{
 "slug": "reviews-on-the-record",
 "what": "The sections of this article, each with a line and a paragraph, for showing what the reader has already read. Built by admin/build/summaries.py from admin/content/articles/summaries/.",
 "sections": [
  {
   "id": "top",
   "label": "The opening",
   "words": 283,
   "line": "Every article reviewed on the record, never twice",
   "para": "One model writes these articles and some things fall through, so they get reviewed as records: published criteria any reviewer reads, one file per article and text, five at a time, newest first, each finding answered on the article. The editor and the desk use the same record, and signed readers come next.",
   "by": "newsroom.writer"
  },
  {
   "id": "in-short",
   "label": "In short",
   "words": 133,
   "line": "Criteria, records, batches, answers, on every page",
   "para": "Published criteria cover facts, voice, claims, conversions, health and people. A review is one file per article, text and reviewer, so nothing is reviewed twice; reviewers take five, newest first. The newsroom answers every finding and every article shows its reviews. The editor and desk use the same record; signed readers are next.",
   "by": "newsroom.writer"
  },
  {
   "id": "why",
   "label": "Why review now",
   "words": 242,
   "line": "One model writes, so a second reader checks",
   "para": "Publishing works, and an article has become a way to answer people. But one model with a lot of context misses things, like today's superlative caught only because I read it. So every article gets a second pass by readers that are not the writer: other models, then me and the desk, cheap enough to run when credits allow.",
   "by": "newsroom.writer"
  },
  {
   "id": "criteria",
   "label": "What a review checks",
   "words": 222,
   "line": "Six published questions any reviewer answers",
   "para": "The criteria are published as a page and as markdown. Facts against sources; voice, including what I would not say; claims and limits; conversions and translations; the article's health today; and people. Each is pass, issues or not checked, and each finding has a severity, section, exact quote, evidence and the smallest fix.",
   "by": "newsroom.writer"
  },
  {
   "id": "record",
   "label": "One article, one text, one reviewer",
   "words": 223,
   "line": "Reviewer, article and text hash: never twice",
   "para": "A review is one immutable file in the article's folder, tied to a hash of the article's sections. A reviewer never gets the same text twice: it takes its next five, newest first, from a status file, reads its own past reviews and the answers first, and leaves. That makes review a background job rather than a project.",
   "by": "newsroom.writer"
  },
  {
   "id": "arrives",
   "label": "How a review arrives",
   "words": 115,
   "line": "A pull request, a commit, by hand, or a vault",
   "para": "Outside agents send a pull request touching only review files, checked before it is merged; agents in the repository commit after the same check; one with no access hands the JSON to the editor. Later, the editor's channel append lane lets a reviewer send without an account, sealed to the newsroom.",
   "by": "newsroom.writer"
  },
  {
   "id": "answers",
   "label": "The newsroom answers",
   "words": 129,
   "line": "Every finding answered, every article shows reviews",
   "para": "The newsroom answers each finding in a status file only it writes: fixed with the version, declined with the reason, or open. The hub lists open findings, the daily pass works through them, and each article now has a Reviews section showing who, when, which version, the verdict and the findings' fate.",
   "by": "newsroom.writer"
  },
  {
   "id": "people",
   "label": "People on the record",
   "words": 244,
   "line": "The editor, the desk and later signed readers",
   "para": "This is comments on the record, not a comments section. My sign-off is an approval at a version; disagreement is a comment with a finding; journalists will review the same way. Notes can be added years later, such as I got this wrong, keeping articles alive. Signed readers who confirm or dispute are next.",
   "by": "newsroom.writer"
  },
  {
   "id": "first-run",
   "label": "The first run",
   "words": 353,
   "line": "Five reviews, eighteen findings, all fixed",
   "para": "Another model from the writer's vendor reviewed the five newest articles in about nineteen minutes: one approval, four with changes, 18 findings, 4 medium and 14 low. The medium ones were real, a misattributed quote among them. All were checked and fixed in this release, and its notes on the criteria became version 1.1.",
   "by": "newsroom.writer"
  },
  {
   "id": "limits",
   "label": "What this does not claim",
   "words": 135,
   "line": "Review makes checks visible, not articles right",
   "para": "Review shows what was checked and found; it does not make an article right. One model has reviewed one batch, and the criteria are version 1.1. Findings are claims to check. A hash is not a signature, nothing reviews the reviewers yet, and signed readers are a proposal.",
   "by": "newsroom.writer"
  },
  {
   "id": "sources",
   "label": "Where this comes from",
   "words": 76,
   "line": "A voice memo, the criteria and the code",
   "para": "From a voice memo on 18 October 2026. The criteria at /reviews/criteria.md, written from the article guide and the wording check; the status file at /reviews/status.json; the code in admin/build/reviews.py; with related articles on the editor's channel, undo cost, the human in the loop and agent policies.",
   "by": "newsroom.writer"
  }
 ]
}