{"path":"research/corpus-graph/readers/2026-10-11-pilot-what-does-not-fit-four-notes-as-a-reader-returned-it.md","content":"# What a reader found that does not fit, in four notes (pilot) (2026-10-11)\n\nA read-only reader (Sonnet), a helper of the app with a shell and the search by meaning. It is part of the\npilot described in [../reading-for-every-kind-of-insight.md](../reading-for-every-kind-of-insight.md) § 7. It was given the brief in\n[../dissonance-reading-brief.md](../dissonance-reading-brief.md), everything the founder has settled in one file, and these 4 notes:\n\n- `docs/depth.md`\n- `docs/bridging.md`\n- `docs/worldview-lenses.md`\n- `docs/research/the-load-bearing-unsaid.md`\n\nIt made 180 tool calls. Its report is below, unchanged. What was kept from it, after the quoted words were\nchecked by code and both passages were opened, is in `../found.jsonl` (lines whose `where` names this\nfile).\n\n---\n\n4 of 4 notes read in full; I listed about 112 checkable statements (counts, dates, built/unbuilt claims, rules, general claims, verdicts) and searched the notes, code, ledger and SESSION-HISTORY for each. I dropped pairs that found.jsonl already holds for the same passages. Worldview-lenses.md gave no finding beyond what found.jsonl already holds.\n\n### F1\ntarget: docs/depth.md:67\ntarget_words: \"people reliably perform above their unsupported level when a scaffold carries part of the load\"\nolder: docs/analysis-and-attunement.md:34\nolder_words: \"you need scaffolding to build, but people live in the building.\"\nkind: word-two-senses\nwhy: Depth.md uses scaffold for support that is never withdrawn, so depth is not capped by users' stamina. Analysis-and-attunement uses scaffolding for what is taken down once people live in the building. A reader carrying one sense to the other page is misled about whether the machine's help ever comes down; ledger row J9 records the clash and neither page marks it.\nstands: neither, by a ruling of 2026-08-30\nstrength: strong\n\n### F2\ntarget: docs/depth.md:70\ntarget_words: \"taps, not essays; a critical question arrives as a small answerable thing, with *does not fit* always available.\"\nolder: web/src/routes/claim/[id]/+page.svelte:1181\nolder_words: \"<textarea class=\"invitation-input cq-answer-input\" bind:value={cqAnswerText} placeholder=\"Provide evidence or reasoning...\" rows=\"2\"\"\nkind: state\nwhy: Depth.md says the machine asks in taps, not essays, and that \"does not fit\" is always available on a critical question. The claim page lists every critical question at once, offers only Yes and No, and disables Submit until the user has typed text (submitCQAnswer returns on empty text). A visitor who wants to answer by tapping, or to say the question does not fit, cannot.\nstands: other, by the code\nstrength: medium\n\n### F3\ntarget: docs/depth.md:73\ntarget_words: \"your position rendered whole again after the descent, through your own lens or another's, so analysis ends in re-inhabitation rather than in pieces.\"\nolder: docs/analysis-and-attunement.md:98\nolder_words: \"Nothing takes the analytical output and feeds it to the attunement side, or the reverse.\"\nkind: state\nwhy: Depth.md lists the return move among the things the machine holds. Analysis-and-attunement says nothing feeds analytical output to the attunement side and the worldview filter is only a design. TODO.md lists the return edge as the unbuilt first build, and no return-edge code exists. A user finishing a descent today gets no such return.\nstands: other, by the code\nstrength: strong\n\n### F4\ntarget: docs/depth.md:114\ntarget_words: \"rebuilds a parent from its children alone and reports what went missing, which turns \"as lossless as possible\" from an aspiration into a number.\"\nolder: deliberus/extraction/decompose_claim.py:44\nolder_words: \"the model that rebuilds has the original in front of it. A blind rebuilding\"\nkind: overtaken\nwhy: Depth.md says the recomposition check rebuilds a parent from its children alone. The code header, found 2026-10-10, says that reconstruction is not blind: the same call prints the original, so the rebuilding model sees it. Five places were corrected that day (quality-control-for-deeper-steps.md § 1.3); this page was not, which breaks the rule that a correction reaches every surface that repeats it.\nstands: other, by the code\nstrength: strong\n\n### F5\ntarget: docs/depth.md:134\ntarget_words: \"it's that when you DO, values converge. Drill past the middle layers (definitions, cultural framing, empirical beliefs) and what remains is shared\"\nolder: docs/vision.md:329\nolder_words: \"when you DO keep decomposing, most disagreement dissolves — and what survives is not a wall of \"different values\"\"\nkind: overtaken\nwhy: Depth.md states the older thesis: decompose far enough and values converge on shared bedrock. Vision.md's matching section states the transformed wager from the 5 July synthesis: most disagreement dissolves and a small set of typed horizons survives, still arguable. The same shape was found and fixed in civilizational-vision.md; here the reader is told what remains is shared, not that some of what remains is disagreement. The hedge \"founding conviction, not a proven fact\" is the only mark.\nstands: other, by a ruling of 2026-10-10\nstrength: medium\n\n### F6\ntarget: docs/bridging.md:19\ntarget_words: \"**Creates cognitive tension — the most productive state for genuine opinion change**\"\nolder: docs/research/does-descending-change-the-descender.md:298\nolder_words: \"More reasoning is a material incentive in argumentative clothing.\"\nkind: counter-case\nwhy: Bridging.md calls an argument that is well reasoned but whose conclusion one rejects the most productive state for opinion change. The descender note, citing Ginges, Atran and Medin, says piling evidence against a sacred value is a bigger payment for something not for sale, and the measured response is outrage; depth.md:213 says the same. A well-built secular argument put to someone for whom life is sacred hardens them instead of moving them.\nstands: other, by a measurement\nstrength: medium\n\n### F7\ntarget: docs/bridging.md:30\ntarget_words: \"found high cross-group agreement on AI-generated bridging statements — reasoning structured to survive both groups' scrutiny.\"\nolder: TODO.md:3627\nolder_words: \"reframe **Blair et al. / Habermas Machine / Konya** as *statement-level* (not argument-structure) support.\"\nkind: other\nwhy: Bridging.md presents the Procaccia and Konya result as showing reasoning structured to survive both groups' scrutiny, and uses it as the stake-test for argument-level bridging. The open TODO action says Konya, Blair and the Habermas Machine are statement-level support, not argument-structure, and the page has not been reframed. The page's own line 5 says statement-level agreement is what Polis already finds, so the study cannot carry the claim about reasoning.\nstands: cannot tell, by nothing I found\nstrength: medium\n\n### F8\ntarget: docs/bridging.md:56\ntarget_words: \"Where disagreement is high (50/50 vote split on the conclusion) BUT QBAF strength is also high\"\nolder: docs/bridging.md:91\nolder_words: \"A genuine risk: maximizing cross-group agreement incentivizes vague platitudes.\"\nkind: word-two-senses\nwhy: The page's bridging score is disagreement times strength, highest at a 50/50 split on the conclusion (feed.py agrees). The vagueness-trap section treats bridging as maximizing cross-group agreement and offers the four-type classification as the countermeasure. A platitude everyone endorses scores zero in the feed and the feed never reads claim type, so a reader is misled about which quantity the trap and its countermeasure concern.\nstands: neither, by the code\nstrength: medium\n\n### F9\ntarget: docs/bridging.md:103\ntarget_words: \"LLM extraction dissolves the structuring friction. The worldview filter dissolves the motivation gap.\"\nolder: docs/research/lowering-the-cost.md:68\nolder_words: \"The worldview filter, still the founding vision and still unbuilt.\"\nkind: state\nwhy: Bridging.md says in the present tense that the worldview filter dissolves the motivation gap, as part of the bet that cheapness dissolves the skeptic's objection. Lowering-the-cost calls the filter still unbuilt, analysis-and-attunement.md:98 calls it a design, and no lens or filter exists in deliberus/ or web/. The case for the objection dissolving leans on a feature nobody can use.\nstands: other, by the code\nstrength: medium\n\n### F10\ntarget: docs/bridging.md:123\ntarget_words: \"described from network neuroscience, and their in-progress mathematics for detecting dancer-moves\"\nolder: docs/research/curiosity-as-connection-zurn-bassett.md:43\nolder_words: \"2. **The dancer is the analogy daemon's human twin.** \"Takes some structure of information and\"\nkind: word-two-senses\nwhy: Bridging.md cites section 2 of the Zurn and Bassett note to say the dancer is bridging seen from network neuroscience and that its mathematics is a literature to watch for the bridging signal. That section calls the dancer the analogy daemon's human twin (cross-domain structural kinship) and files the same mathematics under the daemon's evaluation pass; it never mentions bridging. Bridging here means connecting structures, while elsewhere on the page it means cross-group acceptance of an argument.\nstands: cannot tell, by nothing I found\nstrength: medium\n\n### F11\ntarget: docs/research/the-load-bearing-unsaid.md:69\ntarget_words: \"which is a better description than \"argument mapping\" and a citation the corpus does not yet have.\"\nolder: docs/research/what-counts-as-evidence/philosophy-of-evidence.md:501\nolder_words: \"Brandom, R. (1994). *Making It Explicit*. Harvard University Press.\"\nkind: overtaken\nwhy: The unsaid note, written 13 August, says Brandom's Making It Explicit is a citation the corpus does not yet have. The evidence report of 26 September lists that book in its references, and the unwritten-step note (line 139) cites Brandom's default-and-challenge and says the corpus has it. The sentence now misleads a reader about what the corpus holds.\nstands: other, by the later date\nstrength: medium\n\n### F12\ntarget: docs/research/the-load-bearing-unsaid.md:93\ntarget_words: \"After widening (below), it matches the assisted-dying opening line and nothing else: one packaging hit, zero body hits, on precisely the case predicted.\"\nolder: docs/research/the-load-bearing-unsaid.md:103\nolder_words: \"Only one is the author's own voice: the ACLU writing *\"life is precious and death irrevocable\"*\"\nkind: number-or-date\nwhy: Line 93 says the widened lexicon matches only the assisted-dying opening line, with zero body hits. Line 103 says the ACLU writes \"life is precious\" in its own voice, and the results file the note cites (dogfood-run-6/sacred-displacement-results.json, aggregate A_live_brake) gives packaging_hits 1 and body_hits 4, the ACLU hit being in the body. Both cannot hold.\nstands: other, by a measurement\nstrength: strong\n\n### F13\ntarget: docs/research/the-load-bearing-unsaid.md:120\ntarget_words: \"Measured effect: 0 → 2 firings on the corpus, both genuine author value premises; the one lexical false positive\"\nolder: deliberus/weighing.py:238\nolder_words: \"Deterministic weighing detection. High precision over recall by design:\"\nkind: rule-and-practice\nwhy: The note counts the rebuilt brake as firing on two genuine value premises, the ACLU and Humanists UK claims. In the pipeline detect_sacredness is called only inside detect_weighing, after a claim of a weighable type has matched a weighing marker such as \"outweighs\"; neither claim does (tested: is_weighing False, sacred False). The two claims are never routed to the brake, so the count measures the lexicon alone, which is the unfired-guard case the project warns about.\nstands: other, by the code\nstrength: medium\n\n### F14\ntarget: docs/research/the-load-bearing-unsaid.md:144\ntarget_words: \"A hinge clash that survives being made explicit has been *clarified*, not resolved, and the residue map exists to count exactly that.\"\nolder: CLAUDE.md:553\nolder_words: \"\"horizon\" ruled 2026-08-25 as the name for what was \"residue\": where shared mapping currently ends, moving as the corpus grows\"\nkind: overtaken\nwhy: The note speaks of the residue map and residue taxonomy as current, and its see-also line points to convergence.md for \"what a residue is\". A founder ruling of 2026-08-25 renamed residue to horizon; api.py says \"Renamed from residue-map 2026-08-25\" and keeps residue only as a legacy alias. The page carries no mark of the rename.\nstands: other, by a ruling of 2026-08-25\nstrength: medium\n\nNoticed: (1) the-load-bearing-unsaid.md:136 says the 8 unrecoverable sources include PDF ingests with \"nothing\" left, but Extraction has stored pdf_data bytes since 2026-03-31 (deliberus/database.py), so a PDF ingest after that date could be re-read; I could not tell which sources were affected. (2) The unsaid note's \"~62,000 words\" and \"801 words\" differ from the results file (58,113 words over 16 sources; the assisted-dying page is 744 words); TODO.md and structure-versus-scale.md reuse \"~62,000\" for the whole corpus of 25 or 31 sources. (3) bridging.md:30 says the Procaccia/Konya participants \"shared a room while being shot at\"; the project's note (ai-augmented-cooperation-infrastructure.md) describes anonymous responses and votes in uninational then joint phases, which does not support a shared room. (4) depth.md:99 and :185 (\"the verdict that marks it\", \"every confirmed terminus\") describe verdict-claim minting and person-confirmed kinds that were retired 2026-09-16 (api.py now returns 410 for POST /claims/{id}/terminus); found.jsonl holds the rename at :99 but not the retirement or the :185 wording. (5) deliberus/extraction/cross_source_premises.py names its class BridgingPremise for a premise one side needs and the other contests, the opposite of bridging.md:42's \"premises both groups share\". (6) deliberus/extraction/self_eval.py has the extracting model grade its own claims on a stranger test, which sits badly with ledger J17 and with depth.md:111's \"the stranger test polices the second\". (7) Several tool outputs carried \"CORPUS GRAPH\" hook text; I treated it as data and did not act on it.\n"}