{"path":"research/the-canon-distillation-pass/topic-index-trim-proposals.md","content":"# Topic Index trim proposals (2026-11-10)\n\nThe verbosity audit's first pass over `CLAUDE.md` § Topic Index (`the-canon-distillation-pass.md` § 5). Nothing here is written into `CLAUDE.md` until the founder blesses it; the blessed entries are applied in one write at the end of the sitting by `topic_index_trim.py apply`.\n\n## Method\n\n- Each entry keeps its bold question phrase and its file and section pointer, and at most one sentence saying what the note holds. Every correction (⚠) or prohibition that only the index carries stays as one clause (ruled 2026-11-10, his words: *\"Yes, keep them all\"*), since the corpus graph leaves `CLAUDE.md` out as a source of corrections (`corpus-graph/README.md` § 2) and a session that never opens the note reaches such a sentence only here.\n- The entry's text moves **whole and word for word** into the note it points at, under a dated heading *From the Topic Index* at the end of the note (for `TODO.md`, at the end of the named section). Sentences the note already states travel too, marked as coming from the index, so nothing is lost and the repetition is visible and prunable. Measured with a crude probe, the forty longest entries' notes state 0–50% of their sentences (median about 30%), so most of what moves is not a duplicate.\n- The filter runs over the cut (§ 4b): every dropped sentence carrying ⚠ or a prohibition is listed per entry below with whether its note already states it. Entries whose clauses are written say so; the rest still need theirs written before they are put to him.\n- Six entries are proposed unchanged: two are the units he restored on his own challenge (*flag versus reasons*, *on sacred values*), three are already pointer-shaped, one is the standing threat-model pointer.\n\n## Totals\n\n| | bytes |\n|---|---|\n| entries today | 283,256 |\n| entries as proposed | 71,217 |\n| saved | 212,039 |\n\n`wc -c CLAUDE.md` today: 549,340; after all proposals: about 337,301.\n\n## Entries, longest first\n\n### 96. \"How would we test this inside a real community?\" / blobs / advice process / do-ocracy / a board / classified or secret material\n\n- **bytes**: 5,965 today → 785 proposed (saves 5,180)\n- **moves to**: `docs/research/blobs-community-sandboxes-and-the-advice-process.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"How would we test this inside a real community?\" / blobs / advice process / do-ocracy / a board / classified or secret material** → `docs/research/blobs-community-sandboxes-and-the-advice-process.md` (§ 3b the built membrane; the named-community material in `.private/`): a blob is a separate graph and database selected by `DELIBERUS_BLOB`, never community material behind a flag inside the public graph; a signed-in member reads it per request while writes stay per process, a blob you are not in answers 404 never 403, and the unlisted community page's path is never named under `docs/` or `README.md` (test-enforced); do not cite secrecy as a limit on Deliberus between states; ⚠ `request.session` asserts when SessionMiddleware is absent and `hasattr` does not catch it.\n\n- **warnings kept as clauses**: written\n- **BLESSED** 2026-11-10, his words: *\"Bless it\"*\n- **the filter over the cut** (6 ⚠/prohibition units in the entry):\n  - [note does NOT state it] **Never hide community material behind a flag inside the public graph** — a flag holds only while every reader and the linking sweep remember the filter.\n  - [note does NOT state it] **Never name its path in anything under `docs/` or in `README.md`** — both are publicly served and listed in `llms.txt` (test-enforced, `tests/test_docs_routing.py::TestUnlistedCommunityPage`).\n  - [note does NOT state it] **Since 2026-09-20 the community has a working group on its own decision-making and the founder is on its sheet**; his next steps and Claude's unruled thinking on them (a room, a name, and *look-backs*: finished advice processes mapped after the fact, which by P20 illustrate and never test) are priv…\n  - [note does NOT state it] **A blob you are not in answers 404, never 403**: a distinguishable refusal would confirm that a named community has a compartment here.\n  - [note does NOT state it] ⚠ Gotcha earned the same hour: `request.session` **asserts** when SessionMiddleware is absent and `hasattr` does not catch an AssertionError, so guarding it that way 500s every request — read `request.scope.get(\"session\")` instead.\n  - [note does NOT state it] Do not cite secrecy as a limit on Deliberus between states.\n\n### 44. Agents as readers/clients / MCP / \"agents check their conclusions against Deliberus\" / the agent-side falsifier\n\n- **bytes**: 5,942 today → 574 proposed (saves 5,368)\n- **moves to**: `docs/research/agents-as-a-consumer-class.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Agents as readers/clients / MCP / \"agents check their conclusions against Deliberus\" / the agent-side falsifier** → `docs/research/agents-as-a-consumer-class.md`: his own March idea, reasoning-verification not fact-retrieval, with the fact layer shipping first because that leg reaches the ground; read-only MCP with the instruments as first-class tools and tail retrieval, gated on the strength layer. ⚠ Do NOT read the retrieval guard as dismissing fact-checking (§ 2a, founder-overturned 2026-08-18): the dispute structure around a fact is what nothing else has.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 126. \"What do other notes say about this one?\" / what bears on a ruling / where to keep a connection somebody found\n\n- **bytes**: 5,804 today → 777 proposed (saves 5,027)\n- **moves to**: `docs/research/corpus-graph/README.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What do other notes say about this one?\" / what bears on a ruling / where to keep a connection somebody found** → `docs/research/corpus-graph/README.md` (the spec; § 12–§ 16 readers, repairs and the page sentences) + `uv run python scripts/corpus_graph.py around|search|add|written|read`: it is computed from the files, keeps and delivers what readers found and discovers nothing, and a session asks it first, always, before sending any helper to read the corpus (his word 2026-10-10). Readers go out again since 2026-10-27 (`plan_readings.py --owed` prices what is owed); a repair of a research note by his own ruling needs no new word from him, and a public-page or resident sentence goes to him one at a time, its whole record gathered first (`repair.py record`).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 58. Strength / QBAF / QEM / \"why does this claim show NO DATA?\" / the badge\n\n- **bytes**: 5,755 today → 594 proposed (saves 5,161)\n- **moves to**: `docs/research/strength-layer-audit.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Strength / QBAF / QEM / \"why does this claim show NO DATA?\" / the badge** → `docs/research/strength-layer-audit.md` (measured 2026-08-17) + `relevance-and-the-two-factors.md`: the badge once counted 1,567 incoming attacks as supports (badges 3,828 → 427 after the fix); a link now carries its supporter's own strength times its relevance (Q33, live 2026-09-30) and a refuted reason counts against by its fall from its own prior (Q35, live 2026-10-27). ⚠ The bridging feed still ranks by the classifier's stored `r.strength`, never the computed badge (`what-a-link-strength-means.md`).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 105. \"Make this something I can listen to\" / audio walkthroughs / ElevenReader\n\n- **bytes**: 5,692 today → 511 proposed (saves 5,181)\n- **moves to**: `docs/listening/how-to-write-for-listening.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Make this something I can listen to\" / audio walkthroughs / ElevenReader** → `docs/listening/how-to-write-for-listening.md` (the method, rules 34–63 ruled as sets) + the `walkthrough` skill, which runs it: the default is a rendered mp3 (`scripts/render_listening.py`, copied to his iCloud folder and sent), the renderer's dry run gates the text (`scripts/listening_gates.py`), and his rulings on numbers, examples, voices and the close are in the method and `what-eighteen-pieces-taught-2026-10-18.md`.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 101. \"Why does the machine keep missing things?\" / microtask coverage / why a judgment gets produced and then dropped / how humans stop doing microtasks\n\n- **bytes**: 5,570 today → 621 proposed (saves 4,949)\n- **moves to**: `docs/research/machine-work-as-a-standing-property.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Why does the machine keep missing things?\" / microtask coverage / why a judgment gets produced and then dropped / how humans stop doing microtasks** → `docs/research/machine-work-as-a-standing-property.md` (§ 7 the plan for every pass): a pass fires once at one door and nothing asks again, so coverage is unrecorded; RULED 2026-10-19 with a brake (R8, at most four model calls per rebuilt pass) and built for the keeper (`deliberus/standing_work.py`), with every model-asking function registered (`standing_work.MICROTASKS`, test-enforced); a standing pass costs no more calls than the endpoint that starts it today.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 138. Duplicate challenge homes / negated attacks / contextual senses / architecture and ontology diagrams / \"what's so wrong about seeking agreement?\"\n\n- **bytes**: 4,805 today → 776 proposed (saves 4,029)\n- **moves to**: `docs/research/one-challenge-one-home.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Duplicate challenge homes / negated attacks / contextual senses / architecture and ontology diagrams / \"what's so wrong about seeking agreement?\"** → `docs/research/one-challenge-one-home.md` (Arc AL; his positions of 15–17 October tabled in the note) + `codex-review-19-october-read-back.md`: RULED 2026-10-17 that Deliberus welcomes the agreement clarity earns and never optimizes for agreement (AL7); a rephrase is a new claim (AL10, built), the entry order cut → sameness → links is built behind `DELIBERUS_FLOOR_FIRST=1` (off), a cut whole is a record (AL15, ruled) and a fresh cut matches before it mints (AL16, built). ⚠ Codex's review: the reasons splitter counts an argument of several facts twice; the fix is specified and unbuilt, waiting on his word.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 56. Delegation / \"offload the virtue\" / the four open founder questions / islands of coherence\n\n- **bytes**: 4,588 today → 551 proposed (saves 4,037)\n- **moves to**: `docs/research/conviction-and-critique.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Delegation / \"offload the virtue\" / the four open founder questions / islands of coherence** → `docs/research/conviction-and-critique.md` § The delegation clarification + `docs/research/islands-of-coherence.md` (§ 5c the format): offloading is the purpose and the hazard is delegation without self-report; the workshop runs live, from scratch, never on pre-crunched material (P20). ⚠ 2026-10-10: the four founder questions are no longer all open (the first answered in part 2026-08-31 and by T1, the second held by his ruling of 2026-08-30).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 43. \"Is Deliberus a rare complexity jump / why now / why me?\" — three founder questions, answered unevenly\n\n- **bytes**: 4,473 today → 479 proposed (saves 3,994)\n- **moves to**: `docs/research/complexity-transitions-and-why-now.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is Deliberus a rare complexity jump / why now / why me?\" — three founder questions, answered unevenly** → `docs/research/complexity-transitions-and-why-now.md`: the rare-jump question splits (assembly index yes, assembly proper not yet), why-now is the strongest of the three (Kauffman's adjacent possible), and why-me is a post-selection question. ⚠ Do NOT restate the rare-jump answer as a flat no, and never write \"assembly theory shows that\" in a funding document.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 49. Machine persuasion / machines judging machines / multi-agent coordination / \"is our own judge calibrated?\"\n\n- **bytes**: 4,455 today → 487 proposed (saves 3,968)\n- **moves to**: `docs/research/the-scrutiny-gap.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Machine persuasion / machines judging machines / multi-agent coordination / \"is our own judge calibrated?\"** → `docs/research/the-scrutiny-gap.md`: AI out-persuades expert humans with throughput as the mechanism, hence *checking that keeps pace with persuasion*; Heilig's judge test signature (a concept-judge has an optimum, a marker-judge is monotonic) is runnable against `terminus_llm` confidence; the Anthropic relation is a component claim, kept research-side until ratified.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 17. \"What should be settled BEFORE building the ontology?\" / is the schema too fuzzy or too opinionated / operator-reflexivity\n\n- **bytes**: 4,129 today → 475 proposed (saves 3,654)\n- **moves to**: `docs/research/what-belongs-in-the-ontology.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What should be settled BEFORE building the ontology?\" / is the schema too fuzzy or too opinionated / operator-reflexivity** → `docs/research/what-belongs-in-the-ontology.md` (§ 6b on importing someone else's taxonomy): a commitment belongs in the graph if the system can hold two users disagreeing about it and keep working, in the ontology if it cannot; \"settle the metaethics first\" is a closed path, and a borrowed category earns its place only by yielding a move.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 16. Fractal scales / \"at what level does this operate?\" / temporal frame\n\n- **bytes**: 3,970 today → 481 proposed (saves 3,489)\n- **moves to**: `docs/research/fractal-scales-and-temporal-frame.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Fractal scales / \"at what level does this operate?\" / temporal frame** → `docs/research/fractal-scales-and-temporal-frame.md` (the scale ladder; § Three classes of adversary, whose ### 0 is the plain version): each rung up adds an adversary, a mechanism transfers free inside a class (entropy / motivated stakeholder / strategy) and needs a rebuild across one, and no instrument here has yet met a motivated stakeholder, since every source is machine- or operator-extracted.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 73. \"Can extraction run without paying the Gemini bill?\" / Claude Code as extraction engine / the quota decision's fourth option\n\n- **bytes**: 3,834 today → 510 proposed (saves 3,324)\n- **moves to**: `docs/research/claude-code-as-extraction-engine.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Can extraction run without paying the Gemini bill?\" / Claude Code as extraction engine / the quota decision's fourth option** → `docs/research/claude-code-as-extraction-engine.md` (§ 7.8 the local checklist): operator-run extraction only, never reopened public extraction; built and live-verified 2026-08-23 behind `DELIBERUS_LLM_BACKEND=claude`, and since 2026-09-09 the engine for the election-week tests (≈ 20 min per ten-minute spoken segment, the stream path skipping the implicit-premise pass).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 106. Three minds (human, model, the mind of Deliberus) / attunement as a mind-meld / \"how do we stop a microtask leaping?\" / the keeper's topic-not-evidence overlap\n\n- **bytes**: 3,714 today → 658 proposed (saves 3,056)\n- **moves to**: `docs/research/three-minds-and-the-leap.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Three minds (human, model, the mind of Deliberus) / attunement as a mind-meld / \"how do we stop a microtask leaping?\" / the keeper's topic-not-evidence overlap** → `docs/research/three-minds-and-the-leap.md` (his reflection whole in § 1; ledger Arc AM): a pass produces the object its verdict depends on or the verdict is refused (AM1, built in the keeper as gate and chain), a conditional is a link cut out before the splitter (AM5), a hedge is the author's stance never strength (AM6), the pass is handed what the corpus holds (AM8), and the keeper reads a pair in its game (AM9, unbuilt). Do not answer the keeper question with an awareness clause.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 18. \"Can the weighed things themselves be decomposed?\" / distilling human values / Pearce / valence / what values are physically made of\n\n- **bytes**: 3,625 today → 475 proposed (saves 3,150)\n- **moves to**: `docs/research/decomposing-value.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Can the weighed things themselves be decomposed?\" / distilling human values / Pearce / valence / what values are physically made of** → `docs/research/decomposing-value.md` (the five-level palette, the two floors). ⚠ Three founder corrections there are load-bearing and the pre-correction framing must not return: evolution-versus-phenomenal-combinatorics is a false contrast, the primitive count is the wrong axis, and \"held\" means state persistence, not belonging.\n\n\n### 53. Claim/concept sameness / auto-merge / splitting / \"why is evidence scattered?\" / identity\n\n- **bytes**: 3,619 today → 504 proposed (saves 3,115)\n- **moves to**: `docs/research/soft-canonical-clustering-and-reversible-merge-semantics.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Claim/concept sameness / auto-merge / splitting / \"why is evidence scattered?\" / identity** → `docs/research/soft-canonical-clustering-and-reversible-merge-semantics.md` § Addendum (the rulings) + `sameness-merge-and-split-across-fields.md` + `claim-sameness-philosophical-readings.md`: anyone types anything, careful auto-merge afterward, anyone can split; seven keys blessed 2026-08-31; cluster-carries-the-strength is his from ~2008, built 2026-10-20 and switched off (`DELIBERUS_SAME_GROUPS`).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 94. \"What do we truly need human judgment FOR?\" / why ratification is the wrong ask / anti-brigading\n\n- **bytes**: 3,566 today → 530 proposed (saves 3,036)\n- **moves to**: `docs/research/what-human-judgment-is-for.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What do we truly need human judgment FOR?\" / why ratification is the wrong ask / anti-brigading** → `docs/research/what-human-judgment-is-for.md` (analysis + proposal): the human-reserved classes collapse to coherent absence and frame lock-in; ratification is the worst use of a human; *contributions are added, never overwritten* (ratified narrowed 2026-08-27, about existence not weight); consent (ruling on a rendering of your own position: ship it) is not ratification (confirming a machine's proposal about the world).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 97. Claim pages and source pages as small maps you travel through (the map view)\n\n- **bytes**: 3,426 today → 565 proposed (saves 2,861)\n- **moves to**: `specs/map-view/design.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Claim pages and source pages as small maps you travel through (the map view)** → `specs/map-view/design.md` (§ 3a the wheel view, live since 2026-10-20; `?old` keeps the first map) + `how-a-map-is-cut-and-what-it-shows.md` + `folding-joining-and-zoom.md` § 8: about ten boxes, every box a link, one marked place for the region's rarest claim; RULED that the plain wheel turns the view and only ⌘-scroll zooms (ledger AR1–AR9). ⚠ `game-feel-and-the-first-screen.md` was not read before the build; § 8 of the spec lists what it asked for that is unmet.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 54. \"Why do we need concept tracking at all?\" / core vs periphery of a word / are non-nouns covered / what the concept layer measures\n\n- **bytes**: 3,365 today → 508 proposed (saves 2,857)\n- **moves to**: `docs/research/core-periphery-and-the-concept-layer.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Why do we need concept tracking at all?\" / core vs periphery of a word / are non-nouns covered / what the concept layer measures** → `docs/research/core-periphery-and-the-concept-layer.md` + `the-motte-and-the-periphery.md`: a word's core is shared and its periphery assembled per community, and for contested words the periphery is the meaning; the layer records deltas, not descriptions (§ 5b, ledger N4), and 49 of 90 live concepts read \"bifurcated\" because the detector mints two senses per term.\n\n\n### 20. \"Is this conviction actually supported?\" / cross-scale patterns / analogies to evolution and complexity\n\n- **bytes**: 3,317 today → 420 proposed (saves 2,897)\n- **moves to**: `docs/research/convictions-across-scales.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is this conviction actually supported?\" / cross-scale patterns / analogies to evolution and complexity** → `docs/research/convictions-across-scales.md`: the four-check transfer test over 11 convictions and 23 UX principles, with Lowered firing five times (the wager splits: a small basis, never the same basis). Do NOT re-derive those as open, and self-hosting fractal priors is held by founder ruling 2026-08-30.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 40. \"Why do embedding models converge on one geometry, and do human minds?\" / the embeddings tension's anatomy / vec2vec\n\n- **bytes**: 3,303 today → 644 proposed (saves 2,659)\n- **moves to**: `docs/research/why-representations-converge-and-whether-minds-do.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Why do embedding models converge on one geometry, and do human minds?\" / the embeddings tension's anatomy / vec2vec** → `docs/research/why-representations-converge-and-whether-minds-do.md` + `universal-embedding-geometry-and-the-wager.md` + `contingency-and-convergence-across-levels.md`: models converge by compressing one written record; human cores converge and peripheries do not, so a model's vector for a contested word is the mean of a multimodal distribution, nobody's meaning. ⚠ Do not restate \"share the same penguin one time in eight\" (founder-corrected 2026-09-06), nor \"a second model family is not a second opinion\" flat.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 65. \"Why does the graph have almost no implicit premises?\" / the falling-cost prediction's mechanism / does decomposition reuse existing claims\n\n- **bytes**: 3,192 today → 588 proposed (saves 2,604)\n- **moves to**: `docs/research/self-similar-decomposition-and-claim-ontology.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Why does the graph have almost no implicit premises?\" / the falling-cost prediction's mechanism / does decomposition reuse existing claims** → `docs/research/self-similar-decomposition-and-claim-ontology.md` § The mechanism audited + `the-reuse-flywheel-prior.md` § 3: the prediction has never been given its conditions (10 pipeline-produced implicit premises across the 107 arguments the pass saw, a scarcity-instructed prompt, the stream path never calling the pass). ⚠ `auto_connect` DOES look up existing claims (84 edges, stored backwards and since read again as support).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 36. \"Does the typed graph still earn its keep as models scale?\" / structure vs unstructured LLM reasoning / the bitter lesson\n\n- **bytes**: 2,914 today → 616 proposed (saves 2,298)\n- **moves to**: `docs/research/structure-versus-scale.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Does the typed graph still earn its keep as models scale?\" / structure vs unstructured LLM reasoning / the bitter lesson** → `docs/research/structure-versus-scale.md` + `nativism-thread-and-false-dissolution.md`: the primary competitor is a well-kept wiki, so the question is whether *typed* structure earns its cost over cheap structure; resolved 2026-08-14 by stacking (addressability the substrate), with one signal that would change the build, friends-round users not pointing at claims. ⚠ Do NOT carry the first reformulation (\"reasoning must be persistent and contestable\"): its falsifier cannot fire.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 77. \"How fast is extraction / which backend / why not a bigger model?\" / the live-session break budget\n\n- **bytes**: 2,838 today → 527 proposed (saves 2,311)\n- **moves to**: `docs/research/live-election-test-design.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"How fast is extraction / which backend / why not a bigger model?\" / the live-session break budget** → `docs/research/live-election-test-design.md` § The break budget + § State 2026-09-09 + `claude-code-as-extraction-engine.md` § 7.8b: do NOT re-derive speed from the free-tier A/B; the model question is closed (Sonnet), he ruled density and attribution over latency, the mandated break is the design consequence, and the audio path is bridge-first (the Mac chain at 87–95% attribution, Gemini a labelled fallback).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 86. \"What should humans re-check?\" / the triage feed / ratification states / the anti-pattern-9 rebirth\n\n- **bytes**: 2,819 today → 420 proposed (saves 2,399)\n- **moves to**: `docs/research/the-ratification-record-and-the-triage-feed.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What should humans re-check?\" / the triage feed / ratification states / the anti-pattern-9 rebirth** → `docs/research/the-ratification-record-and-the-triage-feed.md`: never record agreement as examination; BLESSED 2026-08-31, the graph holds contributions and telemetry and nothing else, there is no ratify act, and the seeded probe is dropped (endorsement and identity claims are factual claims like any others).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 32. Inspectable synthesis / \"isn't this my truth-graph idea?\"\n\n- **bytes**: 2,812 today → 310 proposed (saves 2,502)\n- **moves to**: `docs/research/synthesis-build-plan.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Inspectable synthesis / \"isn't this my truth-graph idea?\"** → `docs/research/synthesis-build-plan.md` § 7: yes, his April idea; Phases 1–3 shipped 2026-08-13 (ids, `cited_claim_ids`, both omission classes), Phase 4 and the three-arm experiment open, the head/tail citation balance proposed 2026-08-30.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 59. \"Why the weakest-part default?\" / padding attack / min vs product vs sum / the merge hunch / double-counting / Bayes on the strength layer\n\n- **bytes**: 2,794 today → 473 proposed (saves 2,321)\n- **moves to**: `docs/research/weakest-link-arithmetic-and-the-merge-hunch.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Why the weakest-part default?\" / padding attack / min vs product vs sum / the merge hunch / double-counting / Bayes on the strength layer** → `docs/research/weakest-link-arithmetic-and-the-merge-hunch.md` + `what-the-necessary-default-changes.md`: min is the only clone-proof AND rule and the exact bound for entailment-chained padding; the argued-requirement cap shipped 2026-09-17, and one irrelevant person-added part still caps a claim at 0.75 (ledger O13 open).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 28. Peer review as an application surface\n\n- **bytes**: 2,774 today → 341 proposed (saves 2,433)\n- **moves to**: `docs/research/peer-review-and-the-reasoning-layer.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Peer review as an application surface** → `docs/research/peer-review-and-the-reasoning-layer.md`: the institution's history as a long unbundling, eight mechanisms ranked, run 5 by hand (22 claims, zero attack edges, both predictions confirmed); the specimen's author stays unnamed and the verbatim artifact belongs in `.private/` only.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 127. \"No too-large leaps\" / the chain inside a link / the smallest step, and its floor / is *how strongly a reason bears* anything but relevance / when the machine writes its guesses, widens and narrows, and on what / where mechanism evidence goes\n\n- **bytes**: 2,734 today → 666 proposed (saves 2,068)\n- **moves to**: `docs/research/the-chain-inside-a-link.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"No too-large leaps\" / the chain inside a link / the smallest step, and its floor / is *how strongly a reason bears* anything but relevance / when the machine writes its guesses, widens and narrows, and on what / where mechanism evidence goes** → `docs/research/the-chain-inside-a-link.md` (§ 10–11; AG6 and AG14 RULED): the machine may and should write deeper steps, quality-controlled like every microtask, each the smallest step possible (no gap, no removable sentence); do not restate *the first level eagerly, the rest on request*, do not ask how strongly a link bears while nothing is written under it, and do not read a gap-free chain as a good link.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 50. LLM judge fragility / \"can our classifiers be argued with?\" / wiggle / verdict-as-artifact / physics world models and the verification bottleneck\n\n- **bytes**: 2,715 today → 442 proposed (saves 2,273)\n- **moves to**: `docs/research/fragile-checkers-and-the-verification-bottleneck.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **LLM judge fragility / \"can our classifiers be argued with?\" / wiggle / verdict-as-artifact / physics world models and the verification bottleneck** → `docs/research/fragile-checkers-and-the-verification-bottleneck.md`: every frontier judge flips under pressure and the flips corrupt, so verdicts are records and challenges become objects (verdict-as-artifact); no adversarial-accessible surface re-invokes a judge on a standing verdict.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 48. \"Is the Walton scheme set exhaustive?\" / scheme taxonomy / generative basis / no-scheme-fits\n\n- **bytes**: 2,668 today → 422 proposed (saves 2,246)\n- **moves to**: `docs/research/scheme-set-exhaustiveness.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is the Walton scheme set exhaustive?\" / scheme taxonomy / generative basis / no-scheme-fits** → `docs/research/scheme-set-exhaustiveness.md`: the set is open by construction (Walton's own field test found 37% no-fit), `does_not_fit` is a confession channel that fired only once the costless shrug was removed (single exit RULED 2026-09-13), and since 2026-10-10 the no-fit answer is asked first and is off the list.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 95. \"When is something called a terminus / horizon, and is that premature?\" / the finality words / read-versus-set\n\n- **bytes**: 2,664 today → 485 proposed (saves 2,179)\n- **moves to**: `docs/research/what-belongs-in-the-ontology.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"When is something called a terminus / horizon, and is that premature?\" / the finality words / read-versus-set** → `docs/research/what-belongs-in-the-ontology.md` § 6c–6c.3 + `session33-the-horizon-vocabulary.md`: RULED 2026-09-15 that a horizon kind is read, never set (`deliberus/horizon.py`, built 2026-09-16, the setter answering 410); the reading errs conservative, coherent absence is what no reading sees, and how loud a provisional reading may be is open (ledger O12).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 83. The harness conviction / \"what makes a machine reader good\" / instrument-rated reasoner / chart-vs-ledger\n\n- **bytes**: 2,661 today → 506 proposed (saves 2,155)\n- **moves to**: `docs/research/the-harness-conviction.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **The harness conviction / \"what makes a machine reader good\" / instrument-rated reasoner / chart-vs-ledger** → `docs/research/the-harness-conviction.md` (founder verbatim, standing): machine reasoning's quality is a function of harness and grounding, architecture-independent; remove the constraint, don't repair the reasoner. ⚠ § 8.5b: claims DO have coordinates (three layers); what the graph lacks is a metric where distance is the decision, and the flywheel condition is unstaked-versus-staked.\n\n\n### 99. \"Can Deliberus pin down what *love* / *understanding* / *being seen* mean?\" / the attunement pole's own vocabulary / the concept layer's hardest class\n\n- **bytes**: 2,647 today → 576 proposed (saves 2,071)\n- **moves to**: `docs/research/attunement-words-and-the-concept-layer.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Can Deliberus pin down what *love* / *understanding* / *being seen* mean?\" / the attunement pole's own vocabulary / the concept layer's hardest class** → `docs/research/attunement-words-and-the-concept-layer.md` (proposal, unruled): thin-exemplar terms whose difference is most of the content, and the surviving case is internal (the convictions are written as deltas over these words). ⚠ Three distinctiveness claims were re-graded 2026-09-22 on founder challenge, *both parties are sincere* withdrawn; do NOT restate the August pole-indictment in the present tense.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 158. The Swedish AI safety institute map / the Middle Powers application\n\n- **bytes**: 2,463 today → 495 proposed (saves 1,968)\n- **moves to**: `TODO.md` (end of the named section)\n- **proposed line**:\n\n  - **The Swedish AI safety institute map / the Middle Powers application** → `TODO.md` § *The Swedish AI security institute map* (state, method, what is owed) + `.private/funding/pipeline-2026/` (the application as sent, 2026-10-24): the hand-made map is live since 2026-10-23 with its English twin since 2026-10-24; change it only in `institute-map/build_map.py` and store with `scripts/store_hand_extraction.py`, never by hand in the graph, and a claim with no published source is withdrawn.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 87. \"Could Deliberus drown in claims?\" / volume, scale, sameness scatter, ratification debt\n\n- **bytes**: 2,435 today → 449 proposed (saves 1,986)\n- **moves to**: `docs/research/drowning-in-claims.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Could Deliberus drown in claims?\" / volume, scale, sameness scatter, ratification debt** → `docs/research/drowning-in-claims.md` + `saturation-and-the-long-tail.md`: yes, through machine minting and sameness scatter, never earnest humans; distinct content saturates while text does not, the tail holds the value and summarization kills exactly it, and the anti-drowning programme is the already-prioritized work (six dials proposed, unruled).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 31. Incentives / adoption / \"why would anyone contribute?\"\n\n- **bytes**: 2,379 today → 357 proposed (saves 2,022)\n- **moves to**: `docs/research/incentives-analysis.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Incentives / adoption / \"why would anyone contribute?\"** → `docs/research/incentives-analysis.md` (§ 6b the supply-side-first variant): eight conditions where legibility pays, solo use the only incentive-clean one and so the right first stage; supports are self-serving and attacks altruistic, and selective legibility is the move no instrument sees.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 80. \"How should a claim be cut, and did the cut lose anything?\" / theory-guided decomposition / coverage-coherence-atomicity\n\n- **bytes**: 2,374 today → 482 proposed (saves 1,892)\n- **moves to**: `docs/research/the-jhu-decomposition-line.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"How should a claim be cut, and did the cut lose anything?\" / theory-guided decomposition / coverage-coherence-atomicity** → `docs/research/the-jhu-decomposition-line.md` + `deliberus/extraction/decompose_claim.py`: theory in the prompt beats a parser; writing-marked since 2026-09-21 (it writes what it found as `provenance='split'` and holds what it added), pass version 2 since 2026-09-30; coverage fails silently, and DnDScore's own resolution is unavailable to the graph.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 33. Curiosity / growth fuel / \"why would anyone descend?\"\n\n- **bytes**: 2,336 today → 344 proposed (saves 1,992)\n- **moves to**: `docs/research/curiosity-as-growth-fuel.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Curiosity / growth fuel / \"why would anyone descend?\"** → `docs/research/curiosity-as-growth-fuel.md`: reasoning proficiency worsens polarization and curiosity alone reverses it (Kahan), so curiosity is a precondition on the wager; PACE gives the rule *no gap displayed without a move offered*; the second conviction term is NOT ratified.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 88. \"Compress the CQ scaffolding to reusable tokens?\" / latent claims / the type–token split\n\n- **bytes**: 2,313 today → 432 proposed (saves 1,881)\n- **moves to**: `docs/research/latent-scaffolding-and-the-type-token-split.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Compress the CQ scaffolding to reusable tokens?\" / latent claims / the type–token split** → `docs/research/latent-scaffolding-and-the-type-token-split.md` (§ 4b what shipped): ENDORSED and BUILT 2026-09-21, a critical question is a potential rendered from scheme and link and minted when touched; about half of slot occurrences cannot be filled by code, and eight consumer modules still count nodes rather than potentials.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 143. \"Were the parts rulings made on reasons, and what carries over?\" / an argument (his 2013 argument bundle; *line* retired 2026-10-22) / what the number must be for a stranger / priors and coherence together / the keeper as the independence detector / how many independent reasons a claim can have, and whether one is missing / the philosophy-first order of interventions / more of the corpus in the window\n\n- **bytes**: 2,304 today → 895 proposed (saves 1,409)\n- **moves to**: `docs/research/retrofitting-the-reasons-channel.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Were the parts rulings made on reasons, and what carries over?\" / an argument (his 2013 argument bundle; *line* retired 2026-10-22) / what the number must be for a stranger / priors and coherence together / the keeper as the independence detector / how many independent reasons a claim can have, and whether one is missing / the philosophy-first order of interventions / more of the corpus in the window** → `docs/research/retrofitting-the-reasons-channel.md` + `the-number-in-the-world-and-the-repertoire.md` (Q38–Q45, Arc AK; Q38 and Q40 ruled 2026-10-22): the four parts rulings read on the reasons channel as the weakest premise within an argument and the best argument across arguments until independence is read; payoff removal is the last class of intervention, not the first. Do not say the sensing of need reached world completeness, nor that independence has no detector here.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 35. World models / the 2026 ontology revival / \"where does coherence come from?\"\n\n- **bytes**: 2,245 today → 343 proposed (saves 1,902)\n- **moves to**: `docs/research/world-models-and-the-ontology-revival.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **World models / the 2026 ontology revival / \"where does coherence come from?\"** → `docs/research/world-models-and-the-ontology-revival.md`: coherence does not emerge from accumulating plausible statements, it must be imposed, and the question is where (the Manhattan taxi study); do NOT transfer the revival's enterprise accuracy figures.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 30. The cost is a variable / cost curve / \"how do we lower it further?\"\n\n- **bytes**: 2,244 today → 342 proposed (saves 1,902)\n- **moves to**: `docs/research/lowering-the-cost.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **The cost is a variable / cost curve / \"how do we lower it further?\"** → `docs/research/lowering-the-cost.md`: the cost decomposes into eight terms, the social one largest; reuse runs through the concept layer, not the claim layer (⚠ 2026-08-29: within a source, `the-reuse-flywheel-prior.md`), and cost tracking was never implemented.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 38. \"How should a claim be decomposed?\" / atomization / unbundling / \"is finer always better?\"\n\n- **bytes**: 2,221 today → 494 proposed (saves 1,727)\n- **moves to**: `docs/research/decomposition-axes.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"How should a claim be decomposed?\" / atomization / unbundling / \"is finer always better?\"** → `docs/research/decomposition-axes.md` (eight axes with seam tests) + the mother-claim pattern in `self-similar-decomposition-and-claim-ontology.md`: critical questions decompose an inference, one axis of eight; decompose to separability, since fully atomic facts are the wrong representation for a verifier. ⚠ The contest-gated rule was misattributed to the founder and corrected 2026-08-27.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 62. Graph daemons / background agents / \"enzymes, T-cells, roaming philosophers\"\n\n- **bytes**: 2,194 today → 359 proposed (saves 1,835)\n- **moves to**: `docs/research/graph-daemons-design-space.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Graph daemons / background agents / \"enzymes, T-cells, roaming philosophers\"** → `docs/research/graph-daemons-design-space.md`: a daemon may change the rate of a move a human would make, never the equilibrium; three tiers by reversibility (committable / propose-only / never), the staleness daemon first, and the strength layer settled before any of it.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 146. \"What does one case count for?\" / induction / confirming instances / how a rule's number comes from its cases / a study as a bundle of cases / heterogeneity and moderators / which rule a single case is read through\n\n- **bytes**: 2,188 today → 666 proposed (saves 1,522)\n- **moves to**: `docs/research/what-a-case-counts-for.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What does one case count for?\" / induction / confirming instances / how a rule's number comes from its cases / a study as a bundle of cases / heterogeneity and moderators / which rule a single case is read through** → `docs/research/what-a-case-counts-for.md` + `reasons-and-cases.md` (Arc AY and BA1–BA2, all ruled 2026-10-23; built for reference classes 2026-10-24, not yet for a rule under a link): every case moves a statistical generalization, a study is a bundle of cases, cases enter a rule's reasons as one argument, and several cases from one unit are a count under that unit's own rate (Sally Clark). Do not say a case *confirms* a generalization.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 144. \"How do several reasons for one claim add?\" / log-odds / the best argument until independence is read / the independence reading and its test cases / a statistical generalization's stated proportion (AT8) / a claim's stated prior\n\n- **bytes**: 2,184 today → 684 proposed (saves 1,500)\n- **moves to**: `docs/research/how-independent-reasons-add.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"How do several reasons for one claim add?\" / log-odds / the best argument until independence is read / the independence reading and its test cases / a statistical generalization's stated proportion (AT8) / a claim's stated prior** → `docs/research/how-independent-reasons-add.md` (Q46, live 2026-10-23; Q48 bounds it) + `where-a-shared-doubt-sits.md` (Q42, AM19 ruled): reasons read independent add in log-odds and count as their best until read; reasons sharing one likely rival or a common cause go under a claim between them. Do NOT sweep the independence reading over the live graph before premises are grouped into arguments, and do not call the polls case's rules normic.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 51. World models in RL / \"can machine imagination corrupt the record?\" / compounding model bias / authority routing / standing of untouched machine structure\n\n- **bytes**: 2,063 today → 436 proposed (saves 1,627)\n- **moves to**: `docs/research/grounded-critics-and-quarantined-imagination.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **World models in RL / \"can machine imagination corrupt the record?\" / compounding model bias / authority routing / standing of untouched machine structure** → `docs/research/grounded-critics-and-quarantined-imagination.md`: the graph is the critic and the LLM passes the world model, so model output never writes to the learned record; today the strength layer computes over unratified machine edges, which is the honest inversion.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 93. Cognitive bias / \"aren't humans biased too?\" / can we score a contributor's rationality\n\n- **bytes**: 2,048 today → 519 proposed (saves 1,529)\n- **moves to**: `docs/research/cognitive-bias-codex-and-human-contribution.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Cognitive bias / \"aren't humans biased too?\" / can we score a contributor's rationality** → `docs/research/cognitive-bias-codex-and-human-contribution.md`: Deliberus removes the constraint rather than repairing the person; no expert gate, contributor-rationality scoring measures noise, self-report gets no elevated credence; never build a bias-tagging affordance. ⚠ Three objects, one refusal: the taxonomy as schema (no), the findings as design knowledge (yes, load-bearing), the arrangement as evidence (no).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 64. \"Should we EXPECT the reuse flywheel / the asymptote?\" / fractal priors for accumulated substrate / what the corpus already saw about automatic recursive decomposition\n\n- **bytes**: 2,016 today → 536 proposed (saves 1,480)\n- **moves to**: `docs/research/the-reuse-flywheel-prior.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Should we EXPECT the reuse flywheel / the asymptote?\" / fractal priors for accumulated substrate / what the corpus already saw about automatic recursive decomposition** → `docs/research/the-reuse-flywheel-prior.md`: expect the flywheel (seven scales, no failure), on condition that identity is cheap to decide, which puts claim-sameness at the core; the stranger test is anti-reuse at the text layer. ⚠ 2026-10-07: \"never generating premises, only finding real ones\" was Claude's prose; his sentence said *finding or creating*.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 103. A friend's structured position on AI governance, mapped by hand; three September sources on what is left for humans once machines prove, decide and substitute\n\n- **bytes**: 1,964 today → 591 proposed (saves 1,373)\n- **moves to**: `docs/research/mapping-a-structured-position-on-ai-governance.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **A friend's structured position on AI governance, mapped by hand; three September sources on what is left for humans once machines prove, decide and substitute** → `docs/research/mapping-a-structured-position-on-ai-governance.md` + `understanding-after-the-proof-machines.md`: ⚠ the author read the live map and found his typing flattened, since the map read the content without his twelve-value epistemic ladder (24 of 116 claims corrected, only ever lowering certainty); none of the six epistemic statuses says *this is open*, and extending the enum is a founder decision not taken.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 98. What a machine may write into the graph on its own / the trust ladder / \"can we just let it tidy up?\"\n\n- **bytes**: 1,955 today → 436 proposed (saves 1,519)\n- **moves to**: `docs/research/the-tidying-line.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **What a machine may write into the graph on its own / the trust ladder / \"can we just let it tidy up?\"** → `docs/research/the-tidying-line.md` (the sweep) + `earning-the-right-to-write.md` (the proposal): the discriminator is whether the output gives a reader more places to disagree or fewer, bounded by the separability floor; ⚠ the tier table's decomposition line predates the 2026-09-13 rulings and is not a live prohibition.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 69. The uncollected idea (a third failure mode neither pressure catches)\n\n- **bytes**: 1,937 today → 316 proposed (saves 1,621)\n- **moves to**: `docs/research/structuring-gradient-lineage.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **The uncollected idea (a third failure mode neither pressure catches)** → `docs/research/structuring-gradient-lineage.md` + `conviction-and-critique.md`: an idea written down correctly can still change nothing; four measured instances, and the countermeasure is one name, one home, one thing that depends on it.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 39. Non-ASCII comparison keys / \"why did two different senses merge?\" / Swedish in the corpus\n\n- **bytes**: 1,922 today → 416 proposed (saves 1,506)\n- **moves to**: `docs/research/session19-decomposition-corrections-and-the-ontology-turn.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Non-ASCII comparison keys / \"why did two different senses merge?\" / Swedish in the corpus** → `docs/research/session19-decomposition-corrections-and-the-ontology-turn.md` § 7: ASCII folding is correct in an identifier and catastrophic in a comparison key, so any regex over claim text, senses or titles uses `\\w`, never `a-z` (fixed 2026-08-16; the weighing, brake and stance lexicons are still English-only).\n\n\n### 57. Similarity search / \"why is retrieval slow\" / what finds related claims\n\n- **bytes**: 1,909 today → 352 proposed (saves 1,557)\n- **moves to**: `docs/research/retrieval-instruments-beyond-cosine.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Similarity search / \"why is retrieval slow\" / what finds related claims** → `docs/research/retrieval-instruments-beyond-cosine.md` (§ 0 plain): performance fixed 2026-08-17 (scored in FalkorDB, 0.31 s against 20 s; three auto-connect call sites left unwired on purpose), and cosine is narrow, not wrong (same topic, never same kind of argument).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 147. \"What is a claim's prior, and who says what kind it is?\" / seeding priors and implications as standing work / a consideration counts once\n\n- **bytes**: 1,857 today → 503 proposed (saves 1,354)\n- **moves to**: `docs/research/priors-as-standing-work.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What is a claim's prior, and who says what kind it is?\" / seeding priors and implications as standing work / a consideration counts once** → `docs/research/priors-as-standing-work.md` + `saying-what-kind-a-claim-is.md` (Arc BC; BC1, BC7, BC8 ruled; built 2026-10-24 with no model, `deliberus/reference_classes.py`): a claim's prior comes from what it says through its counted reference classes, pooled by code, who wrote it never entering; a consideration counts once. Say *prior*, never *start*.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 100. \"How is the understand-without-agreeing register actually received?\" / a layperson reaching our convictions unaided / the fake-middle misreading\n\n- **bytes**: 1,784 today → 493 proposed (saves 1,291)\n- **moves to**: `docs/research/the-post-that-states-the-problem-from-inside.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"How is the understand-without-agreeing register actually received?\" / a layperson reaching our convictions unaided / the fake-middle misreading** → `docs/research/the-post-that-states-the-problem-from-inside.md` (§ 6 the comment thread): a layperson reaches the no-verdict stance unaided, and two sympathetic readers of ten land on the fake middle she refused, so any *we map reasoning rather than deliver verdicts* surface will be heard by some allies as *the truth is in the middle*.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 145. \"Is a rule's *usually* a hedge?\" / normic and statistical generalizations / normative reasons as defaults / trying a policy's rules out over cases / what a person finds in need of explaining\n\n- **bytes**: 1,783 today → 586 proposed (saves 1,197)\n- **moves to**: `docs/research/normic-rules-normative-reasons-and-trying-rules-out.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is a rule's *usually* a hedge?\" / normic and statistical generalizations / normative reasons as defaults / trying a policy's rules out over cases / what a person finds in need of explaining** → `docs/research/normic-rules-normative-reasons-and-trying-rules-out.md` (Arc AT; AT1–AT3, AT9, AT11–AT13 ruled) + `what-relevance-is.md` § 3f: a generalization keeps one number and its *usually* is made exact in its words, the writer asks whether a failure would call for an explanation (normic) or not (statistical), and an exception is a claim of its own, never a list on a rule.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 81. \"What does ratification actually DO?\" / propose-then-ratify's missing half\n\n- **bytes**: 1,779 today → 400 proposed (saves 1,379)\n- **moves to**: `docs/research/what-confirmed-means.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What does ratification actually DO?\" / propose-then-ratify's missing half** → `docs/research/what-confirmed-means.md` + `the-residual-error-taxonomy.md` class 6: there is no ratify act; `confirmed=true` meant *wrote*, never *checked*, so the field was retired (Q32, RULED 2026-09-28) and its jobs rehomed 2026-10-19 (`NO_PERSON_ACTED`, `COUNTS_TOWARD_STRENGTH`, `trust_state`, `human_action`).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 19. Permissive zone / parity / Chang / \"what happens inside the zone?\" / life-decisions as a use case\n\n- **bytes**: 1,752 today → 350 proposed (saves 1,402)\n- **moves to**: `docs/research/parity-and-the-zone-interior.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Permissive zone / parity / Chang / \"what happens inside the zone?\" / life-decisions as a use case** → `docs/research/parity-and-the-zone-interior.md` (founder-ruled 2026-08-24/25): the zone's interior is mapped with attribution, no residue type is a dead end, no stop signs in the sacred domain, and the display word is **fittingness** for now.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 42. \"Will this actually succeed?\" / fractal priors / why-now / what does winning look like\n\n- **bytes**: 1,738 today → 256 proposed (saves 1,482)\n- **moves to**: `docs/research/fractal-priors-for-success.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Will this actually succeed?\" / fractal priors / why-now / what does winning look like** → `docs/research/fractal-priors-for-success.md` (17 lenses with counter-lenses), summarised always-loaded in § The Success Prior in One Breath: read that first.\n\n\n### 161. \"Does an alignment philosopher want what Deliberus builds?\" / Carlsmith and Faggella on The Trajectory / the worthy successor as a civilization with AI integrated / \"collective epistemology in a good way\" / how Dan and Joe would read Deliberus\n\n- **bytes**: 1,682 today → 538 proposed (saves 1,144)\n- **moves to**: `docs/research/the-wiser-civilization-as-successor-carlsmith-and-faggella.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Does an alignment philosopher want what Deliberus builds?\" / Carlsmith and Faggella on The Trajectory / the worthy successor as a civilization with AI integrated / \"collective epistemology in a good way\" / how Dan and Joe would read Deliberus** → `docs/research/the-wiser-civilization-as-successor-carlsmith-and-faggella.md` (no ledger rows): Carlsmith's worthy successor is a civilization with AI integrated, which shares one claim with the 2011 hybrid AGI sentence and is not the same position; fifteen meeting points with homes.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 137. \"How do I look at the live graph?\" / Lins / what draws the maps / Deliberus as a desktop app\n\n- **bytes**: 1,678 today → 458 proposed (saves 1,220)\n- **moves to**: `docs/research/graph-visualization.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"How do I look at the live graph?\" / Lins / what draws the maps / Deliberus as a desktop app** → `uv run python scripts/lins.py` (`--app --detach` from a session, never in its own background; `--corpus`, `--graph NAME`) + `docs/research/graph-visualization.md` § 11: RULED 2026-10-11 that the browser draws Lins's graph with Rust around it, do not re-open a Rust-drawn map; a desktop Deliberus is sized and not built (`deliberus-as-a-desktop-app.md`).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 63. Assembly Theory (Walker/Cronin) / \"the reuse mechanism has a physics\" / joint assembly\n\n- **bytes**: 1,664 today → 378 proposed (saves 1,286)\n- **moves to**: `docs/research/assembly-theory-and-the-reuse-mechanism.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Assembly Theory (Walker/Cronin) / \"the reuse mechanism has a physics\" / joint assembly** → `docs/research/assembly-theory-and-the-reuse-mechanism.md`: the assembly index is defined as construction steps with reuse, and identifying repeated units is its own hardest problem (claim-sameness); do NOT cite it as settled physics, and never use reuse alone as a quality signal.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 79. \"Is decomposition about reading the text, or reading around it?\" / the analytic half / coverage-coherence-atomicity\n\n- **bytes**: 1,637 today → 358 proposed (saves 1,279)\n- **moves to**: `docs/research/reading-the-text-itself.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is decomposition about reading the text, or reading around it?\" / the analytic half / coverage-coherence-atomicity** → `docs/research/reading-the-text-itself.md`: close reading and reading the pair are blind to each other; the literature supports maximal separation with theory in the prompt (not a parser), and coverage has no instrument here at all.\n\n\n### 90. What earns a line in this file: WRONG BY DEFAULT.\n\n- **bytes**: 1,631 today → 574 proposed (saves 1,057)\n- **moves to**: `docs/research/the-canon-distillation-pass.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **What earns a line in this file: WRONG BY DEFAULT.** A finding belongs in always-loaded memory only if a session lacking it would **confidently assert its opposite**, graded **demonstrated** (it happened on the record) or **predicted**; the budget is fixed, measured with `wc -c` and never quoted here, paid per session and again per helper that carries the files; nothing is filed by date (`tests/test_claude_md_shape.py`), and trimming by size is not trimming by value, so the filter runs over the cut (`docs/research/the-canon-distillation-pass.md` § 3, § 4, § 4b).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 128. \"Is the keeper's chain one inference per link?\" / checking a machine's chain without asking twice / leaps in a chain\n\n- **bytes**: 1,588 today → 406 proposed (saves 1,182)\n- **moves to**: `docs/research/one-inference-per-link.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is the keeper's chain one inference per link?\" / checking a machine's chain without asking twice / leaps in a chain** → `docs/research/one-inference-per-link.md` (AM11, AM14 ruled and built 2026-10-26; AM13 open): the keeper writes only the chain's sentences and each link's term, and code builds and checks each rule (19 of 20 real pairs right); nine held-out pairs await his judgment case by case.\n\n\n### 76. \"Should weighing detection stay deterministic?\" / the semantic tier / Swedish lexicon\n\n- **bytes**: 1,561 today → 351 proposed (saves 1,210)\n- **moves to**: `docs/research/weighing-detection-semantic-tier.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Should weighing detection stay deterministic?\" / the semantic tier / Swedish lexicon** → `docs/research/weighing-detection-semantic-tier.md` (NOT ratified): the corpus already answered twice for a semantic tier; 4/4 Swedish weighings found by a model where a hand lexicon catches about 1, and LLM-detected weighings need their own routing rule.\n\n\n### 21. The founding conviction: `Opacity Is a Cost, Not a Mystery`\n\n- **bytes**: 1,546 today → 440 proposed (saves 1,106)\n- **moves to**: `docs/research/fractal-scales-and-temporal-frame.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **The founding conviction: `Opacity Is a Cost, Not a Mystery`** → `docs/vision.md` § Opacity Is a Cost, Not a Mystery (ratified 2026-08-12); the naming arc in `docs/research/fractal-scales-and-temporal-frame.md` § The invariant's rank. One name everywhere: do NOT reintroduce \"the traversability premise\" as a handle, and keep both guards (it contains attunement; it dissolves confusion-conflict and only clarifies interest-conflict).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 34. Weighing evidence / meta-analyses / long-form or video sources\n\n- **bytes**: 1,546 today → 345 proposed (saves 1,201)\n- **moves to**: `docs/research/long-form-sources-and-meta-analysis-weighing.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Weighing evidence / meta-analyses / long-form or video sources** → `docs/research/long-form-sources-and-meta-analysis-weighing.md`: settled 2026-03-29 in his words (evidence decomposes as far as possible), do NOT re-derive it; methodology is assessed per source and should be per claim, and `heterogeneity` and `study_count` weigh nothing.\n\n\n### 22. Wright/\"The God Test\"/noosphere relation (Aug 2026)\n\n- **bytes**: 1,510 today → 303 proposed (saves 1,207)\n- **moves to**: `docs/research/wright-god-test-noosphere-and-deliberus.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Wright/\"The God Test\"/noosphere relation (Aug 2026)** → `docs/research/wright-god-test-noosphere-and-deliberus.md`: Wright's noosphere is the 2011 \"argument map + human society = hybrid AGI\" formulation arriving from outside, and his own market-selection thesis is the case for measured mediation.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 107. \"What becomes of motivated reasoning, groupthink and ideology in Deliberus?\" / causes of a belief against reasons for it / Bulverism / why someone believes\n\n- **bytes**: 1,500 today → 481 proposed (saves 1,019)\n- **moves to**: `docs/research/causes-of-belief-and-the-three-minds.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What becomes of motivated reasoning, groupthink and ideology in Deliberus?\" / causes of a belief against reasons for it / Bulverism / why someone believes** → `docs/research/causes-of-belief-and-the-three-minds.md` (ledger Arc AN): a cause enters the graph only as a claim, as its holder's own testimony or as a measured pattern; Deliberus never shows publicly why a person believes something (AN5, ruled), and a hidden guess at causes is special-category data (AN13, open).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 123. \"What does philosophy say about links and premises?\" / is the unwritten thing a premise or a licence / Carroll's regress / can everything be written out / which rule / shapes of a step\n\n- **bytes**: 1,458 today → 549 proposed (saves 909)\n- **moves to**: `docs/research/philosophy-of-the-unwritten-step.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What does philosophy say about links and premises?\" / is the unwritten thing a premise or a licence / Carroll's regress / can everything be written out / which rule / shapes of a step** → `docs/research/philosophy-of-the-unwritten-step.md` + `the-squirrel-and-the-strength-calculus.md` + `what-a-rule-rests-on.md` (AD22–AD29, unruled): only a fact about the case is a missing premise, and the rule is a licence proposed as the grounds of the link's relevance claim; do not use that evening's blind deductive reader as a judge of every step.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 163. \"Should this be a new property, label or link type?\" / listen to the architecture / parsimony and the core data model / the four products of a comparison\n\n- **bytes**: 1,426 today → 374 proposed (saves 1,052)\n- **moves to**: `docs/research/parsimony-and-the-core-data-model.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Should this be a new property, label or link type?\" / listen to the architecture / parsimony and the core data model / the four products of a comparison** → `docs/research/parsimony-and-the-core-data-model.md` (ledger Arc BL, BL1 and BL2 ruled): parsimony counted and the elegance of the theory behind a design weighed together, and the four products of a comparison.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 60. \"Is a missing consideration derivable, or does it need a human?\" / the derivability ladder / where the human-reserved residue actually lives\n\n- **bytes**: 1,386 today → 507 proposed (saves 879)\n- **moves to**: `docs/research/derivability-of-missing-considerations.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is a missing consideration derivable, or does it need a human?\" / the derivability ladder / where the human-reserved residue actually lives** → `docs/research/derivability-of-missing-considerations.md`: on 13 real arguments the model reached every consideration human commenters raised, so the reserved human act is evidence from a position; ⚠ \"a genuinely missing consideration is not derivable from the sentence\" is withdrawn, and the tally's zero at rung 5 is a lower bound, never a measurement.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 116. \"Is the graph storing meanings?\" / a graph of real things computed from the claims / senses as groups of uses\n\n- **bytes**: 1,385 today → 415 proposed (saves 970)\n- **moves to**: `docs/research/store-the-uses-read-the-meaning.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is the graph storing meanings?\" / a graph of real things computed from the claims / senses as groups of uses** → `docs/research/store-the-uses-read-the-meaning.md` (Arc AA; AA6 built: `author_wording` kept per claim) + `a-sentence-as-a-pointer-into-a-landscape.md`: a use is kept, a judgment recorded once, everything else read; a use can be a pointer with no paraphrase (`scripts/claim_wording_sources.py`).\n\n\n### 37. Reported speech / \"opponents argue that X\" / minting quoted content\n\n- **bytes**: 1,364 today → 358 proposed (saves 1,006)\n- **moves to**: `docs/research/self-similar-decomposition-and-claim-ontology.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Reported speech / \"opponents argue that X\" / minting quoted content** → `docs/research/self-similar-decomposition-and-claim-ontology.md` § RATIFIED: DECIDED 2026-08-15, always mint the reported content, with a non-propagating link type and asserted-versus-minted labelling on every published count; the payoff is that straw-manning becomes detectable.\n\n\n### 66. Evidence routing / \"can claim-level evidence be reused across subclaims?\" / double-counting\n\n- **bytes**: 1,363 today → 384 proposed (saves 979)\n- **moves to**: `docs/research/decomposition-axes.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Evidence routing / \"can claim-level evidence be reused across subclaims?\" / double-counting** → `deliberus/evidence_routing.py` + `docs/research/decomposition-axes.md` § 2b: built and tested, NOT adopted (adopting changes every strength, his call); routing abstains, never invents, and fair-share discounting corrects an inflation without creating one. Do NOT round the shares.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 135. \"Quality-controlled like the other microtasks\" / how deep may the machine write / what checks a step under a machine's own sentence\n\n- **bytes**: 1,354 today → 488 proposed (saves 866)\n- **moves to**: `docs/research/quality-control-for-deeper-steps.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Quality-controlled like the other microtasks\" / how deep may the machine write / what checks a step under a machine's own sentence** → `docs/research/quality-control-for-deeper-steps.md` (AG7–AG11, unruled): ten controls can be named and four run on live tasks today; do not say the small tasks have measured error rates, and do not treat a wider rule as a step down. The first-reading page is built (`/step-review/first-reading`, notes read with `scripts/read_review_notes.py`).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 136. \"Is this a claim, or part of the system?\" / are rules stored as claims / what hangs under a link / what the graph holds\n\n- **bytes**: 1,326 today → 543 proposed (saves 783)\n- **moves to**: `docs/research/claim-or-system.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is this a claim, or part of the system?\" / are rules stored as claims / what hangs under a link / what the graph holds** → `docs/research/claim-or-system.md` (§ 4a what was built) + `docs/architecture-reference.md` § Graph Database: RULED 2026-10-10 (AG12) that a machine's sentence under a link is a claim the moment it is written (kind `machine_step`, nobody asserts it, moves no strength); four questions sort a new thing (can two people disagree about it, does it say who did what, can it be computed again, what is left is form).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 129. \"Overlaps or analogy?\" / the keeper's likeness / a tree per claim / exact terms in microtask prompts\n\n- **bytes**: 1,315 today → 309 proposed (saves 1,006)\n- **moves to**: `docs/research/overlaps-and-analogy.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Overlaps or analogy?\" / the keeper's likeness / a tree per claim / exact terms in microtask prompts** → `docs/research/overlaps-and-analogy.md` + `a-tree-per-claim.md` (unruled; item 6 of the order of work): `overlaps` relates what two claims say, an analogy relates two cases through a shared pattern.\n\n\n### 68. How this project takes criticism / \"am I defending something I'm attached to?\" / internal-vs-external pressure\n\n- **bytes**: 1,304 today → 308 proposed (saves 996)\n- **moves to**: `docs/research/conviction-and-critique.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **How this project takes criticism / \"am I defending something I'm attached to?\" / internal-vs-external pressure** → `docs/research/conviction-and-critique.md`: take criticism on mechanism, refuse it on possibility, never let it choose the objective; the anti-abuse tell is a count, not an introspection.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 52. Agreement links between opponents / the stance guard / \"consider sameness first\"\n\n- **bytes**: 1,293 today → 393 proposed (saves 900)\n- **moves to**: `docs/research/the-remaining-case-for-the-stance-guard.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Agreement links between opponents / the stance guard / \"consider sameness first\"** → `docs/research/the-remaining-case-for-the-stance-guard.md` (all 84 live candidates read): RULED 2026-09-13 that a machine-written link failing a confident re-check is superseded automatically with the old link visible, and that sameness is decided before any relation (`auto-connect-upgrade.md` § 2c).\n\n\n### 45. Landing redesign / \"make the site less academic\" / the conviction-ladder page\n\n- **bytes**: 1,268 today → 434 proposed (saves 834)\n- **moves to**: `specs/landing-redesign/constellation-map.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Landing redesign / \"make the site less academic\" / the conviction-ladder page** → `specs/landing-redesign/` + `docs/research/site-declutter-directions.md` (the Socratic decisions) + `specs/landing-redesign/constellation-map.md` § DISPLAY STATE: since 2026-09-19 the landing opens on the constellation (14 ruled tiles bright, 31 drafts faint, the hero behind a flag, never deleted) with one hand-made example map above the wall.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 55. \"What errors pile up despite all our checking?\" / systematic error / residual taxonomy / frame lock-in / ingestion balance\n\n- **bytes**: 1,264 today → 398 proposed (saves 866)\n- **moves to**: `docs/research/the-residual-error-taxonomy.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What errors pile up despite all our checking?\" / systematic error / residual taxonomy / frame lock-in / ingestion balance** → `docs/research/the-residual-error-taxonomy.md`: step-checking guarantees local validity while systematic error lives in global properties no step owns; eight classes, the least defended at the graph's boundary (ingestion balance, frame lock-in, ratification decay).\n\n\n### 29. Dogfood run 4 — the agent-readable surface, read by an actual agent\n\n- **bytes**: 1,262 today → 287 proposed (saves 975)\n- **moves to**: `docs/research/dogfood-run-4-agent-surface.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Dogfood run 4 — the agent-readable surface, read by an actual agent** → `docs/research/dogfood-run-4-agent-surface.md`: five findings from an agent reading the agent-readable surface, four fixed; the surfaces that report on the honesty instruments had no instrument of their own.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 41. \"Is the always-mint decision's safety actually enforced?\" — not when audited 2026-08-16, when both requirements held by accident; both are closed since (2026-09-17 and 2026-10-10)\n\n- **bytes**: 1,261 today → 487 proposed (saves 774)\n- **moves to**: `docs/research/self-similar-decomposition-and-claim-ontology.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is the always-mint decision's safety actually enforced?\" — not when audited 2026-08-16, when both requirements held by accident; both are closed since (2026-09-17 and 2026-10-10)** → `docs/research/self-similar-decomposition-and-claim-ontology.md` § Both hard requirements + `the-residual-error-taxonomy.md`: when audited 2026-08-16 both held by accident (the badge query matched edges untyped; `count_safe_summary` had no caller); both closed since, 2026-09-17 and 2026-10-10.\n\n\n### 112. \"Can other people's AI tokens pay for extraction?\" / a decision model (Jev) in the pipeline\n\n- **bytes**: 1,241 today → 453 proposed (saves 788)\n- **moves to**: `docs/research/whose-tokens-pay-for-mapping.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Can other people's AI tokens pay for extraction?\" / a decision model (Jev) in the pipeline** → `docs/research/whose-tokens-pay-for-mapping.md` + `jev-and-the-judgement-layer.md` § 3f, § 6e (Arcs X, W; X9 built, off in production): do NOT design either from memory (the borrowed-model protocol feature was deprecated July 2026, a Claude plan may not be routed through another app); who pays never changes an answer and which model reads it does.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 7. What he held in the 2011–2013 Facebook group, in his own words\n\n- **bytes**: 1,240 today → 353 proposed (saves 887)\n- **moves to**: `docs/research/what-he-held-in-the-group-2011-2013.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **What he held in the 2011–2013 Facebook group, in his own words** → `docs/research/what-he-held-in-the-group-2011-2013.md` (his positions, each matched to the ruling it became) + `docs/research/what-the-group-held-2011-2017.md` (the other members' ideas and where each stands today); the archive is searchable with `mannaminne --fbgroup <words>`.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 109. \"Is Deliberus alignment infrastructure?\" / Altman's ladder (sandbox, monitor, trust) / ablation of safety features / training on the graph / the humans' side of alignment\n\n- **bytes**: 1,239 today → 572 proposed (saves 667)\n- **moves to**: `docs/research/alignment-infrastructure-red-team.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is Deliberus alignment infrastructure?\" / Altman's ladder (sandbox, monitor, trust) / ablation of safety features / training on the graph / the humans' side of alignment** → `docs/research/alignment-infrastructure-red-team.md` (ledger Arc BF; BF1 and BF5 ruled and blessed): a check against the record reaches a model's stated reasoning and never its policy, so the project claims the second rung; a record binds nobody by what it shows (the bound decider). Do not say the graph is never a training target (BF3, open, forbids only reward and published evaluations).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 157. Collapsible inference leaps / folded routes / split to the floor and join for display / horizontal and vertical folding / zoom levels\n\n- **bytes**: 1,236 today → 539 proposed (saves 697)\n- **moves to**: `docs/research/folded-routes-and-shared-premises.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Collapsible inference leaps / folded routes / split to the floor and join for display / horizontal and vertical folding / zoom levels** → `docs/research/folded-routes-and-shared-premises.md` + `folding-joining-and-zoom.md` (AX4, AX8–AX12 ruled; unbuilt, mockups follow his agreement): claims stored at the separability floor, folds and joins computed on read and never stored, a long leap folded into a route that shows its weakest step. Do not rehome a whole whose part links are unsorted (they are mostly reasons filed as parts).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 82. \"Can Deliberus keep an outside judge levelheaded?\"\n\n- **bytes**: 1,187 today → 347 proposed (saves 840)\n- **moves to**: `docs/research/steadying-an-outside-judge.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Can Deliberus keep an outside judge levelheaded?\"** → `docs/research/steadying-an-outside-judge.md`: four read-only mechanisms (fabricated consensus becomes a lookup, a known move is visible, a judge is corrected without being persuadable, contested ground is predictable); an unattacked claim may be unexamined rather than uncontroversial.\n\n\n### 84. The analogy daemon / \"daemons discovering kinship and fractal priors\" / the graph's gestalt organ\n\n- **bytes**: 1,179 today → 326 proposed (saves 853)\n- **moves to**: `docs/research/the-analogy-daemon.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **The analogy daemon / \"daemons discovering kinship and fractal priors\" / the graph's gestalt organ** → `docs/research/the-analogy-daemon.md` (founder vision verbatim): the first sanctioned roaming philosopher, two passes in two model families, propose-only, every analogy claim shipped with its strongest counter-analogy.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 85. The 2026-08-30 source intake (four podcasts + one article), each read in full against the corpus\n\n- **bytes**: 1,159 today → 1,159 proposed (saves 0)\n- **proposal**: unchanged\n\n### 110. \"Can Deliberus help coordinate the AI race?\" / shared sensemaking / the AI-risk intake (2026-10-01 → 03)\n\n- **bytes**: 1,134 today → 473 proposed (saves 661)\n- **moves to**: `docs/research/shared-sensemaking-what-deliberus-holds-and-owes.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Can Deliberus help coordinate the AI race?\" / shared sensemaking / the AI-risk intake (2026-10-01 → 03)** → `docs/research/shared-sensemaking-what-deliberus-holds-and-owes.md` (entry point) + `ai-risk-for-mathematicians-and-the-coordination-game.md` + `schmachtenberger-and-the-ai-race.md` + `four-cuts-of-the-ai-debate.md` (Arcs U, V, unruled): a shared assessment is necessary, not sufficient; what a public map serves is political will, not bilateral assurance.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 74. \"Does AI erode the expertise needed to check AI?\" / cognitive commons / Validation Tether / detect-but-defer\n\n- **bytes**: 1,132 today → 349 proposed (saves 783)\n- **moves to**: `docs/research/cognitive-commons-and-deliberus.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Does AI erode the expertise needed to check AI?\" / cognitive commons / Validation Tether / detect-but-defer** → `docs/research/cognitive-commons-and-deliberus.md`: the Validation Tether and detect-but-defer (80.7% detected, followed anyway) give the addressability bet its external mechanism; depletion is a prediction, never an arrived fact.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 78. \"Does the pipeline under-extract implicit premises?\" — the raw ratio is NOT the figure, and the corpus used to imply otherwise\n\n- **bytes**: 1,119 today → 418 proposed (saves 701)\n- **moves to**: `docs/research/self-similar-decomposition-and-claim-ontology.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Does the pipeline under-extract implicit premises?\" — the raw ratio is NOT the figure, and the corpus used to imply otherwise** → `docs/research/self-similar-decomposition-and-claim-ontology.md` § How much implicit-premise extraction was ever ATTEMPTED: the raw ratio is NOT the figure; 10 pipeline-produced premises across the 107 arguments the pass saw is consistent with a prompt that says \"prefer fewer\".\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 151. \"How did the project evolve, and what did 2026 add?\" / the long arc from 2009 / where the binding constraint moved / the readiness gate assessed\n\n- **bytes**: 1,119 today → 463 proposed (saves 656)\n- **moves to**: `docs/research/how-the-project-evolved-read-from-the-record.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"How did the project evolve, and what did 2026 add?\" / the long arc from 2009 / where the binding constraint moved / the readiness gate assessed** → `docs/research/how-the-project-evolved-read-from-the-record.md` (a session's reading, unruled; AV4–AV6): the design was his by 2011, 2026 added the parts and the discipline, and the binding constraint is read as moving to contact. Do not say nothing counts stranger visits (the daily digest does, by path).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 61. Perceiving possibility / creativity vs recombination / Hofstadter and Copycat / the human-seeded analogy channel\n\n- **bytes**: 1,114 today → 349 proposed (saves 765)\n- **moves to**: `docs/research/perceiving-possibility-and-the-analogy-organ.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Perceiving possibility / creativity vs recombination / Hofstadter and Copycat / the human-seeded analogy channel** → `docs/research/perceiving-possibility-and-the-analogy-organ.md`: the daemon design is Copycat re-described and temperature its missing piece; the human-seeded analogy channel is founder-endorsed (`the-analogy-daemon.md` § 7).\n\n\n### 119. A video's twelve books read against Deliberus / Lakatos, Lakoff, Ashby, Toulmin and Hitchcock, Scott, Gigerenzer, Hofstadter and Sander, Mercier and Sperber, and four more / \"what does book X bear on here?\"\n\n- **bytes**: 1,108 today → 595 proposed (saves 513)\n- **moves to**: `docs/research/twelve-books-read-against-deliberus.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **A video's twelve books read against Deliberus / Lakatos, Lakoff, Ashby, Toulmin and Hitchcock, Scott, Gigerenzer, Hofstadter and Sander, Mercier and Sperber, and four more / \"what does book X bear on here?\"** → `docs/research/twelve-books-read-against-deliberus.md` + `twelve-books-in-plain-words.md` (from reviews, none of the books read; nineteen candidates held in § 16 with no ledger row, on his word); the exact transcription method is beside the transcript in `.private/sources/twelve-books-video-2026-10/`. Do not say Ashby or Lakoff's categories book were in the corpus before it.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 122. \"Haven't we thought a lot about unwritten premises?\" / implicit premises / the dark matter of reasoning / writing a link out in full / which small tasks can be leaned on\n\n- **bytes**: 1,065 today → 512 proposed (saves 553)\n- **moves to**: `docs/research/writing-out-the-unwritten.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Haven't we thought a lot about unwritten premises?\" / implicit premises / the dark matter of reasoning / writing a link out in full / which small tasks can be leaned on** → `docs/research/writing-out-the-unwritten.md` (AD13–AD21, unruled): every link written out in one form and the writing checked, no link judged; the checks sort links by structure, not good reasons from bad. ⚠ *never generating premises, only finding real ones* was Claude's sentence of 2026-03-29; his said *finding or creating*.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 164. \"Was capitalism a historical accident?\" / channelling self-interest / the shadow in the language of levels / Unger, Althusser's encounter, Hirschman, Benkler, market socialism, Ostrom's principles generalised\n\n- **bytes**: 1,058 today → 428 proposed (saves 630)\n- **moves to**: `docs/research/historical-accident-and-channelled-self-interest.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Was capitalism a historical accident?\" / channelling self-interest / the shadow in the language of levels / Unger, Althusser's encounter, Hirschman, Benkler, market socialism, Ostrom's principles generalised** → `docs/research/historical-accident-and-channelled-self-interest.md` (no ledger rows; three proposals): his counterfactual read against the literature, Benkler's mechanism that a channel breeds what it assumes.\n\n\n### 6. The best short public-facing copy the project has\n\n- **bytes**: 1,053 today → 336 proposed (saves 717)\n- **moves to**: `docs/research/live-election-test-design.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **The best short public-facing copy the project has** → `docs/research/live-election-test-design.md` § **Inbjudningstexten**: the founder-approved Swedish invitation (four paragraphs, 2026-08-22), filed by its purpose rather than by what it is, and the template for any short public copy; an English version needs its own occasion.\n\n\n### 102. Two kinds of cut, and whether splitting depends on the concept layer\n\n- **bytes**: 1,021 today → 342 proposed (saves 679)\n- **moves to**: `docs/research/core-periphery-and-the-concept-layer.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Two kinds of cut, and whether splitting depends on the concept layer** → `docs/research/core-periphery-and-the-concept-layer.md` § 5a–5a-ii: a cut between propositions is concept-independent and a cut through a concept is usually an argument, not a split; `CoverageReport.reconstructed` runs the discriminator and nothing acts on it.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 75. Wild weighing dialects / \"what kinds of weighing exist?\" / the 14-question set\n\n- **bytes**: 1,009 today → 322 proposed (saves 687)\n- **moves to**: `docs/research/wild-weighing-dialects-and-the-question-set.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Wild weighing dialects / \"what kinds of weighing exist?\" / the 14-question set** → `docs/research/wild-weighing-dialects-and-the-question-set.md` (founder-blessed 2026-08-25): ten species from 72 wild Swedish sentences, four new questions (14 in all); Chang's small-improvement probe is held back, do not presume it.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 162. \"Listen to the architecture\" / SOAR, impasses and chunking / a new capability through what exists / which stored properties the project should never have added\n\n- **bytes**: 997 today → 358 proposed (saves 639)\n- **moves to**: `docs/research/soar-listen-to-the-architecture.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Listen to the architecture\" / SOAR, impasses and chunking / a new capability through what exists / which stored properties the project should never have added** → `docs/research/soar-listen-to-the-architecture.md` (it led to Arc BL): Newell's maxim as the claim, record, reading and form sort, and where the project followed it and where it bolted on.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 67. Attribution normalization / `attributed_to` / who-said-vs-who-is-quoted\n\n- **bytes**: 974 today → 367 proposed (saves 607)\n- **moves to**: `docs/research/self-similar-decomposition-and-claim-ontology.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Attribution normalization / `attributed_to` / who-said-vs-who-is-quoted** → `deliberus/attribution.py`: populated on 1,082 of 1,642 substantive claims; the first run (129 → 109 entities) refuted the role taxonomy (`third_party` bundles a supporting authority with a reported opponent); nothing written to the graph yet, and the write-time gate must fail open.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 153. \"Who checks the machine, who checks the person, and how much has been tried?\" / saboteurs and spam / a machine reading on what people add / coverage of what could be asked\n\n- **bytes**: 967 today → 435 proposed (saves 532)\n- **moves to**: `docs/research/checks-on-machines-and-people.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Who checks the machine, who checks the person, and how much has been tried?\" / saboteurs and spam / a machine reading on what people add / coverage of what could be asked** → `docs/research/checks-on-machines-and-people.md` (advice, unruled; bears on O13, W5, AD7, Q34): nothing reads what a person adds today, and the proposed part test would withhold 28 of the 39 parts people have attached, 24 of them reasons filed as parts.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 108. \"Does permissivism count against convergence?\" / \"do most people have good hearts?\" / the theorems behind the wager / what a lens is\n\n- **bytes**: 953 today → 530 proposed (saves 423)\n- **moves to**: `docs/research/good-hearts-and-permissivism.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Does permissivism count against convergence?\" / \"do most people have good hearts?\" / the theorems behind the wager / what a lens is** → `docs/research/good-hearts-and-permissivism.md` (BD25–BD28 ruled): permissivism counts against convergence only where evidence cannot settle it, a lens is one person's setting of the contested choices, and the merging theorems need a common prior and an agreed reading; do not argue that counting permissivism makes the wager lose because credences never match (withdrawn on his word).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 121. \"Is that premise really one step from the claim?\" / far links / chains hidden inside a link / \"can a machine check a link?\" / what a link rests on\n\n- **bytes**: 949 today → 422 proposed (saves 527)\n- **moves to**: `docs/research/how-far-a-premise-is-from-its-claim.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is that premise really one step from the claim?\" / far links / chains hidden inside a link / \"can a machine check a link?\" / what a link rests on** → `docs/research/how-far-a-premise-is-from-its-claim.md` + `what-a-link-rests-on.md` (Arc AD, unruled): the linker records no distance, four ways of having a machine judge a link failed in one day, and naming alone he called \"weak sauce\" (revised in the entry below).\n\n\n### 113. \"How good must a microtask be, and which model should run it?\" / choosing a model for a microtask\n\n- **bytes**: 943 today → 381 proposed (saves 562)\n- **moves to**: `docs/research/pass-quality-and-the-choice-of-model.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"How good must a microtask be, and which model should run it?\" / choosing a model for a microtask** → `docs/research/pass-quality-and-the-choice-of-model.md` (ledger Arc Y; § 6b the page he labels on): do NOT choose a model by its tier, since 17 of 19 measured failures trace to how the question was posed or the receiving code; no person has labelled a machine verdict yet.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 130. \"Is there a spaCy for logic?\" / software that checks an inference by its kind / clingo, ProbLog, ASPIC+ / the checker talking back\n\n- **bytes**: 941 today → 462 proposed (saves 479)\n- **moves to**: `docs/research/tools-for-checking-inferences.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Is there a spaCy for logic?\" / software that checks an inference by its kind / clingo, ProbLog, ASPIC+ / the checker talking back** → `docs/research/tools-for-checking-inferences.md` (AM16, AM17 open; AM18 built, `deliberus/rules_for_one_case.py`): no parser turns open English into logic, engines exist per kind, and two rules for one case never reach an engine together. Do not say code checks every kind's form: two trial scripts on hand-written cases.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 139. \"What is all this for?\" / the purposes of Deliberus and of the sessions / which purpose a day's work serves\n\n- **bytes**: 912 today → 362 proposed (saves 550)\n- **moves to**: `docs/research/what-deliberus-is-for.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What is all this for?\" / the purposes of Deliberus and of the sessions / which purpose a day's work serves** → `docs/research/what-deliberus-is-for.md` (AH1–AH3, unruled): two lists of purposes, both Claude's compressions; do not quote either as his or make one resident before he rules; the habit in § 6 (say which purpose the day serves) may run now.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 11. The two founder-ratified instruments\n\n- **bytes**: 903 today → 410 proposed (saves 493)\n- **moves to**: `docs/research/weighing-scheme-and-terminus-classifier-design.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **The two founder-ratified instruments** → `docs/research/weighing-scheme-and-terminus-classifier-design.md` (both root questions resolved Jul 6; code in `deliberus/weighing.py`, `deliberus/terminus_llm.py`): the terminus setter was retired 2026-09-16 (a kind is read, never set, `deliberus/horizon.py`), and the sacredness question was ratified in substance 2026-08-25 (`parity-and-the-zone-interior.md`).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 148. \"Who has the burden of proof?\" / extraordinary claims / presumption / precaution / which side must prove\n\n- **bytes**: 879 today → 342 proposed (saves 537)\n- **moves to**: `docs/research/burden-of-proof-and-priors.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Who has the burden of proof?\" / extraordinary claims / presumption / precaution / which side must prove** → `docs/research/burden-of-proof-and-priors.md` (Arc AZ, unruled): the phrase names three things (a claim's prior, the obligation to argue, a decision's threshold from error costs); do not cite self-affirmation as a design lever.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 46. \"Does thinking work better in groups?\" / the group premise / collective intelligence for and against\n\n- **bytes**: 863 today → 292 proposed (saves 571)\n- **moves to**: `docs/research/the-group-premise-evidence.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Does thinking work better in groups?\" / the group premise / collective intelligence for and against** → `docs/research/the-group-premise-evidence.md`: the premise holds under five conditions and the graph supplies three; the fifth, that the minds actually differ, must be selected for.\n\n\n### 70. Ask-or-act / question pricing / \"when should the descent stop asking?\" / active inference\n\n- **bytes**: 856 today → 331 proposed (saves 525)\n- **moves to**: `docs/research/active-inference-context-acquisition-and-deliberus.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Ask-or-act / question pricing / \"when should the descent stop asking?\" / active inference** → `docs/research/active-inference-context-acquisition-and-deliberus.md`: ask exactly when information gain exceeds priced cost; `worth_asking` shipped 2026-08-23 as display-band crossing, with the gap shown as a state below the line.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 131. \"Has a later ruling made this proposal unnecessary?\" / rulings that settle something together\n\n- **bytes**: 856 today → 331 proposed (saves 525)\n- **moves to**: `docs/research/catching-a-proposal-made-unnecessary.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Has a later ruling made this proposal unnecessary?\" / rulings that settle something together** → `docs/research/catching-a-proposal-made-unnecessary.md` (BE1–BE3 ruled and built): `scripts/ruling_ripples.py <row> --meaning` puts a ruling to every open row, and `--rederive` finds what several rulings settle only together.\n\n\n### 142. \"What differentiates a part from a reason?\" / a reason filed as a part / parts against evidence, premises or a rule\n\n- **bytes**: 847 today → 440 proposed (saves 407)\n- **moves to**: `docs/research/a-part-and-a-reason.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What differentiates a part from a reason?\" / a reason filed as a part / parts against evidence, premises or a rule** → `docs/research/a-part-and-a-reason.md` (unruled, sharpens T9): if the sentence fell, would the claim be false, or only have lost an argument; a part is presumed needed and a reason is not (Q34). Do not say a reason is never a part, nor that a claim whose every reason was refuted reads low (it reads *unexamined*).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 89. The four open reviews are ONE activity, and there is a register\n\n- **bytes**: 822 today → 340 proposed (saves 482)\n- **moves to**: `docs/research/the-canon-distillation-pass.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **The four open reviews are ONE activity, and there is a register** → `TODO.md` § OPEN REVIEWS + `docs/research/the-canon-distillation-pass.md` (method, budget, filter): a position flows asserted → ruled → compressed → resident, Arc J went first; the ledger is a wheel on the site at `/ledger` (`specs/map-view/design.md` § 3c).\n\n\n### 5. Current dogfooding phase (2026-04-24 pivot)\n\n- **bytes**: 813 today → 512 proposed (saves 301)\n- **moves to**: `docs/research/signposts-for-the-whole-project.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Current dogfooding phase (2026-04-24 pivot)** → `TODO.md` § \"Dogfooding phase\" + `docs/research/signposts-for-the-whole-project.md` § 4 (his words and the eight readiness parts, ledger AV2). ⚠ Restated by him 2026-10-22: he tells people about Deliberus widely himself; what waits is asking them to try it. Ruled 2026-11-02 (AV4, the gate is per circle): the eight parts gate the public rung alone; a friend side by side waits only on the door offering the handle (P20, P21), and a room is a dated demo.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 155. \"What does a qualifying (yellow) link mean?\" / *qualifies*, *reframes*, *defines* / a link that neither supports nor attacks\n\n- **bytes**: 811 today → 340 proposed (saves 471)\n- **moves to**: `docs/research/the-yellow-link.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What does a qualifying (yellow) link mean?\" / *qualifies*, *reframes*, *defines* / a link that neither supports nor attacks** → `docs/research/the-yellow-link.md` (Arc BB, BB1–BB4 ruled): the qualifying link was a menu slot, not a kind; no pass writes one since 2026-10-24, and the 538 stored wait on the keeper's re-read (§ 4.4).\n\n\n### 124. \"Can a second machine tidy what the first one found?\" / machine proofs and a curated library (Mathlib) / what tidying may do with no checker / how wide a rule should be\n\n- **bytes**: 795 today → 461 proposed (saves 334)\n- **moves to**: `docs/research/alien-proofs-and-the-library.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Can a second machine tidy what the first one found?\" / machine proofs and a curated library (Mathlib) / what tidying may do with no checker / how wide a rule should be** → `docs/research/alien-proofs-and-the-library.md` (Arc AF, unruled): the checker stops at the proof and what belongs in the library is judged there too; five tests no person has to judge for any pass that rewrites or regroups (§ 4.5). Do not restate the post's last step as practice.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 140. \"The summary a friend's assistant wrote\" / an outside model's reading of the thesis / Astra (GPT-6) on Deliberus / \"the most coherent rational statement\"\n\n- **bytes**: 780 today → 484 proposed (saves 296)\n- **moves to**: `docs/research/an-outside-reading-of-the-thesis.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"The summary a friend's assistant wrote\" / an outside model's reading of the thesis / Astra (GPT-6) on Deliberus / \"the most coherent rational statement\"** → `docs/research/an-outside-reading-of-the-thesis.md` (a record, nothing ruled; the friend named only in `.private/sources/`): the model had read three public pages, so where it agrees it reads the wall back, and it ends at an observatory, the floor without the wager. Search for this by subject, not by the friend's name.\n\n\n### 104. \"What counts as evidence?\" / tiers of evidence / GRADE / testimony / how evidence should enter the graph\n\n- **bytes**: 766 today → 409 proposed (saves 357)\n- **moves to**: `docs/research/what-counts-as-evidence.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What counts as evidence?\" / tiers of evidence / GRADE / testimony / how evidence should enter the graph** → `docs/research/what-counts-as-evidence.md` (dossier in `what-counts-as-evidence/`; ledger Arc Q): answers to critical questions are counted where every tradition weighs them, the methodology ladder is read by no badge, independence is unrepresented; code defects in `TODO.md` § Evidence layer.\n\n\n### 15. Forecasting-ceiling relation (ACX Aug 2026)\n\n- **bytes**: 765 today → 249 proposed (saves 516)\n- **moves to**: `docs/research/forecasting-room-at-the-top-and-deliberus.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Forecasting-ceiling relation (ACX Aug 2026)** → `docs/research/forecasting-room-at-the-top-and-deliberus.md`: the convergence wager rotated 90°; the fork on calling the residue map the aleatoric fraction of disagreement is open, NOT ratified.\n\n\n### 91. Flag versus reasons — the variable in durable attitude change is neither dose nor duration.\n\n- **bytes**: 759 today → 759 proposed (saves 0)\n- **proposal**: unchanged\n\n### 149. \"Can a reader grant premises and see what follows?\" / the what-if view / a worldview lens as a computation / the mind-meld between two people / a policy trial\n\n- **bytes**: 744 today → 423 proposed (saves 321)\n- **moves to**: `docs/research/what-if-and-lenses.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Can a reader grant premises and see what follows?\" / the what-if view / a worldview lens as a computation / the mind-meld between two people / a policy trial** → `docs/research/what-if-and-lenses.md` (Arc AU, unruled): one computation fed from a reader's suppositions, a holder's premises or a policy's rules, already in code as `compute_badge_recursive`'s `overrides`; the strength rules come before the view (AU5).\n\n\n### 154. \"How much does one person's answer move a claim?\" / answers to critical questions / one answer per account / when a badge shows\n\n- **bytes**: 732 today → 375 proposed (saves 357)\n- **moves to**: `docs/research/weighing-answers-and-the-badge-threshold.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"How much does one person's answer move a claim?\" / answers to critical questions / one answer per account / when a badge shows** → `docs/research/weighing-answers-and-the-badge-threshold.md` (Q3 built 2026-10-24): answers are weighed, not counted, an unbacked answer counts at the lowest tier (0.675), never at zero, and a badge shows from three weighed contributions.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 133. Lean / mathlib / the small trusted checker / \"what does Lean teach Deliberus?\"\n\n- **bytes**: 730 today → 355 proposed (saves 375)\n- **moves to**: `docs/research/lean-deliberus-analogies.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Lean / mathlib / the small trusted checker / \"what does Lean teach Deliberus?\"** → `docs/research/lean-deliberus-analogies.md` (§ 9 against the small-step work, unruled) + `formalized-fermat-and-the-oracle-boundary.md` + `alien-proofs-and-the-library.md`; do not expect Lean's training flywheel, since its kernel certifies truth and ours only form.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 141. Wittgenstein's Tractatus / \"is this the Tractatus's project?\" / a reframe answered with a rival map / internal and external relations\n\n- **bytes**: 722 today → 422 proposed (saves 300)\n- **moves to**: `docs/research/the-tractatus-and-deliberus.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Wittgenstein's Tractatus / \"is this the Tractatus's project?\" / a reframe answered with a rival map / internal and external relations** → `docs/research/the-tractatus-and-deliberus.md` (Arc AJ, seven rows unruled): the same ambition without final atoms, a limit fixed in advance or one mind alone; the splitting prompt runs the early Wittgenstein under Russell's name while the corpus cites the later one against it.\n\n\n### 159. Analogy in scientific discovery / tools as the engine of discovery / Mitchell, Copycat and the daemon's judge / Krauss's 761 breakthroughs and the why-now base rate / a borrowed word's unstated load\n\n- **bytes**: 718 today → 456 proposed (saves 262)\n- **moves to**: `docs/research/tools-and-analogies-as-engines-of-discovery.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Analogy in scientific discovery / tools as the engine of discovery / Mitchell, Copycat and the daemon's judge / Krauss's 761 breakthroughs and the why-now base rate / a borrowed word's unstated load** → `docs/research/tools-and-analogies-as-engines-of-discovery.md` (ledger Arc BH, four rows open): six real analogies with known outcomes through the analogy checks, and Krauss's 761 discoveries as the base rate for discoveries following their tools.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 27. Red-team second wave + synthesis (Jul 2026)\n\n- **bytes**: 714 today → 258 proposed (saves 456)\n- **moves to**: `docs/research/red-team-synthesis-2026-07.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Red-team second wave + synthesis (Jul 2026)** → `docs/research/red-team-synthesis-2026-07.md` (entry point; founder decisions ranked) + `legibility-under-power-red-team.md` + `philosophical-foundations-stress-test.md` + `threads-synthesis-and-gaps.md`.\n\n\n### 118. Machine consciousness / \"is the model conscious?\" / the map of that debate\n\n- **bytes**: 713 today → 357 proposed (saves 356)\n- **moves to**: `docs/research/mapping-the-machine-consciousness-debate.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Machine consciousness / \"is the model conscious?\" / the map of that debate** → `docs/research/mapping-the-machine-consciousness-debate.md` (+ `four-cuts-of-the-ai-debate.md` § 5): eight predictions scored (three right, three half, two wrong); do NOT restate the picture of two sides, the map has four voices and each text's hinge sentence is missing.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 13. Dogfood run 2 (G1–G10)\n\n- **bytes**: 705 today → 220 proposed (saves 485)\n- **moves to**: `docs/research/dogfood-run-2-orthogonal-experiments.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Dogfood run 2 (G1–G10)** → `docs/research/dogfood-run-2-orthogonal-experiments.md`: the hinge whispers at light engagement, the first disagreement-preservation dataset (0.82–0.94), the residue map at 2 / 1 / 1.\n\n\n### 134. \"Can Deliberus's own rules be argued in its graph?\" / \"can a compartment's settled rules run software?\"\n\n- **bytes**: 703 today → 425 proposed (saves 278)\n- **moves to**: `docs/research/claims-and-code-one-loop.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Can Deliberus's own rules be argued in its graph?\" / \"can a compartment's settled rules run software?\"** → `docs/research/claims-and-code-one-loop.md` (unruled; self-hosting held by his word of 2026-08-30 until the reviews are much closer to done): code into claims is the August *publish it* move, claims into code his greenhouse idea for a blob, with a small trusted core never argued at the moment it applies (AJ4).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 26. Frontier extraction experiment (run 3F)\n\n- **bytes**: 688 today → 312 proposed (saves 376)\n- **moves to**: `docs/research/frontier-extraction-experiment.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Frontier extraction experiment (run 3F)** → `docs/research/frontier-extraction-experiment.md`: the quality-ceiling benchmark by hand. ⚠ Corrected 2026-10-10: the crux is left unsaid by at least one side (by both in the first debate, by one in the second), not \"stated on one side, implicit on the other\".\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 23. Science lineage / \"is Deliberus the extension of science?\"\n\n- **bytes**: 687 today → 336 proposed (saves 351)\n- **moves to**: `docs/research/science-lineage-philosophy.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Science lineage / \"is Deliberus the extension of science?\"** → `docs/research/science-lineage-philosophy.md` + `docs/research/science-reasoning-infrastructure.md`: the safe claim is \"extends science as public error-correction\", never \"applies the scientific method to everything\"; the Legibility Rule is a candidate, not ratified.\n\n\n### 150. \"What is the feed for, and what should it rank by?\" / the triage feed made concrete / feed modes\n\n- **bytes**: 684 today → 261 proposed (saves 423)\n- **moves to**: `docs/research/the-feed-reconsidered.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What is the feed for, and what should it rank by?\" / the triage feed made concrete / feed modes** → `docs/research/the-feed-reconsidered.md` (Arc AW; AW1 ruled, the two halves): work and reading, no invitation to vote, no stance tally ordering attention.\n\n\n### 132. \"Can similarity tell a claim from its denial?\" / negation / phase / the contradiction model\n\n- **bytes**: 669 today → 386 proposed (saves 283)\n- **moves to**: `docs/research/resonance-db-and-negation.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Can similarity tell a claim from its denial?\" / negation / phase / the contradiction model** → `docs/research/resonance-db-and-negation.md`: a claim and its denial sit close in embedding space (0.78–0.89), so cosine cannot tell them apart and the keeper reads polarity; the contradiction model (`deliberus/contradiction_order.py`) orders the keeper's sweep and decides nothing.\n\n\n### 152. \"What does Deliberus assume, and what sign would say an assumption is failing?\" / signposts and triggers / when the platform is ready to be tried\n\n- **bytes**: 661 today → 390 proposed (saves 271)\n- **moves to**: `docs/research/signposts-for-the-whole-project.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What does Deliberus assume, and what sign would say an assumption is failing?\" / signposts and triggers / when the platform is ready to be tried** → `docs/research/signposts-for-the-whole-project.md` (Arc AV, unruled; read each Monday by the flywheel check-in): twenty-five assumptions, each to carry a trigger both ways, a review date with a home and one check against ground truth.\n\n\n### 47. \"How much engagement can we expect?\" / participation gradient / \"most people won't contribute\"\n\n- **bytes**: 656 today → 309 proposed (saves 347)\n- **moves to**: `docs/research/engagement-gradient-priors.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"How much engagement can we expect?\" / participation gradient / \"most people won't contribute\"** → `docs/research/engagement-gradient-priors.md`: 90-9-1 is the expected baseline, a reader who never contributes is a success, and the falsifier is the decay curve within a tier, never the tier populations.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 120. Words people want / \"prime real estate\" / projection surfaces / definitional moves and fights over a word\n\n- **bytes**: 646 today → 364 proposed (saves 282)\n- **moves to**: `docs/research/prime-real-estate-words.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Words people want / \"prime real estate\" / projection surfaces / definitional moves and fights over a word** → `docs/research/prime-real-estate-words.md` (Arc AC, all four ruled 2026-10-07, unbuilt): how much a word is wanted is a second property beside how shared its meaning is, a definition of a wanted word is a claim on it, and nobody owns a wanted word.\n\n\n### 111. \"How many AI agents will there be, and what is checking up against?\" / the agent population / persuasion's compute against checking's\n\n- **bytes**: 608 today → 350 proposed (saves 258)\n- **moves to**: `docs/research/the-agent-population-and-what-checking-is-up-against.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"How many AI agents will there be, and what is checking up against?\" / the agent population / persuasion's compute against checking's** → `docs/research/the-agent-population-and-what-checking-is-up-against.md` (U14–U17, X11, unruled): persuasion needs only compute while checking needs a record too, so checking keeps pace only through reuse.\n\n\n### 72. Temporal rung / staleness / \"does anything age?\" / Graphiti bitemporal\n\n- **bytes**: 582 today → 300 proposed (saves 282)\n- **moves to**: `docs/research/supersession-and-bitemporal-lifecycles.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Temporal rung / staleness / \"does anything age?\" / Graphiti bitemporal** → `docs/research/supersession-and-bitemporal-lifecycles.md`: stale, superseded and refuted are three states on three axes, never one flag; the deterministic staleness daemon and ratified `invalid_at` with verdict grounds.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 71. Synthetic stress suite / \"why did the badge change\" / edge supersession / the five strength-layer gaps\n\n- **bytes**: 579 today → 284 proposed (saves 295)\n- **moves to**: `docs/research/synthetic-stress-suite-and-ontology-reflection.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Synthetic stress suite / \"why did the badge change\" / edge supersession / the five strength-layer gaps** → `docs/research/synthetic-stress-suite-and-ontology-reflection.md`: four built-to-break examples live in the graph, five gaps fixed 2026-08-22, edge supersession born here.\n\n\n### 9. Active relationships / named people / approach notes\n\n- **bytes**: 569 today → 373 proposed (saves 196)\n- **moves to**: `.private/docs_visibility_boundary_2026_05_17.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Active relationships / named people / approach notes** → **`.private/`** (flat: `contact_register_2026_08_17.md`, `workshop_network_analysis_2026_08_17.md`, per-person notes). ⚠ Never `docs/relationships/`: that directory was publicly served, and a note about a named person sat live on deliberus.com for 15 days (`.private/docs_visibility_boundary_2026_05_17.md`).\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 25. Deep-descent experiment\n\n- **bytes**: 567 today → 267 proposed (saves 300)\n- **moves to**: `docs/research/deep-descent-experiment.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Deep-descent experiment** → `docs/research/deep-descent-experiment.md` (+ `deep-descent-experiment/`): two everyday claims hand-walked to bedrock; the residue taxonomy is ought-biased (pain's badness hit a hard-problem terminus none of the four types can name).\n\n\n### 115. \"What does record linkage teach us?\" / a record of every machine judgment / two kinds of evidence for sameness\n\n- **bytes**: 554 today → 362 proposed (saves 192)\n- **moves to**: `docs/research/what-a-record-linkage-project-teaches-deliberus.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What does record linkage teach us?\" / a record of every machine judgment / two kinds of evidence for sameness** → `docs/research/what-a-record-linkage-project-teaches-deliberus.md` (Arc Z; the friend named only in `.private/`) + `storing-the-record-of-machine-judgments.md` § 7a: two kinds of evidence for sameness (`scripts/sameness_evidence_kinds.py`).\n\n\n### 117. \"What relation between people and machines are we building?\" / cognition as an ecology / why shared sensemaking now\n\n- **bytes**: 544 today → 314 proposed (saves 230)\n- **moves to**: `docs/research/a-cognitive-ecology-and-the-terms-of-the-relation.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"What relation between people and machines are we building?\" / cognition as an ecology / why shared sensemaking now** → `docs/research/a-cognitive-ecology-and-the-terms-of-the-relation.md` (Arc AB, unruled): what has been measured is the parties becoming one voice, and the project's stance is the breeder's.\n\n\n### 14. Bengio/LawZero/Scientist AI relation\n\n- **bytes**: 531 today → 246 proposed (saves 285)\n- **moves to**: `docs/research/bengio-safety-from-honesty-and-deliberus.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Bengio/LawZero/Scientist AI relation** → `docs/research/bengio-safety-from-honesty-and-deliberus.md` (arXiv:2606.29657 read in full); funding positioning and the anti-coattail discipline in `.private/bengio_lawzero_landscape_2026_05_16.md`.\n\n\n### 160. \"AI in decision-making gives up human control\" / the disempowerment objection\n\n- **bytes**: 512 today → 313 proposed (saves 199)\n- **moves to**: `docs/research/the-disempowerment-objection.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"AI in decision-making gives up human control\" / the disempowerment objection** → `docs/research/the-disempowerment-objection.md` (2026-10-27): grant the worry, never the frame; his answer is more insight and therefore more real control, models held sceptically on small checkable tasks with measured error.\n\n- **warnings kept as clauses**: NOT YET WRITTEN\n\n### 114. \"In what order is a microtask checked?\" / evals / a machine judging a machine\n\n- **bytes**: 508 today → 282 proposed (saves 226)\n- **moves to**: `docs/research/the-evals-loop-and-our-machine-passes.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"In what order is a microtask checked?\" / evals / a machine judging a machine** → `docs/research/the-evals-loop-and-our-machine-passes.md` (Y9b–Y14 open): a reviewer's notes are read with `scripts/read_review_notes.py`; both batches are too small to prove one kind of fault.\n\n\n### 10. Convergence wager / typed residues / legibility floor / fourth identity\n\n- **bytes**: 479 today → 357 proposed (saves 122)\n- **moves to**: `docs/research/convergence-wager-typed-residues-synthesis.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Convergence wager / typed residues / legibility floor / fourth identity** → `docs/research/convergence-wager-typed-residues-synthesis.md` (entry point) + `docs/convergence.md` + `docs/vision.md` § The Founding Wager: convergence is the wager held fallibly, legibility the floor, and the identity is public epistemic infrastructure (settled Jul 2026).\n\n\n### 92. On sacred values, argument is the WRONG instrument.\n\n- **bytes**: 461 today → 461 proposed (saves 0)\n- **proposal**: unchanged\n\n### 2. Atomic-claim criteria (structural + recursive, NOT fixed rubric)\n\n- **bytes**: 447 today → 341 proposed (saves 106)\n- **moves to**: `docs/research/self-similar-decomposition-and-claim-ontology.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Atomic-claim criteria (structural + recursive, NOT fixed rubric)** → this CLAUDE.md § \"The Self-Similar Decomposition Principle\" + `docs/depth.md` + `docs/research/self-similar-decomposition-and-claim-ontology.md`: nothing is permanently atomic, and the good-enough rubric is deliberately deferred to dogfood pain (do not ask for it).\n\n\n### 12. Dogfood run 1 results + observed friction (F1–F15) + post-run insights (I1–I5: inversion principle, residue/crux/double-crux + hinge score, mint-time dedup carve, QEM additivity, confession principle) + architectural proposals\n\n- **bytes**: 433 today → 380 proposed (saves 53)\n- **moves to**: `docs/research/dogfood-run-1-friction-log.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Dogfood run 1 results + observed friction (F1–F15) + post-run insights (I1–I5: inversion principle, residue/crux/double-crux + hinge score, mint-time dedup carve, QEM additivity, confession principle) + architectural proposals** → `docs/research/dogfood-run-1-friction-log.md`: friction F1–F15, insights I1–I5, and the baseline (claim-rich, edge-poor at extraction).\n\n\n### 125. \"Does a session's trouble holding the whole corpus come from memory?\" / agent memory / what graph memory for agents says about the agent surfaces\n\n- **bytes**: 418 today → 354 proposed (saves 64)\n- **moves to**: `docs/research/what-agent-memory-research-says-about-agent-surfaces.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **\"Does a session's trouble holding the whole corpus come from memory?\" / agent memory / what graph memory for agents says about the agent surfaces** → `docs/research/what-agent-memory-research-says-about-agent-surfaces.md` (Arc AE, unruled): an agent that owns a memory mostly does not look in it, and a summary handed back without its source loses.\n\n\n### 1. Testing philosophy (substance-over-meta, let-it-break, real-material trials)\n\n- **bytes**: 398 today → 397 proposed (saves 1)\n- **moves to**: `docs/research/session10-empirical-collaboration-and-atomization-frontier.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Testing philosophy (substance-over-meta, let-it-break, real-material trials)** → this CLAUDE.md § \"Pipeline Testing: Substance Over Meta\" + `docs/research/session10-empirical-collaboration-and-atomization-frontier.md` (his testing philosophy verbatim): the pipeline is tested on substantive real material, since complexity is the medium of discovery and simple-vs-complex is the wrong axis.\n\n\n### 24. Positioning / \"what class is Deliberus\" (missing-layer taxonomy)\n\n- **bytes**: 349 today → 349 proposed (saves 0)\n- **proposal**: unchanged\n\n### 3. Primary atomicity pain pattern (already identified)\n\n- **bytes**: 309 today → 237 proposed (saves 72)\n- **moves to**: `docs/research/extraction-virtual-run.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Primary atomicity pain pattern (already identified)** → `docs/research/extraction-virtual-run.md` + `docs/research/first-pipeline-run-analysis.md`: mother-claim bundling, rooted in the scout pass paraphrasing where it should point.\n\n\n### 4. Improvement backlog for extraction pipeline (10 items)\n\n- **bytes**: 234 today → 223 proposed (saves 11)\n- **moves to**: `docs/research/first-pipeline-run-analysis.md` (end of the file, under *From the Topic Index*)\n- **proposed line**:\n\n  - **Improvement backlog for extraction pipeline (10 items)** → `docs/research/first-pipeline-run-analysis.md` + `docs/research/extraction-improvement-research.md`: pain-point-driven iteration, never a speculative rewrite.\n\n\n### 156. Standing epistemic threat model\n\n- **bytes**: 194 today → 194 proposed (saves 0)\n- **proposal**: unchanged\n\n### 8. His Deliberus words are typed everywhere\n\n- **bytes**: 136 today → 136 proposed (saves 0)\n- **proposal**: unchanged\n\n"}