article and two alerts. The records are synthetic, on purpose: a fictional campus in a fictional county, because fake permits under real agencies' names would be exactly the kind of record a newsroom like this should never publish. The method is the point, and a vault is a natural home for it: the observation history is a versioned record, the evidence travels with every claim, and anyone with the read key can open it, check it, and copy it for their own beat.
1 unchanged paragraph
!trail mesa-ranch
6 unchanged paragraphs, under In short {#in-short}
• Five graphs of one knowledge. A fractal graph you open level by level, and a knowledge graph you play in time.
17 unchanged paragraphs, under The argument {#argument}, From slides to a working desk {#desk}, What the observations found {#found}
The same knowledge, at every level {#graphs}
[figure ji-fractal.webp] The fractal semantic graph in the vault: the question opens into the actors and the campus, the campus into twelve event families, and here Power and utilities into its seven event types; the breadcrumb leads back up, and every node's details open on the right. | architecture
An atlas, an ontology and an observation history are three views of one graph, and the vault draws it five ways. The fractal semantic graph starts at the question and opens any node into the graph inside it: the campus into its twelve event families, a family into its event types, an event type into the observations that matched it, an observation into its records. Every level has the same shape, which is the point of a fractal graph, and it is the same idea as reading code as a fractal semantic graph: you never need the whole thing at once, only the level you are on and the way down. A radial view puts every level on one page, with the evidence in orange. A bipartite graph joins the sixteen records to the seventeen observations they support. And a delta chain lays the observations on a time axis, with an arc from each one to the earlier observation it was compared with.
[figure ji-knowledge.webp] The knowledge graph, in time: the ontology instantiated with the campus, its buildings, the projects, the companies, the county, the water district and the residents, played one observation at a time; at the acquisition, the new relationship is orange. | architecture
The one I like most is the knowledge graph played in time. It is the ontology instantiated, the campus, its three buildings, the two projects, four companies and the public bodies, and each relationship carries the observation that created it, and, if one did, the observation that ended it. Press play and it is delta science you can watch: the generation station appears with its permit, the development agreement ties the county to the campus, two buildings appear in August, the developer is bought, and the ownership edge moves from one company to another. That is what a newsroom's memory should look like: not a pile of articles, but a graph whose every edge has a date and a source.
15 unchanged paragraphs, under Where AI fits {#ai}, Why it is synthetic {#synthetic}, Why a vault {#vault}…
• Updated on 15 October. The vault became seventeen pages with a menu and a details panel, and gained the five graph pages.
3 unchanged paragraphs, under Where this comes from {#sources}