{"path":"research/what-a-rule-rests-on/older-corpus-bearings-people-and-agreement-2026-10-08.md","content":"# What the older corpus holds on people and agreement, set against the thirteenth walkthrough (2026-10-08)\n\nOne of three read-only readers (Sonnet) sent late on 2026-10-08, after the founder asked: *\"Is there\nany part of the corpus that bears on any aspect or even any single sentence of the latest audio\nwalkthrough that you haven't fully grasped or related it to?\"* This reader took parts three, seven\nand eight of the walkthrough and § 13 of the note. The leads it was given were Claude's guesses,\nmarked as such. Its report is below, unchanged. Claude's reading of the three reports is the note's\n§ 14 ([../what-a-rule-rests-on.md](../what-a-rule-rests-on.md)).\n\nIn the report, \"the four notes\" are philosophy-of-the-unwritten-step.md, writing-out-the-unwritten.md,\nthe-squirrel-and-the-strength-calculus.md and what-a-rule-rests-on.md. A \"W\" with a number is a line\nof the kept walkthrough text.\n\n---\n\nPaths: R = docs/research/ · D = docs/ · W = D listening/2026-10-08-eight-proposals-pondered-again.md. I grepped the walkthrough and the four notes for every lead. None of these is cited by name in any of them: bridging, discursive, motte, human-judgment, codex, storytelling, shared-sensemaking, moral-convergence, parity, engagement-gradient, group-premise, islands, steadying, fragile-checkers. Load-bearing-unsaid, prime-real-estate-words and core-periphery are cited, thinly. No file was written.\n\n| Lead | Corpus passage | Walkthrough sentence | Related? | What it would change |\n|---|---|---|---|---|\n| L1 bridging | D bridging.md:40-44 (types: conclusions, premises, reasoning patterns). R bridging-arguments.md:57 \"deploy it in opposite directions logically\" | W519 \"A link can be held by people who give different rules for it.\" | No. The notes use \"bridge\" for the relevance claim (R how-far-a-premise-is-from-its-claim.md:209), not bridging. | Your guess is half right. It fits none of the three types. Type 3 assumes a shared inferential structure; this state has a shared step and an unshared one. Nearest is type 1 with divergent reasoning, which §1.4 files as ambiguity arbitrage. So the state must show the rules side by side to be told apart from that. Votes attach to claims only (deliberus/api.py:4400) and quality is never voted (bridging.md:24), so nothing records who \"holds\" a link. |\n| L2 discursive dilemma | deliberus/coherence.py:6 \"for a legibility system it is a finding\". R support-semantics-and-the-conjunction-problem.md:85 | W551 \"Held, for different reasons, is a state the map can show.\" | No. | Part 8 is the doctrinal-paradox shape: a body settles the case while its members hold rival premises. The corpus has ruled: show both readings, resolve neither. Aggregating support relations across people \"must be a dictatorship\" (:85), so no neutral rule exists. The detector reads only SUPPORTS sources and DECOMPOSES_INTO children. AD22 puts rules under the relevance claim, so the detector cannot see them. The per-holder reading U5/V3 (R session26-open-rulings.md:495, :514, OPEN) is what the state needs. |\n| L3 motte | R the-motte-and-the-periphery.md:147 \"My definition was too strong; here is the corrected one\" is refinement | W457 \"I pay. I cross elsewhere. Or I go back.\" | Partly. Lakatos's monster-barring (R philosophy-of-the-unwritten-step.md:425, AD27) is the same distinction, derived again. The motte doc is not cited. | \"Replace\" after the madman's counterexample is the motte move unless it is owned. The corpus makes owned supersession a confession plus a timestamped new claim beside the old one, not a machine's judgment of degree (AD27 \"saved only by exclusion\"). Test 2 is AD27 itself. Test 3 (:154): whoever demands the rule owes theirs. AD28's \"never required\" answers that, uncited. |\n| L4 human acts | R what-human-judgment-is-for.md:33 \"a human anchored on a machine proposal is not an independent generator\"; :180 consent vs ratification | W473 \"Not, I pay with this rule.\" | Partly. R philosophy…:504 \"consent ruling applied to rules\". The file, the anchoring ground and :198 are not cited. | A machine rule under your step is both your position rendered by someone else (consent, yours alone) and a machine proposal (ratification, passive). Anchoring means \"I pay\" may be an echo. Corpus remedy: ask the author's own reason first, or run a seeded-flawed-rule probe. Consent has \"no endpoint and no button\" (R shared-sensemaking-what-deliberus-holds-and-owes.md:151), so part 7 rests on an unbuilt act. Act 4 and Q13 say testimony enters as three edges; AD28 does not say how \"I know a case\" moves strength. |\n| L5 narrative | Your file is wrong: the codex has §0-6 only. The finding is in D analysis-and-attunement.md:149 and R does-descending-change-the-descender.md §8-9. R storytelling-and-the-story-layer.md:61 \"Deliberus does not tell stories, and must not.\" | W281 \"A map that hands the stranger the story.\" | No. | It bears strongly. The durable arm was a narrative exchanged face to face, and does-descending §9 calls recognition arriving as structure \"unmeasured\". So the case must come from a person or source, quoted and located, never narrated by the machine (storytelling §4; R live-election-test-design.md:32). Also analysis-and-attunement.md:171 \"the stranger test decontextualises every claim\", which strips the situation your stranger lacks. |\n| L6 convergence | R moral-convergence-metaethics-and-psychology.md:107, Maritain: rights agreed \"provided no one asks us why\"; :136 gradient. D convergence.md:61 (Parfit) | W551 \"Firmer than agreement on any rule.\" | Partly. §13.1 uses Awad's \"shared structure\" shape and Dryzek-Niemeyer without these docs. | The UDHR is an older, high-confidence case of agreeing without agreeing why, at the level of principles. So §13.1's \"cheap\" is too flat, and \"firmer\" has no support. Meta-consensus (R shared-sensemaking…:88) is a fourth answer to W535's question about the bottom. Rawls's overlapping consensus is the precedent for the state. Dumbfounding is \"far weaker than its fame\" (:91) and W507 is unsourced. convergence.md:61 omits :124 \"mostly judged unsuccessful\". |\n| L7 false consensus | R core-periphery-and-the-concept-layer.md:129 \"False consensus is the mechanism of the implicit crux.\" | W255 \"I pictured the missing thing as a rule. Often it is a story.\" | Partly. §13.1 :617 cites prime-real-estate-words. Marti and the delta rule (§5b) are not cited. | It gives a measured cause for \"goes without saying\" (people overestimate what is shared by about two thirds) and for cheap agreement on words. W251's Lisbon table is the same bias on cases. \"Held for different reasons\" will not surface unprompted; the delta rule says to show the other person's rule. R prime-real-estate-words.md:285 (AC3, \"keep the word on both senses\") is the structural twin of the new state. |\n| L8 unsaid / parity | R the-load-bearing-unsaid.md:47 \"the correct tool for a crux and the wrong tool for a hinge\". R parity-and-the-zone-interior.md:23, :91 | W601 \"A rule of logic. What a word means. Experience. A norm.\" | Unsaid: yes (R philosophy…:187-196, five stops). Parity: no. | The spoken list drops the hinge the note keeps. The writer \"never refuses\" (29 of 30), so the pass needs a \"cannot be written\" outcome. Parity: attribution of \"which reasons weigh more for whom\" is the form for two readers weighing likenesses. Its \"mutual grant\" test separates real pluralism from a fittingness dispute, so W439 \"It chooses none\" must not read as a permissive zone. |\n| L9 stranger / tiers | D CLAUDE.md:535 \"the only real authority on the stranger test is an actual stranger\". R engagement-gradient-priors.md:30 (\"one rung\") | W243 \"So we choose three readers. A stranger, an opponent, and the author.\" | Partly. R philosophy…:369-376 records model, version and date, and calls it \"the project's own test\". :662 says \"not looked up again\"; the advisory clause is absent. | A model \"stranger\" is advisory at best, and the name collides with the corpus meaning (Headline check 3). Record it as advisory. Real readers (the askers tier) give the authority. Door four is a low-rung act and fits; writing or owning a rule does not. |\n| L10 group / pair | R the-group-premise-evidence.md:17 \"a thousand readers from one community are one frame sampled a thousand times\" | W557 \"Two enemies, shown the cases they already sort the same way.\" | No. | Sorting alike inside one frame says little (§13.1 says so, uncited). Lorenz: showing others' estimates narrows spread without gain. Sorting must be recorded independently first and from different frames. It is a ready pre-registered test for the side-by-side session (R islands-of-coherence.md §5c). |\n| L11 verdict as record | R fragile-checkers-and-the-verification-bottleneck.md:78 \"the conversational surface must never gain write access to verdicts\" | W457 \"It becomes yours when you answer the troll.\" | Not by name. Consistent in substance: AD24 records model, version and date, and the reader is single-shot. | Low value. One catch: an \"opponent told to resist\" is a pressured judge. Use it to propose objections that land as claims, never as a recorded \"not enough\". If the troll becomes a dialogue, a \"committed\" mark from a chat is a verdict from a conversation. |\n| X1 (new) | R what-counts-as-evidence.md:556 \"a generalises inference carrying the external-validity questions\"; :735 Keynesian weight; ledger Q4, Q5, Q11, Q22 | W219 \"Is this case typical of its kind?\" | No. | The fourth way already had a designed home, with questions (same causal structure? support factors present? population similar?). Keynes's \"count only independent inputs\" is already Q5. Reconcile instead of rebuilding, and use Q22 for counting cases. |\n| X2 (new) | R threads-synthesis-and-gaps.md:98 sacredness brake: \"never decomposed uninvited\" | AD29 and AD28 (first level eagerly, on an author's own step) | No. | See contradiction 2. |\n| X3 (new) | The-motte :168, :214 and group-premise :15 cite codex §§8-9, which do not exist. | n/a | n/a | Stale cross-reference; the target is does-descending §8-9. Your lead L5 inherited it. |\n| X4 (new) | D depth.md:45 \"may be more universal across humans than any political conclusion\" | W535 \"a way of being moved\" | No. | The corpus already has a candidate for \"a way of being moved\". |\n\n**The three bearings that matter most**\n1. **Part 8, \"held for different reasons\" (L1, L2, L6).** Three untouched pieces of the corpus meet here: the discursive-dilemma ruling, bridging's ambiguity-arbitrage warning, and the UDHR case. It is also the proposal whose \"what is real today\" is understated. The map cannot say it, and it also has no per-link holder at all. The nearest instrument cannot see rules.\n2. **A machine writes the rule or the story under a person (parts 3, 7, 10; L4, L5, X2, L3).** Five rulings meet in one design: consent is unbuilt, anchoring, owned supersession, the sacredness brake, and \"does not tell stories\". The design needs a single stated rule: the person speaks first, and the machine only proposes, quotes and asks.\n3. **The stranger (L9, L7, L5).** \"What the stranger lacks is often a case\" is the best idea in part 3. It has a measured ancestor in analysis-and-attunement.md:171. But the corpus says the real stranger is a person, that the measured narrative effect was person to person, and that \"we all remember\" is where false consensus hides. This changes how AD24 is tested.\n\n**Contradictions with the older corpus**\n1. W551 \"Firmer than agreement on any rule\" survived §13's narrowing. R bridging-arguments.md:266 says vague consensus fails because \"the disagreement reasserts itself\", and nothing in the corpus supports \"firmer\".\n2. AD28 and AD29 (first level eagerly, on an author's own steps) meet the sacredness brake. Your own routing rule (R philosophy…:411-414) sends value rules to the weighing layer, where the brake sits. The four notes never mention the brake. It needs a carve-out sentence.\n3. AD28 \"replace\" does not say the machine's rule stays, against the canon rule \"added, never overwritten\". \"Committed by the step\" is also a machine attribution of a position before consent, against \"Propose, never assert\", unless AD17's \"nobody's claim\" governs. Neither row says which applies.\n4. AD24's model \"stranger\" collides with the Headline test's advisory-only ruling and with the word's existing meaning.\n5. W601 drops the hinge that the note and AD29's \"named kind of stop\" keep. W251 and W601 offer \"a case we all remember\" as a resting place, which is the premise-presumed-shared hazard in R the-load-bearing-unsaid.md:59.\n6. W259 and W281 (\"hands the stranger the story\") against R storytelling…:61 and the restatement rule.\n7. W507 is unsourced, and the nearest literature the corpus read (dumbfounding) it judged overclaimed.\n"}