{"path":"research/corpus-graph/readers/2026-10-10-governance-modalities-and-premise-runs-as-a-reader-returned-it.md","content":"# What a reader found to bear on eleven notes on governance, modalities and premise runs, one of them a public page (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/governance-and-financing.md`\n- `docs/research/80000-hours-ai-decision-making.md`\n- `docs/research/progressive-disclosure.md`\n- `docs/open-problems.md`\n- `docs/research/moral-convergence-metaethics-and-psychology.md`\n- `docs/research/fractal-priors-for-convergence.md`\n- `docs/research/interaction-modalities-and-the-pointing-test.md`\n- `docs/research/dogfood-run-7-swedish-election-structured-vs-unstructured.md`\n- `docs/research/kialo-deep-dive.md`\n- `docs/research/cross-source-premise-run.md`\n- `docs/research/reading-the-text-itself.md`\n\nIt made 95 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/reading-the-text-itself.md:59\ntarget_words: \"an unaligned split still multiplies energy (one claim at 0.70 contributes 0.06; twelve children contribute 0.72)\"\nolder: docs/research/session26-open-rulings.md:45\nolder_words: \"padding pinned by regression tests — 3 vs 12 untagged children now score identically\"\nkind: later-correction\nwhy: The target (27 Aug) ranks a strength-layer guard before any new decomposition axis because splitting multiplies energy. Ledger B10, ruled 2026-08-31 (\"Go with option C, both\"), flipped the default the same day. Untagged parts now take their weakest member, so 3 and 12 children score alike (80 parents, 132 edges affected). For untagged parts the guard exists. What remains is tagging, the backfill and evidence routing. Pebble protection now has to be claimed by marking parts corroborative.\nstrength: strong\n\n### F2\ntarget: docs/research/dogfood-run-7-swedish-election-structured-vs-unstructured.md:198\ntarget_words: \"Reported, not fixed: it re-labels a shipped surface, which is a founder call.\"\nolder: docs/research/strength-layer-audit.md:83\nolder_words: \"So the grey `UNEXAMINED` state added today is a restoration, not an invention.\"\nkind: already-answers\nwhy: Run 7 (17 Aug) leaves the confident blue badge at 0.500 as an unmade founder call (also § 6.4 item 2). The strength-layer audit of the same day records the founder's question (\"structural sycophancy\") and says a grey UNEXAMINED state was added that day. It traces it to the hand-drawn VALID sketch, which had a distinct \"none addressed\" state. It cites the same 115-of-116 figure. Whoever rules on run 7's item 2 should know it was acted on.\nstrength: strong\n\n### F3\ntarget: docs/research/cross-source-premise-run.md:94\ntarget_words: \"These proposals were already excluded from badge computation, but for two accidental reasons\"\nolder: docs/research/what-confirmed-means.md:4\nolder_words: \"RULED 2026-09-28 (ledger Q32; founder: *\\\"Yes.\\\"*), which settles Q29 and makes C8 moot\"\nkind: later-correction\nwhy: The target stores the 11 proposals as confirmed=false and makes the badge query filter on `confirmed`, the deliberate guard. Q32 (RULED 2026-09-28) retires that field. It has five meanings in one bit and is set by whoever wrote the link. A machine proposal's standing moves to the trust ladder's state. The guard over the 50 cross-source edges needs a new home. Unruled Z7 goes further: proposals get their own kind of link or table, so a query cannot return them.\nstrength: strong\n\n### F4\ntarget: docs/research/interaction-modalities-and-the-pointing-test.md:55\ntarget_words: \"With two known speakers there is nothing to cluster: it is a per-segment embedding matched by cosine against two profiles\"\nolder: docs/research/live-election-test-design.md:68\nolder_words: \"the earlier shorthand \\\"enrollment deletes the clustering step\\\" was an architecture claim, not a description of the code\"\nkind: later-correction\nwhy: The target says the expensive stage is the one enrollment deletes. The live-election design, reading the swhisper code on 2026-08-27, retracts that shorthand. The profile matcher runs after pyannote to repair its clustering, and the pipeline has no profile reference. A streaming path is new code (keep segmentation, drop clustering, promote the matcher). Its thresholds were calibrated on household voices only. The above-real-time throughput argument stands, but \"the eight profiles already exist, so live is nearly free\" does not.\nstrength: strong\n\n### F5\ntarget: docs/research/progressive-disclosure.md:34\ntarget_words: \"Over 1 million users engage with argument trees, with research showing argument-mapping tools are\"\nolder: docs/research/kialo-deep-dive.md:69\nolder_words: \"Kialo does not publicly disclose active user counts. The FT article (Jan 2018, 4 months after launch) mentioned\"\nkind: contradicts\nwhy: The target twice (also line 338) uses \"over 1 million users\" as evidence that layered argument trees work at scale. The Kialo deep dive, written the same day, says Kialo does not disclose active users. Its 1,000,000+ figure is total votes (Jul 2023, line 61), not users. The wording appears to have drifted from votes to users; session7-deep-review-and-strategic-assessment.md:255 repeats it. Do not cite it as adoption evidence. The deep dive's read-and-vote versus write (90-9-1) split is the usable fact.\nstrength: strong\n\n### F6\ntarget: docs/research/interaction-modalities-and-the-pointing-test.md:95\ntarget_words: \"the modality set for session one is pre-registered before the session, in writing, with the affordance being tested named for each\"\nolder: docs/research/live-election-test-design.md:3\nolder_words: \"the pre-registered modality set (expansion logged so attempts stay countable), the matching rule\"\nkind: already-answers\nwhy: The closing caution makes the pointing test depend on a written pre-registration. The live-election design (2026-08-22) is that document. It has three modalities (typed, spoken post-hoc with diarization, live only if energy allows), logged expansion, pre-registered signals and a readiness gate. It also records that no planned live test ran (founder 2026-09-13) and that live recorded debates are \"no longer a high priority\". The owed pre-registration exists. The open fact is that session one has not happened.\nstrength: strong\n\n### F7\ntarget: docs/research/governance-and-financing.md:385\ntarget_words: \"the structural unlock for everything below; see [sff-2026-and-fiscal-sponsorship-path.md](../../sff-2026-and-fiscal-sponsorship-path.md) for full reasoning\"\nolder: docs/frontier.md:31\nolder_words: \"any charity that has agreed in advance will do (not necessarily EA Sweden; SFF's 2026 round also paid through Epistea and Ashgro\"\nkind: later-correction\nwhy: The target's summary sequence makes an EA Sweden fiscal-sponsor conversation the unlock for everything below, and lists LTFF in parallel (line 387). The public frontier page now says only SFF needs a receiving organisation and the individual-eligible grants do not wait on it. It says LTFF no longer exists. It says the founder ruled on 2 September to send individual applications, and that nothing had been sent as of 4 October. The top and line-359 notes only partly carry this through.\nstrength: strong\n\n### F8\ntarget: docs/research/fractal-priors-for-convergence.md:31\ntarget_words: \"The list is the structure. The how-much is the weighting.\"\nolder: docs/research/decomposing-value.md:156\nolder_words: \"needs/satisfiers distinction is named as his main contribution. The gallery's shape is his thesis verbatim.\"\nkind: same-idea-earlier\nwhy: The target presents \"same short list, different weights\" as a pattern found by sweeping domains. The follow-up note (same day, names the gallery, not named back) checked Max-Neef against the primary source. His theory holds nine needs, finite, few and the same in all cultures, with all variation in satisfiers. So the gallery restates a named, contested anthropological thesis. That matters when counting the Schwartz, Haidt, Curry and gallery rows as independent strands.\nstrength: strong\n\n### F9\ntarget: docs/open-problems.md:21\ntarget_words: \"The verdict survived every filing for 91% of them, the typical score did not move at all\"\nolder: docs/research/what-the-necessary-default-changes.md:13\nolder_words: \"whole's strength. Before this ruling, the answer for untagged parts was **add them up**. Now it is\"\nkind: later-correction\nwhy: The public page quotes the August slicing study, which ran on the 22 Aug code. On 2026-08-31 (ruled option C) untagged parts began to take their weakest member. From 17 Sept a confirmed untagged part also caps its whole. So the study's \"as filed\" reading is no longer today's computation for untagged edges, and the two sensitive claims' spread describes the old default. The page promises a re-runnable instrument. These figures should be re-run before being quoted as current.\nstrength: strong\n\n### F10\ntarget: docs/open-problems.md:79\ntarget_words: \"the same handful of moral foundations shows up in everyone while political tribes load them differently\"\nolder: docs/research/moral-convergence-metaethics-and-psychology.md:83\nolder_words: \"MFT's own authors have since rebuilt the instrument (MFQ-2), finding that \\\"the nomological network of moral foundations varied across cultural contexts\\\"\"\nkind: narrows\nwhy: The public sentence is stronger than the in-house review supports. § 4.1 reports panel studies where foundations are less stable than political attitudes and ideology predicts later foundations (\"Ideology Justifies Morality\"). It reports inflated effect sizes in the flagship dataset and an instrument rebuilt after cross-cultural variation. The review's firm universal-structure evidence is Schwartz, Curry and Moral Machine, not Moral Foundations Theory. § 7 adds that these are aggregate findings.\nstrength: medium\n\n### F11\ntarget: docs/open-problems.md:79\ntarget_words: \"dig far enough beneath any two worldviews' disagreement and what remains is small, nameable, and largely shared\"\nolder: docs/research/fractal-priors-for-convergence.md:63\nolder_words: \"So the gallery supports \\\"what remains is typed and nameable\\\" far more strongly than it supports \\\"what remains is small\\\"\"\nkind: narrows\nwhy: The page's \"for reasons\" are the cross-domain structure-versus-weights pattern, which is the gallery in this note. This note says the gallery shows the shape of disagreement, not its size. Two cultures can share the whole value circle and still go to war over the ranking: \"Weight-fights are real fights.\" So \"small\" must rest on the descents and the structured-elicitation result. The pattern does not carry it.\nstrength: medium\n\n### F12\ntarget: docs/research/fractal-priors-for-convergence.md:65\ntarget_words: \"and not with structural or axiom-choice residues, which would mean two parties running on genuinely different goods lists.\"\nolder: docs/research/frontier-extraction-experiment.md:62\nolder_words: \"Neither batch produced a fittingness residue, which is itself a finding\"\nkind: narrows\nwhy: The target predicts the residue map will fill mainly with fittingness and contextual residues. Structural ones would signal different goods lists. The 9 July hand runs (H9) found the opposite mix. Both batches bottomed out in structural residues (which evidential standard governs; burden of proof under asymmetric error costs) plus a risk-posture one, and no fittingness residue. Those structural residues are disputes about standards of proof, not goods lists, so the prediction's gloss also needs repair. The sample is two debates.\nstrength: medium\n\n### F13\ntarget: docs/research/progressive-disclosure.md:388\ntarget_words: \"subtle indicators that deeper layers exist (e.g., \\\"12 premises analyzed\\\", \\\"3 counter-arguments\\\")\"\nolder: docs/ux-principles.md:141\nolder_words: \"and the reader is given no signal that a budget was applied\"\nkind: supplies-a-missing-piece\nwhy: The target proposes making depth visible and keeping every level self-sufficient, as design. P6 records that the three levels already ship in the synthesis path and dropped material silently. 6/10/16 claims were retrieved, then cut to 3/5/8 supporting and 2/3/5 challenging, with an answer cap of 520 characters at every depth. The depth ladder was \"a promise the schema could not keep\". Fixed 13 Aug with a sentence naming how many ideas were in scope and unused, ending in a move. This is the target's principle tested in production.\nstrength: medium\n\n### F14\ntarget: docs/research/cross-source-premise-run.md:74\ntarget_words: \"it is a recurring class the taxonomy does not have, and it has a mechanical shape: one side generalises from a studied instance\"\nolder: docs/research/starting-values-and-the-problem-of-the-prior.md:461\nolder_words: \"It turns the reference class problem from a hidden choice into a visible, arguable link, which is the\"\nkind: supplies-a-missing-piece\nwhy: The target finds the reference-class premise in two unrelated debates, with neither side arguing the transfer, and lacks a home for it. The starting-values note (30 Sept, nothing ruled) designs one. A reference class becomes an ordinary attackable claim, joined to the claims it applies to by a contestable membership link (Hájek's reference class problem). That is a ready representation for the class the taxonomy lacks.\nstrength: medium\n\n### F15\ntarget: docs/research/kialo-deep-dive.md:150\ntarget_words: \"Weak pro-arguments can outweigh serious cons simply by accumulating points\"\nolder: docs/research/strength-layer-audit.md:130\nolder_words: \"For conjunction it is wrong in the dangerous direction: a mother claim can score well while one *necessary* part is weak\"\nkind: same-mechanism-elsewhere\nwhy: The Kialo critique (weak reasons winning by accumulation, nested cons undervalued) is listed only as Kialo's flaw. Deliberus's own sibling energy was additive with no correlation term, so strong siblings outvote a weak required part and twelve children outscore one. The 31 Aug padding ruling closed it for untagged parts. The Kialo critique is outside prior art for that ruling. Whether the nested-con undervaluation has an analogue here has not been asked.\nstrength: medium\n\n### F16\ntarget: docs/research/80000-hours-ai-decision-making.md:65\ntarget_words: \"the exact target audience (EA/rationalist) is being told to look for projects in this space, and they're not finding them\"\nolder: docs/research/cross-thread-synthesis.md:37\nolder_words: \"The rationalist/EA/AI safety community is the proposed beachhead. This community has strong priors on epistemology, political economy\"\nkind: narrows\nwhy: The target reads the profile as market validation of an EA/rationalist beachhead. The audit of the same day (28 March) names that beachhead as an early-adopter bias: the community's norms become the baseline for judging later contributions. A later note, curiosity-as-growth-fuel.md:109, argues that community is the wrong filter, since curiosity cuts across communities. The profile shows demand and funder attention. It does not show this audience should seed the graph.\nstrength: medium\n\n### F17\ntarget: docs/research/moral-convergence-metaethics-and-psychology.md:111\ntarget_words: \"with two-sided depolarization on all 26 initially polarized proposals among those starting at the extremes\"\nolder: docs/research/empirical-deliberation-convergence-evidence.md:32\nolder_words: \"participants moved toward the center on **22 of 26**, with the shift statistically significant on **19**\"\nkind: narrows\nwhy: The same-day in-house evidence record reports America in One Room as centre-ward movement on 22 of 26 proposals, 19 significant. It says the effect was strongest on complex substantive policy and weaker on identity-saturated symbolic issues. It concludes (line 93) that value-weighting disagreement moves very little and that the modal success is agreement on action without agreement on reasons. The two counts may describe different subsets, so check before quoting. The issue-type gradient bears directly on § 7's selection-effects objection.\nstrength: medium\n\n### F18\ntarget: docs/research/governance-and-financing.md:73\ntarget_words: \"Errikos Pitsos (founder/CEO) personally funded early operations. Now described as \\\"entirely funded by a private philanthropic foundation.\\\"\"\nolder: docs/research/kialo-deep-dive.md:276\nolder_words: \"For-profit company founded 2011, based in Brooklyn and Berlin\"\nkind: contradicts\nwhy: The target files Kialo as a nonprofit patron-funded by one foundation (about 50 staff) and tells Deliberus to avoid \"Kialo-style patron dependency\". The deep dive splits it. Kialo Inc. is a for-profit parent that is \"unfunded\" per Tracxn with a sole founder and no revenue. Kialo Edu is the nonprofit arm funded by the foundation. The patron applies to the education arm only. The parent's funding is unknown, and the structure is the dual-entity shape the target recommends elsewhere.\nstrength: medium\n\nNoticed: Candidates read at both ends and left out for the 18 cap: progressive-disclosure.md:141 (inspectability premium) against strength-layer-audit.md:95 (the instrument displayed unsupported confidence on 115 of 116 claims); progressive-disclosure.md:382 (three levels) against engagement-gradient-priors.md:30 (six tiers, \"reachable ONE rung from the tier below\", 90%+ readers); interaction-modalities:63 (mid-conversation structure as steering) against live-election-test-design.md:32 (DeepMind implicates restating, not showing structure); kialo-deep-dive.md:326 (education wedge works) against adoption-problem.md:243 (education viable but not primary, possibly a ceiling); 80000-hours:45 against ai-safety-and-tao-augmentation-research.md:70 (that note records the profile as five categories, September 2025, a different title, no \"structured transparency\" row); open-problems.md:51 (checker-circularity \"nobody has broken\") against ledger Y4 (session26-open-rulings.md:593, a different-family checker caught what the writer's family passed, but errors correlate ~60%); open-problems.md:59 (AI reader owes) against agents-as-a-consumer-class.md:152 (candidate answer: a paid run); governance:377 (self-funded horizon) against whose-tokens-pay-for-mapping.md (input form off, bill unpaid; money \"never the binding constraint\"). Governance line 387 still lists LTFF as a live parallel track, although its own line 359 note says LTFF closed. Pair (cross-source-premise-run, the-load-bearing-unsaid) was skipped as already in found.jsonl.\n"}