{"path":"research/corpus-graph/readers/2026-10-10-evidence-and-meaning-notes-as-a-reader-returned-it.md","content":"# Older notes that bear on eight recent notes about evidence, starting values and meaning (2026-10-10)\n\nOne of three read-only readers (Sonnet) sent on 2026-10-10, after the founder asked for the corpus graph to\nbe filled: *\"If it looks promising, fill up the graph with more substantial connections that you see\nclearly after reading through the corpus.\"* This reader was given eight recent notes that hold unruled\nproposals and was asked which older notes bear on a specific passage of each and are not named by it. It\nread all eight in full (its own count), with 95 tool calls. Its report is below, unchanged.\n\nWhat was kept from it, after both passages of each finding were opened and the quoted words checked by code,\nis in `../found.jsonl` (lines whose `where` names this file). Why readers were sent, and what the graph does\nwith what they find: [../README.md](../README.md) § 9.\n\nIn the report, T1 to T8 are the target notes in the order given to the reader: T1\nwhat-counts-as-evidence.md · T2 starting-values-and-the-problem-of-the-prior.md · T3\nstore-the-uses-read-the-meaning.md · T4 prime-real-estate-words.md · T5\nattunement-words-and-the-concept-layer.md · T6 a-cut-through-a-concept-is-an-argument.md · T7\nhow-a-map-is-cut-and-what-it-shows.md · T8 major-transitions-heredity-and-evolvability.md.\n\n---\n\n### F1\ntarget: docs/research/a-cut-through-a-concept-is-an-argument.md:114\ntarget_words: \"The corpus treats the recomposition check as *the* discriminator. Argumentation theory has been\"\nolder: docs/research/weakest-link-arithmetic-and-the-merge-hunch.md:99\nolder_words: \"criterion of Freeman's structural kind, and Yu & Zenker are the named reason to expect a genuine\"\nkind: already-answers\nwhy: § 5 presents \"the parts-versus-premises test is unsettled in the field\" as news and leaves the recomposition check as the corpus's only criterion. The 2026-08-31 § 1d had already read the same dispute (Yu & Zenker's circularity charge, Freeman's structural reply). It records a second criterion already in use (a child that shares the parent's subject and fragments its predicate is a part) and an explicit stance of confessing on the grey zone. A session would name both criteria and that stance instead of treating the dispute as a fresh complication.\nstrength: strong\n\n### F2\ntarget: docs/research/a-cut-through-a-concept-is-an-argument.md:210\ntarget_words: \"**A re-filed cut should land on `argument_from_verbal_classification`**, which is already in the\"\nolder: docs/research/dogfood-run-6-israel-palestine-cross-domain.md:257\nolder_words: \"`argument_from_verbal_classification` where the taxonomy has `verbal_classification`\"\nkind: contradicts\nwhy: Run 6 (J14) found that exactly this spelling was a plausible but wrong hand-written name. The scheme enum value is `verbal_classification`. The first buildable should name the real enum value, and the session should check that the scheme's critical questions are the ones § 3 lists.\nstrength: strong\n\n### F3\ntarget: docs/research/what-counts-as-evidence.md:227\ntarget_words: \"**Hinges** (Wittgenstein; Pritchard) are commitments that make evaluation possible and are not\"\nolder: docs/research/philosophical-foundations-stress-test.md:266\nolder_words: \"(which nodes most move a conclusion), close to the opposite of a Wittgensteinian hinge (which no argument moves).\"\nkind: same-idea-earlier\nwhy: The July stress test (§ 7, lines 234-273) took the same Pritchard reading and proposed a fifth, challengeable residue type, \"hinge commitment / bedrock practice\", so that a stopping point is not scored as a copout. It also flagged that the platform's \"hinge score\" is sensitivity, near the opposite of a hinge. T1 uses both senses with no gloss (line 227 here, \"the existing hinge machinery\" at line 737). A session would adopt or answer the residue type and gloss or rename the hinge-score use in the resilience readout.\nstrength: strong\n\n### F4\ntarget: docs/research/major-transitions-heredity-and-evolvability.md:328\ntarget_words: \"individuality-axis transition in progress, which is the squeeze the frame's co-author declines. It\"\nolder: docs/vision.md:51\nolder_words: \"Deliberus is the next instance — a platform that lets humanity's collective reasoning exceed any individual's capacity,\"\nkind: contradicts\nwhy: The founding doc asserts the individuality-axis claim. It sets Deliberus in the line \"multicellular life, culture, civilization\" as collaboration at \"a higher level of organization\". civilizational-vision.md:98-104 (§ The Next Evolutionary Transition, from the 2012 notes) says the same. T8 narrows only fractal-priors-for-success § 9 and says the phrase is on no other surface, but the idea is on the front-door docs under other wording. A session would extend the narrowing (and ledger S3/S7) to those two docs, or the corpus states the opposite of its own \"never a major transition\" rule.\nstrength: strong\n\n### F5\ntarget: docs/research/starting-values-and-the-problem-of-the-prior.md:878\ntarget_words: \"- **In aim, yes**: the number is meant to be a truth estimate, the graph's best guess at how likely a\"\nolder: docs/research/weakest-link-arithmetic-and-the-merge-hunch.md:153\nolder_words: \"And QEM strengths are deliberately *not* calibrated probabilities — the gravity-toward-prior\"\nkind: narrows\nwhy: The earlier position (§ 1f; ledger P5 superseded) says no strength should be read as P(claim is true). bengio-safety-from-honesty-and-deliberus.md:70 adds that strength is participation-dependent: a claim can be strong because nobody attacked it yet. Honest starts fix the start, not that dependence. A session would write \"in aim, yes\" with the stated conditions (outcome checking via the oracle bridge, and a guard for unexamined-because-unattacked) and record that this reverses a settled \"deliberately not\" rather than extending it.\nstrength: strong\n\n### F6\ntarget: docs/research/what-counts-as-evidence.md:810\ntarget_words: \"found* counts only in proportion to how exposed a claim was to people with reason to attack it.\"\nolder: docs/research/incentives-analysis.md:79\nolder_words: \"So contribution will systematically over-collect supports and under-collect attacks.\"\nkind: supplies-a-missing-piece\nwhy: Q10 treats absence of attacks as a function of exposure only. § 4.2 explains a second cause: attacking your own claim is altruism. It has the datapoint (run 5's referee report: 22 claims, 14 support edges, zero attack edges). It names a cheap instrument, a support-to-attack ratio published beside strength, which fits the \"weigh, don't count\" rows. A session would add the ratio and treat zero-attack sources as biased inputs, not as exposure-free evidence.\nstrength: strong\n\n### F7\ntarget: docs/research/major-transitions-heredity-and-evolvability.md:605\ntarget_words: \"which a region's strongest claims are replaced over time. The wager predicts churn falling with\"\nolder: docs/research/convergence-wager-red-team.md:95\nolder_words: \"A convergence signal without the epistemic event it is supposed to indicate.\"\nkind: narrows\nwhy: The red team (Shumailov model collapse; reasoning outsourced to the same few models) warned that measured convergence can be real while the event it should indicate is absent. T8's own § 7 describes machine structure compounding same-hand bias. Falling churn at depth could therefore be monoculture, not truth-tracking. A session would require the churn metric to be split by provenance and model family before it counts for or against the wager.\nstrength: moderate\n\n### F8\ntarget: docs/research/starting-values-and-the-problem-of-the-prior.md:320\ntarget_words: \"found minimal convergence between experts and superforecasters after months of exchange: persistent\"\nolder: docs/research/empirical-deliberation-convergence-evidence.md:65\nolder_words: \"Skeptics assumed \\\"the world usually changes slowly, making rapid extinction unlikely\\\"\"\nkind: supplies-a-missing-piece\nwhy: T2 cites the FRI result only as \"disagreement in priors\". The corpus note (lines 58-70) records the actual cruxes. They were two competing reference classes plus different evidential standards. 80 hours moved medians by almost nothing, it was not a failure of understanding, and agreement grew as the horizon extended to 1000 years. That is a measured worked case for reference-class claims as public objects and for the \"prior-dependent\" state in § 6.6, and the session would use it as the example.\nstrength: moderate\n\n### F9\ntarget: docs/research/store-the-uses-read-the-meaning.md:461\ntarget_words: \"For senses: a sense is a group of attributed, dated uses, and one with no use behind it cannot exist\"\nolder: docs/research/sameness-merge-and-split-across-fields.md:37\nolder_words: \"But concept senses are themselves individuated by their use across claims — the sense IS the usage pattern.\"\nkind: supplies-a-missing-piece\nwhy: This is the same idea (sense as usage pattern) reached on 2026-08-26, with two things T3's \"one rule for claims, senses and real things\" lacks. First, the circularity: claim identity needs sense identity and sense identity is individuated by usage across claims, resolved as alternating refinement. Second, a false sense merge corrupts every claim using the term (line 44). It also measures a ceiling on sense granularity (line 113: human agreement 56% fine against 91% coarse, and for 96 of 191 words no middle granularity). A session would state the circularity and use the ceiling to support T3's own \"cannot tell\" caution.\nstrength: moderate\n\n### F10\ntarget: docs/research/attunement-words-and-the-concept-layer.md:142\ntarget_words: \"**It has no strategic adversary**, so a null result is informative rather than confounded.\"\nolder: docs/research/interaction-modalities-and-the-pointing-test.md:33\nolder_words: \"A null under one interface is data about that interface.\"\nkind: narrows\nwhy: § 4 makes the co-present delta session an informative null that \"should stop the programme\". The older ruling-level argument is that absence of an adversary does not remove the interface confound: a null across one modality is data about that modality, and only a null across a pre-registered set of good-faith modalities is the signal. A session would pre-register the modality set (subtraction taps, written exemplars, co-present) and phrase the stop condition accordingly.\nstrength: moderate\n\n### F11\ntarget: docs/research/what-counts-as-evidence.md:787\ntarget_words: \"target and indirect for another). Study results are **corroborative** children of the body — never\"\nolder: docs/research/support-semantics-and-the-conjunction-problem.md:62\nolder_words: \"Card-Krueger and monopsony theory still stands. **Contributions accumulate.** Implemented, and this is\"\nkind: contradicts\nwhy: In the live decomposition vocabulary `corroborative` means contributions add. The older note calls this \"where the splitting incentive lives\", and session20 ships `alternative` (strongest part decides, unpaddable) as the third value. T1 means Schum's saturating corroboration. Tagging study results `corroborative` as worded would sum them. A session would map Schum's corroborative onto `alternative`, or add and name a saturating value, before any tag is written.\nstrength: moderate\n\n### F12\ntarget: docs/research/prime-real-estate-words.md:271\ntarget_words: \"needs a holder, as a sense does. A first reading costs nothing: published ratings of how pleasant\"\nolder: docs/research/dogfood-run-6-israel-palestine-cross-domain.md:183\nolder_words: \"A brake keyed to lexical absolutes will fire on undergraduate moral philosophy and stay silent on Jerusalem.\"\nkind: narrows\nwhy: The corpus measured that word-list detection of loaded or sacred standing is register-bound. weighing-detection-semantic-tier.md:38 adds that a lexicon's silence is ambiguous between absence and blindness. AC1's rating-lexicon reading, and the word-list scan of the 45 tiles, should be labelled candidate-only. A hit from the lexicon or a miss in the scan must not be read as \"not wanted\". The holder-based reading (two senses on record, shown, neither moved) is the sturdier signal.\nstrength: moderate\n\n### F13\ntarget: docs/research/starting-values-and-the-problem-of-the-prior.md:11\ntarget_words: \"evidence, right? What's the starting value/prior expectation given an \\\"entirely fresh\\\" claim? Or is there\"\nolder: docs/inductive-reasoning-research.md:275\nolder_words: \"What serves as the \\\"prior\\\" for a brand-new claim on Deliberus? Base rates from prediction markets?\"\nkind: same-idea-earlier\nwhy: The question was already filed as open question 5 of the inductive-reasoning survey, with three candidate answers: prediction-market base rates, expert estimates, community polls. T2 evaluates all three (§ 2.6 and § 10.2, § 10.2 item 4, § 9 question 6) without citing the survey. A session would cite it as the origin and say which candidates were kept or dropped.\nstrength: moderate\n\n### F14\ntarget: docs/research/how-a-map-is-cut-and-what-it-shows.md:102\ntarget_words: \"what matters*), which is a cut by **topic**. The first pass cuts by **argument**: a conclusion and its\"\nolder: docs/research/graph-visualization.md:38\nolder_words: \"Requires meaningful clustering criteria (topic, stance, author)\"\nkind: contradicts\nwhy: The March semantic-zoom research chose \"Galaxy: topic clusters\" for the zoomed-out level (lines 109-146). mobile-argument-ux.md § 1.4 repeats it. It listed topic, stance and author as the clustering criteria, with no \"argument\". T7 measures that a topic cut misdescribes argument structure (35% of links cross the outline cut on the hand-made map, 73% on one transcript). Under the corpus's own rule that a correction is not done until every repeating surface is corrected, T7 should name those two notes and add \"argument\" as the criterion.\nstrength: moderate\n\n## Looked for and not found\n- For T7, an older design or measurement for grouping a claim's supporters into \"needed together\" versus \"each stands alone\" bundles on the claim page. Only the 2013 sketch (which T7 names) and the strength-layer tags on decomposition parts exist; the sibling linked-convergent note is dated 09-22, the same week.\n- For T5, any older measurement of within-community variation on love/understanding-type words. Only Marti and Jackson (both already cited) exist, so the \"thin exemplar\" claim has no older rival evidence.\n- For T3, any older design of the object-level graph of real things. The corpus holds only the April founder request, and T3's \"no design existed\" holds. The only nearby older work (sameness-merge's entity-resolution survey, soft-canonical-clustering) is about identity, not that layer.\n- For T4, an older treatment of a word's emotive standing or of persuasive definition. Stevenson, Weaver and Gallie as moves appear only in the new notes, apart from Gallie in the stress test. The concept-layer side of wanted words is therefore new to the corpus.\n\nTargets read in full: 8 of 8.\n"}