{"path":"research/corpus-graph/readers/2026-10-10-scales-critiques-and-outside-frames-as-a-reader-returned-it.md","content":"# What a reader found to bear on eleven notes on scales, critiques and outside frames (2026-10-10)\n\nA read-only reader (Sonnet), one of those sent through the notes no reader had yet read, after the founder\nasked for the graph's coverage to be widened (README § 12). It was given the brief in\n[../reader_brief.md](../reader_brief.md) and these 11 notes:\n\n- `docs/research/storytelling-and-the-story-layer.md`\n- `docs/research/cyborg-manifesto-and-the-harness.md`\n- `docs/research/the-scrutiny-gap.md`\n- `docs/research/convictions-across-scales.md`\n- `docs/research/complexity-transitions-and-why-now.md`\n- `docs/research/fractal-scales-and-temporal-frame.md`\n- `docs/research/bengio-safety-from-honesty-and-deliberus.md`\n- `docs/research/convergence-wager-red-team.md`\n- `docs/research/wright-god-test-noosphere-and-deliberus.md`\n- `docs/research/non-zero-sum-economics-and-civilizational-cooperation.md`\n- `docs/research/legibility-under-power-red-team.md`\n\nIt made 136 tool calls. Its report is below, unchanged. What was kept from it, after the quoted words were\nchecked by code and both passages of each finding were opened, is in `../found.jsonl` (lines whose `where`\nnames this file).\n\n---\n\n11 of 11 notes read in full.\n\n### F1\ntarget: docs/research/the-scrutiny-gap.md:133\ntarget_words: \"Whether the vision language adopts *checking that keeps pace with persuasion* as the stated precondition\"\nolder: docs/vision.md:96\nolder_words: \"One consequence of the conviction is now stated as a precondition rather than a hope (ratified 2026-08-18)\"\nkind: already-answers\nwhy: The scrutiny gap lists adopting this phrase in the vision language as an open founder call. The public vision page says it was ratified on 2026-08-18, the day the note was written, and the always-loaded convictions list carries it too. Anyone treating the call as pending, or drafting other wording, would be reopening settled conviction text. The note's other founder call, whether to speak publicly about the Anthropic adjacency, is still an unchecked item in the to-do file.\nstrength: strong\n\n### F2\ntarget: docs/research/legibility-under-power-red-team.md:165\ntarget_words: \"The sacredness brake creates an incentive gradient: declare it sacred and it cannot be decomposed uninvited.\"\nolder: docs/research/parity-and-the-zone-interior.md:47\nolder_words: \"I don't see any fundamental need to put up stop signs for ourselves in facilitating either/both however\"\nkind: narrows\nwhy: On 2026-08-25 the founder said even the sacred should not be artificially protected against analysis; the weighing-design note records this as ratified in substance. The brake survives only as consent and timing: who opens a descent, and when. Declaring something sacred therefore buys delay, not immunity, which weakens the free-immunity premise here and the same premise in convergence-wager-red-team.md:174 and fractal-scales-and-temporal-frame.md:147. Another note parks any sacredness detector as detection before payoff removal.\nstrength: strong\n\n### F3\ntarget: docs/research/the-scrutiny-gap.md:79\ntarget_words: \"Calibration presupposes a defined quantity, and the prompt's entire instruction was\"\nolder: docs/research/what-a-link-strength-means.md:51\nolder_words: \"of quantity as the self-evaluation score removed on 2026-08-18, a number requested with no rubric, so\"\nkind: same-mechanism-elsewhere\nwhy: The note removes one undefined model-written number and sorts the remaining numbers into classes. The 2026-09-28 audit found another of the same kind: the stored link strength, written by prompts that say only \"0.0-1.0\". It is printed as a percentage, ranks four feeds and gates the automatic linker. The audit calls it the same kind of quantity as the removed score. The founder ruled \"Yes\" to replace it with a reading computed from answered questions. The scrutiny gap's list was therefore not complete.\nstrength: strong\n\n### F4\ntarget: docs/research/non-zero-sum-economics-and-civilizational-cooperation.md:424\ntarget_words: \"Curry et al. (2019) found seven moral rules across 60 societies with ZERO counter-examples.\"\nolder: docs/research/decomposing-value.md:167\nolder_words: \"wherever these behaviours appear, their moral valence is uniformly positive, no society reverses it\"\nkind: later-correction\nwhy: The corpus later corrected how this result is stated, after the overclaim reached the live public page. Both the 60-society and the 256-society studies found most of the seven rules in most societies, not all of them everywhere. The strong result is uniformly positive valence wherever a rule appears. \"Confirmed\" and \"replicated\" overstate it as support for convergence; the sound reading is shared structure with divergent weights. Anyone reusing this bullet for funders would repeat the corrected error.\nstrength: strong\n\n### F5\ntarget: docs/research/non-zero-sum-economics-and-civilizational-cooperation.md:164\ntarget_words: \"(Lean kernel analogy) — the community can argue about the platform's own foundations\"\nolder: docs/research/fractal-priors-for-success.md:57\nolder_words: \"(no epistemic constitution, no amendment procedure, operator-only governance: the red team's Hole 7 and Ostrom converge\"\nkind: contradicts\nwhy: This table row scores Deliberus as meeting Ostrom's collective-choice principle. The August success-prior note scores the same principle as the named gap: no epistemic constitution, no amendment procedure, operator-only governance. The legibility red team's Hole 7 says the operator's constitutive moves are challengeable in-graph by no one. On 2026-08-30 the founder also held off on storing beliefs about Deliberus inside Deliberus. The row states as present what the corpus records as missing.\nstrength: strong\n\n### F6\ntarget: docs/research/complexity-transitions-and-why-now.md:129\ntarget_words: \"An exact-match or high-cosine lookup would find almost nothing, and the corpus already has the measurement proving it.\"\nolder: docs/research/auto-connect-upgrade.md:224\nolder_words: \"structural kinship ≈ 3.5/6 with a diagnosed failure mode**: claims with a single incident edge carry near-information-free signatures\"\nkind: narrows\nwhy: This note says reuse must match structurally. Two days later the linker gained a structural-kinship generator and a concept-route generator, measured on the Israel/Palestine source. Of 435 merged candidate pairs, 253 were never surfaced by cosine. By hand, concept-route precision was about 6 of 6 but structural kinship only about 3.5 of 6. So match through shared concepts first; structure alone is the weakest generator. As of 2026-08-29 no live edge carried either generator's mark.\nstrength: strong\n\n### F7\ntarget: docs/research/storytelling-and-the-story-layer.md:89\ntarget_words: \"if the platform ever generates narrative renderings of graph regions\"\nolder: docs/research/synthesis-build-plan.md:103\nolder_words: \"Deliberus does not actually decline — it synthesizes at `/query` — so the choice is not whether to take the cost reduction\"\nkind: narrows\nwhy: The note treats prose renderings of graph regions as a hazard that might arrive. The build plan shows one already ships: the query endpoint writes prose answers from the graph. Its design answer to the same DeepMind steering result is a synthesis that must cite claims and declare what it dropped, to be compared against a restating synthesis in a planned test. It also warns that a published synthesis is one-way. A ruling on this hazard should address that existing surface.\nstrength: strong\n\n### F8\ntarget: docs/research/fractal-scales-and-temporal-frame.md:215\ntarget_words: \"If it finds nothing, the metacrisis claim looks more like a mood than a structure.\"\nolder: docs/research/dogfood-run-6-israel-palestine-cross-domain.md:98\nolder_words: \"What died is the assumption that the shipped machinery can detect it.\"\nkind: narrows\nwhy: The note proposes testing the metacrisis claim by cross-domain auto-connect and reading a null as evidence against it. Run 6 had already run that kind of test: 4 cross-domain candidate pairs against 402 inside the debate, none at threshold. It explained why: homologous arguments score about 0.60 on cosine because cosine measures vocabulary. The claim survives as a claim about structure; what failed was the detector. A null from embeddings is uninformative, so only the later structural and concept-route generators could test it.\nstrength: strong\n\n### F9\ntarget: docs/research/bengio-safety-from-honesty-and-deliberus.md:69\ntarget_words: \"Held up today by two non-LLM confession channels (verbatim source retained beside every extraction; human correction)\"\nolder: docs/research/the-load-bearing-unsaid.md:134\nolder_words: \"The system stored no raw source text anywhere — only extracted structure plus verbatim fragments.\"\nkind: later-correction\nwhy: On 2026-08-13 the corpus found it had kept only extractor-chosen verbatim fragments, never the source text; eight sources were unrecoverable. Whole-source retention began that day, and the 16 backfilled sources carry the backfill date. The July claim of verbatim source beside every extraction held only for quoted fragments, which are what the extractor judged to be argument. Anyone relying on this confession channel should know it was thinner than stated, and that the framing was the discarded part.\nstrength: strong\n\n### F10\ntarget: docs/research/convictions-across-scales.md:98\ntarget_words: \"Deliberus currently **hosts** both poles and couples them barely at all\"\nolder: docs/research/coupling-the-poles.md:19\nolder_words: \"Analytical output must have a typed path back into an attunement surface.\"\nkind: later-correction\nwhy: Written two days after this check fired, the note answers it. It reads five biological couplings and yields one design per surface, chiefly a return edge: render a decomposed argument through the other side's lens. A later note calls the indictment stale and says to cite it with its date. The founder judged the coupling work already begun in the new modalities. Build the return edge first, lens second, is ledger L9, still open. What is left to rule on is the design, not the diagnosis.\nstrength: strong\n\n### F11\ntarget: docs/research/bengio-safety-from-honesty-and-deliberus.md:56\ntarget_words: \"Deliberus's extraction already distinguishes attributed source claims from assertions, but the paper's *latent-variable* discipline\"\nolder: docs/research/self-similar-decomposition-and-claim-ontology.md:232\nolder_words: \"since the reporter's accuracy and the content's truth are independent — disproving the content must never suggest the quote was fabricated.\"\nkind: already-answers\nwhy: The founder ratified always-mint on 2026-08-15: reported content becomes its own claim, with the reported party as asserter, the source as provenance, and a link that carries no strength. That is the paper's separation of who said X from whether X holds, already decided. What is unsettled: the same note's audit found the no-strength rule held only because those edges carry no scheme, enforced nowhere. The open work is that enforcement, not the split.\nstrength: strong\n\n### F12\ntarget: docs/research/legibility-under-power-red-team.md:163\ntarget_words: \"so minting many mediocre supporting subclaims moves strength until saturation binds\"\nolder: docs/research/session26-open-rulings.md:45\nolder_words: \"padding pinned by regression tests — 3 vs 12 untagged children now score identically\"\nkind: narrows\nwhy: The hole says no cost analysis exists for padding the semantics with supporting subclaims. For decomposition parts the founder ruled on 2026-08-31 (\"Go with option C, both\") that untagged parts are weakest-part-governed. It went live that day. In the worked example twelve padded parts moved the old default from 0.502 to 0.671; under the new rule 1, 3 and 12 parts all give 0.502. Still open: evidence-edge farming, account cost, and spurious requirements as the new payoff.\nstrength: strong\n\n### F13\ntarget: docs/research/non-zero-sum-economics-and-civilizational-cooperation.md:422\ntarget_words: \"High-numeracy partisans are 45% more likely to misread data when the correct answer threatens their identity (vs 25% for low-numeracy).\"\nolder: docs/research/convergence-wager-red-team.md:71\nolder_words: \"a large preregistered replication found ideologically congruent responding but *not* the numeracy amplification\"\nkind: later-correction\nwhy: The bullet goes on to say intelligence amplifies polarization. The July red team records that a large preregistered replication (Persson et al. 2021) found congruent responding but not the numeracy amplification, while other replications succeeded. Identity-protective responding is robust; \"smarter means worse\" is uncertain. A design that assumes the most capable users rationalize best rests on the weaker half. The caveat belongs wherever this bullet is reused.\nstrength: medium\n\n### F14\ntarget: docs/research/non-zero-sum-economics-and-civilizational-cooperation.md:276\ntarget_words: \"The `@[simp]` flywheel (from Lean) means every well-vetted argument makes the system better at automatically connecting future arguments to existing ones.\"\nolder: docs/research/the-reuse-flywheel-prior.md:168\nolder_words: \"only 4 concepts span more than one SOURCE (4%), and no concept in the graph spans\"\nkind: narrows\nwhy: This describes the flywheel as running. The August measurement says it has not started: claim-level reuse is near absent (56 similarity edges across about 4,800 claims), and the concept layer's 88% reuse is within one source. Only 4 concepts span more than one source, none more than two. That note concludes identity quality, not volume, sets the rate. Contributions do not yet make linking easier; the flywheel is an expectation with a stated falsifier, not a property.\nstrength: medium\n\n### F15\ntarget: docs/research/wright-god-test-noosphere-and-deliberus.md:25\ntarget_words: \"But by his own selection logic, an uninstrumented empathy-coach is bred to *flatter*\"\nolder: docs/research/curiosity-as-growth-fuel.md:223\nolder_words: \"did not significantly improve group consensus in either study, yet participants consistently preferred facilitated discussion\"\nkind: supplies-a-missing-piece\nwhy: The note infers that an approval-bred coach flatters, and that Gemini taking the \"wise elder\" role is no evidence it helps. A DeepMind study (N=879, real stakes) measured that dissociation: facilitation did not improve consensus yet participants consistently preferred it, and synthesis-build-plan.md records allocations steered by up to 5.5 points. The inference thus has a measured instance. It also shows preference cannot be the check, which the instrument argument needs.\nstrength: medium\n\n### F16\ntarget: docs/research/legibility-under-power-red-team.md:110\ntarget_words: \"If open CQs depress computed or perceived strength, flooding-by-descent works as an attack.\"\nolder: docs/research/derivability-of-missing-considerations.md:206\nolder_words: \"already takes the first position for CQs on inferences (an edge with unanswered CQs contributes\"\nkind: narrows\nwhy: The hole says the design has not chosen how unanswered questions count. For critical questions on inferences it has: an edge with unanswered questions contributes zero, so flooding cannot depress computed strength. Dogfood run 7 found no critical-question claim had ever received evidence. The same note leaves one fork open for the founder: whether unanswered derivable requirements cap a claim at neutral. A settled setting for edges and an open one for requirements is where a ruling on this hole should start.\nstrength: medium\n\n### F17\ntarget: docs/research/the-scrutiny-gap.md:115\ntarget_words: \"A published claim graph **is** a public listings board.\"\nolder: docs/research/the-group-premise-evidence.md:33\nolder_words: \"showing people others' estimates narrowed the spread of answers without improving accuracy\"\nkind: same-mechanism-elsewhere\nwhy: The note argues agents reading a shared public board converge on it, so their agreement is an artefact of shared reading, and uses this to harden read-only. The same mechanism is measured in people: Lorenz et al. 2011, where showing participants others' estimates narrowed the spread without improving accuracy. The evidence note makes independence before pooling a condition. The rule is not only about machine readers; whatever human contributors see before they commit, such as a strength badge or others' positions, deserves the same scrutiny.\nstrength: medium\n\n### F18\ntarget: docs/research/cyborg-manifesto-and-the-harness.md:45\ntarget_words: \"selective legibility as a rhetorical mode** — saying a thing while disowning it is the\"\nolder: docs/research/the-remaining-case-for-the-stance-guard.md:39\nolder_words: \"irony, scare quotes and a bare restatement of the opponent's premise before rebutting it are the same shape\"\nkind: same-mechanism-elsewhere\nwhy: The note sets the platform's attributable plain-assertion register against irony as a speaker's way of disowning words. The stance-guard analysis found the mirror failure inside the pipeline: irony and scare quotes are shapes the reported-speech detector misses, and always-mint cannot mint what was not detected, so an ironic line can enter the graph as a sincere position. The ruled response (2026-09-13) is a stance field at extraction plus a re-read when documented opponents appear to agree. The plain-assertion stance needs that backstop.\nstrength: medium\n\nNoticed: I created one helper script in the session scratchpad (outside the repo, not a project file) before reading the read-only rule; I did not delete it. Other pairs found, read at both ends, not in found.jsonl, left out for the 18 cap: non-zero-sum-economics-and-civilizational-cooperation.md:457 (Community Notes \"validates bridging\") against legibility-under-power-red-team.md:159 (5-20% bad raters can suppress notes); non-zero-sum:218 (bridging signal = disagreement × QBAF strength) against dogfood-run-7-swedish-election-structured-vs-unstructured.md:197 (the feed does not read the badge, which is 0.500 almost everywhere); complexity-transitions-and-why-now.md:17 (no document may borrow assembly theory's authority) against competitive-landscape.md:146 (a public page says assembly theory \"contributes one prediction\"); fractal-scales-and-temporal-frame.md:209 (stranger test strips frame) against coupling-the-poles.md:85 (nothing destroyed, frame kept in source_span/raw_text); storytelling-and-the-story-layer.md:36 (stories to culture chain) against empirical-deliberation-convergence-evidence.md:45 (storytelling and sympathetic-exemplar interventions moved animosity most, but not anti-democratic attitudes); wright-god-test-noosphere-and-deliberus.md:29 (\"strength never by a model's opinion\") against what-a-link-strength-means.md:16 (stored link strength is a model's undefined number that ranks feeds). The same coupling-the-poles.md:73-75 line \"a rendering of one region from a named position, in that position's own words\" would also settle the tension in storytelling §5 between lens renderings and \"never restate\". The ledger row for AD24 asks about \"the map tells no stories\"; synthesis-build-plan is the thing it has not looked at.\n"}