{"path":"research/corpus-graph/readers/2026-10-10-decomposition-concepts-and-first-take-as-a-reader-returned-it.md","content":"# What a reader found to bear on seven notes on decomposition, concepts and the first take (2026-10-10)\n\nA read-only reader (Sonnet), one of those sent through the notes no reader had yet read, after the founder\nasked for the graph's coverage to be widened (README § 12). It was given the brief in\n[../reader_brief.md](../reader_brief.md) and these 7 notes:\n\n- `docs/research/self-similar-decomposition-and-claim-ontology.md`\n- `docs/research/codex-session-20260403.md`\n- `docs/research/semantic-disambiguation-and-concept-tracking.md`\n- `docs/research/execution-plan-clarification-first-roadmap.md`\n- `docs/research/claim-sameness-philosophical-readings.md`\n- `docs/research/disambiguation-ux-and-voice-first-contribution.md`\n- `docs/research/truth-graph-evidence-system.md`\n\nIt made 93 tool calls. Its report is below, unchanged. What was kept from it, after the quoted words were\nchecked by code and both passages of each finding were opened, is in `../found.jsonl` (lines whose `where`\nnames this file).\n\n---\n\n7 of the 7 notes read in full.\n\n### F1\ntarget: docs/research/disambiguation-ux-and-voice-first-contribution.md:33\ntarget_words: \"This connects to P14 (the interpassivity warning): the confirmation step must provoke engagement, not rubber-stamping.\"\nolder: docs/research/the-ratification-record-and-the-triage-feed.md:19\nolder_words: \"standing formulation *review always invited, never conscripted*. Friction designed to force\"\nkind: contradicts\nwhy: The record-rule note argues that \"the confirmation step must provoke engagement\" conscripts engagement. It says the real fault is a click recorded as a check. Its end-state, blessed 2026-08-31, has no ratify act: assent is inferred telemetry and dissent is a contribution. Someone building the confirm step should record \"accepted\" and earn \"examined\" from observed behaviour, not design friction to force a look. The correcting tap stays valuable as a contribution. Ledger rows R9 and R14 are still open.\nstrength: strong\n\n### F2\ntarget: docs/research/codex-session-20260403.md:1874\ntarget_words: \"I’ve rewritten the fix into the parser-safe format Kamal actually expects.\"\nolder: docs/research/session12-production-recovery-and-secrets-fix.md:15\nolder_words: \"The Codex session on April 3 (Session 11) deployed from MERIAN with a broken `.kamal/secrets` file.\"\nkind: later-correction\nwhy: Session 12 shows the env-first `${VAR:-grep}` rewrite pushed here as the fix was the bug. Kamal strips the `${`, which corrupted every container secret. deliberus.com then showed zero extractions with falkordb in error, which is the empty landing page the transcript's last message asks about. Session 12 reverted to the simple grep and calls session 11's report optimistic. Anyone reusing this transcript's fix should read the revert. Project CLAUDE.md now states the one safe pattern.\nstrength: strong\n\n### F3\ntarget: docs/research/self-similar-decomposition-and-claim-ontology.md:238\ntarget_words: \"**The requirement is declared in prose and enforced nowhere**, and it is the cleanest first constraint for the approved ontology layer\"\nolder: docs/research/the-squirrel-and-the-strength-calculus/strength-rulings-and-code-as-read-2026-10-08.md:181\nolder_words: \"takes `(s)-[r]->(c)` where `r.scheme` is non-empty, `type(r)` is NOT in ['REPORTS','CLASSIFIES_TERMINUS','ARGUES_REQUIREMENT','SIMILAR_TO','ARGUES_RELEVANCE']\"\nkind: later-correction\nwhy: The badge query now names its types and excludes REPORTS explicitly (TODO records it done 2026-09-17). Non-propagation no longer rests on an absent scheme. It is still a hand-written exclusion list guarded by a test that greps for a string (world-models-and-the-ontology-revival.md:77 calls this a missing \"Gate 2\"). `count_safe_summary` still has no caller. Whoever rules on the ontology layer should treat the first requirement as built but not yet declared once.\nstrength: strong\n\n### F4\ntarget: docs/research/execution-plan-clarification-first-roadmap.md:458\ntarget_words: \"decompose only when a meaningful burden of proof or ambiguity remains\"\nolder: docs/research/kartpaus-auto-enrichment-design.md:15\nolder_words: \"contest-gating survives as a *budget* heuristic for where to spend machine effort\"\nkind: contradicts\nwhy: The 2026-08-25 ruling wants all claims decomposed recursively and eagerly, landing as machine drafts shown only on contest. The note says contest-gating was never ratified and survives only as a spending heuristic. Ledger B2 rules the floor is separability (\"Both blessed\", 2026-08-31). So \"only when a burden remains\" is not the rule, and the machine's stop is the worth-asking action price. A builder of Phase 3 should not gate descent on need.\nstrength: strong\n\n### F5\ntarget: docs/research/self-similar-decomposition-and-claim-ontology.md:305\ntarget_words: \"Concept reuse at 87% is the saturated upstream layer appearing.\"\nolder: docs/research/the-reuse-flywheel-prior.md:168\nolder_words: \"But **only 4 concepts span more than one SOURCE (4%), and no concept in the graph spans\"\nkind: later-correction\nwhy: Measured 2026-08-29: about 88% of 90 concepts are reused by more than one claim, but only 4 span more than one source and none spans more than two. So the reuse figure is within-source and is not evidence that a shared upstream layer is appearing. Concept identity has the same matching problem as claim identity. Anyone scoring the prediction or the April mechanism should drop concept reuse as support. Caveat: measured on the old cosine-only linker.\nstrength: strong\n\n### F6\ntarget: docs/research/execution-plan-clarification-first-roadmap.md:41\ntarget_words: \"concept pages and APIs now surface emergent lifecycle states like underdefined, emerging, bifurcated, fragmented, and locally stable\"\nolder: docs/research/core-periphery-and-the-concept-layer.md:171\nolder_words: \"as a diagnosis. **The distribution is largely the detector's template**, not a measurement of how\"\nkind: contradicts\nwhy: Measured 2026-09-06: 49 of 90 live concepts read \"bifurcated\" because the contested-concept pass mints two senses per term. The lifecycle state reads the detector's template back, not how people use the word. Sense attribution is modelled but not stored. The plan's \"emergent\" states, and its criterion that stability is emergent, rest on that signal. A builder should not present them as a finding until \"bifurcated\" requires a lineage partition.\nstrength: strong\n\n### F7\ntarget: docs/research/codex-session-20260403.md:90\ntarget_words: \"recursive decomposition under strong pruning, via retrieval-first reuse of existing subtrees, scheme-bounded when\"\nolder: docs/research/session26-open-rulings.md:799\nolder_words: \"It may and should write deeper steps. But the microtasks in charge of doing so need to be quality-controlled\"\nkind: later-correction\nwhy: The founder agreed in April, but \"never generating premises, only finding real ones\" was Claude's sentence; his March wording was \"finding or creating\". It was corrected 2026-10-07 (ledger AD21; the-reuse-flywheel-prior.md:43). On 2026-10-10 he ruled that the machine may and should write deeper steps, with quality-controlled small tasks (AG6). So the \"fear correct for generation\" split is not his line. The live question is the quality control and depth limit. self-similar-decomposition-and-claim-ontology.md:282 repeats the April synthesis.\nstrength: strong\n\n### F8\ntarget: docs/research/self-similar-decomposition-and-claim-ontology.md:216\ntarget_words: \"The split wants at least `cited_support` beside `reported_opposition`, and the honest way to set it is to look at the 94 before choosing\"\nolder: docs/research/the-remaining-case-for-the-stance-guard.md:45\nolder_words: \"one enum on the extracted claim (*asserts · rejects · reports*), the same on the `REPORTS` edge, and the `third_party` attribution role split accordingly.\"\nkind: already-answers\nwhy: The 2026-09-13 note cites this same first run and rules a stance field (asserts, rejects, reports). The field goes on the claim and on the REPORTS edge, and the `third_party` role is split accordingly. `stance.py` is demoted to a batch check. So the vocabulary to build is stance, not a guessed `cited_support`/`reported_opposition` pair. Session 35's count had the field on zero claims, so it is ruled but not built.\nstrength: strong\n\n### F9\ntarget: docs/research/truth-graph-evidence-system.md:189\ntarget_words: \"Default public truth-query retrieval and public feed surfaces suppress `raw`, `draft`, and `superseded` material.\"\nolder: docs/research/dogfood-run-4-agent-surface.md:57\nolder_words: \"The only shared component unique to the feed is `public_claim_filter()` from `deliberus/graph/maturity.py`.\"\nkind: later-correction\nwhy: This guardrail shipped 2026-04-08, and its filter was used by the feed alone. It emitted a FalkorDB-invalid `NOT IN [list]`, so every public feed mode returned 500 for about four months. The tests mocked the query, so nothing failed. The fix was deployed 2026-08-13 (TODO.md:2931). Anyone building promotion by maturity should know the flag held only where a surface remembered the filter. Ledger O14 and Z7 now state that rule.\nstrength: strong\n\n### F10\ntarget: docs/research/semantic-disambiguation-and-concept-tracking.md:72\ntarget_words: \"Deliberus dissolving a false disagreement through structural clarity.\"\nolder: docs/research/prime-real-estate-words.md:292\nolder_words: \"third outcome beside *it was a confusion* and *they disagree about the world*, and today the graph\"\nkind: narrows\nwhy: Ruled 2026-10-07 (ledger AC3): where two people want one word, showing the senses can leave \"what divides them is who gets the word\". That is a third outcome beside a confusion and a real disagreement, and today the graph would file it with confusions. So the aha moment holds for words nobody wants. On a wanted word the graph should keep both senses with their holders' names and report the contest, not report a dissolved disagreement.\nstrength: strong\n\n### F11\ntarget: docs/research/execution-plan-clarification-first-roadmap.md:367\ntarget_words: \"that can treat very high-confidence near-duplicates as one operative cluster by default without destroying provenance or reversibility\"\nolder: docs/research/soft-canonical-clustering-and-reversible-merge-semantics.md:329\nolder_words: \"**anyone types anything; careful auto-merge afterward; anyone (author or passerby) can split when truly warranted.**\"\nkind: already-answers\nwhy: The 2026-08-26 ruling settles this deliverable for claims. Auto-merge is allowed afterward, under four conditions: originals never destroyed; merges visible where the claim is read; split an accountable recorded act; author notified. An evidence gate precedes switch-on: a few hundred reviewed pairs, seeded wrongs, a blind subset. The bar rises for value claims. Concept senses are not covered (ledger P12). Phase 2 should build to these conditions.\nstrength: strong\n\n### F12\ntarget: docs/research/disambiguation-ux-and-voice-first-contribution.md:106\ntarget_words: \"two people who disagree on conclusions but share a definition are a bridging candidate.\"\nolder: docs/research/core-periphery-and-the-concept-layer.md:174\nolder_words: \"**Sense attribution is modelled but not stored.** `ConceptSense.attributed_to` exists in the Pydantic\"\nkind: supplies-a-missing-piece\nwhy: Sharing a definition is a shared core. The founding 2013 fact is that people give the same definition and still draw opposite conclusions. The data the reveal needs is also not stored: only `r.sense` is written on USES_CONCEPT, and nothing in the lifecycle rules reads which lineage holds which sense. The \"enemy shares your vocabulary\" step cannot be computed yet. Treat it as a design target, not a bridging signal.\nstrength: strong\n\n### F13\ntarget: docs/research/claim-sameness-philosophical-readings.md:166\ntarget_words: \"Each lives at its own layer. An embedding suggestion is not a weaker version of a human commitment\"\nolder: docs/research/what-a-record-linkage-project-teaches-deliberus.md:171\nolder_words: \"For us: the similarity search, the sameness check and a second family of model all read the same\"\nkind: narrows\nwhy: Embedding, model check and second model family all read the same two sentences. That is one field read three times, and agreement among them is counted twice. Record linkage needs a second axis: different kinds of evidence, such as place in the source's argument, asserter, concept link and attack links. Counted 2026-10-06, a further kind exists at both ends for 145 of 147 live pairs. Layering by \"specificity\" misses the independence question (ledger Z2, open).\nstrength: medium\n\n### F14\ntarget: docs/research/codex-session-20260403.md:30\ntarget_words: \"all those topic/object-level conclusions will be an increasingly saturated level in the graph consisting of shared\"\nolder: docs/research/the-reuse-flywheel-prior.md:189\nolder_words: \"The flywheel's shape as measured on the OLD linker**: per-source cross-source connection yield does not rise with\"\nkind: narrows\nwhy: Measured on 31 sources, cross-source connection yield does not rise with corpus size. It tracks whether an adversarial counterpart is already present (0.65 and 0.68 edges per claim for the death-penalty pair, against 0.02 for a lone piece). The flywheel is \"paired, not cumulative\", and the note says it runs at the rate identity work improves. The saturating upstream layer is not yet visible. Caveat: it was measured on the old cosine-only linker.\nstrength: medium\n\n### F15\ntarget: docs/research/semantic-disambiguation-and-concept-tracking.md:70\ntarget_words: \"Concept senses that look highly similar may be treated as belonging to one provisional operative cluster for retrieval, feeds, and display\"\nolder: docs/research/sameness-merge-and-split-across-fields.md:42\nolder_words: \"**Concepts want splitting, with careful merging.** The wild produces *false unities*: one word covering incompatible senses.\"\nkind: narrows\nwhy: The literature note finds the default error runs opposite ways. Claims arrive as duplicates and want merging with careful splitting. Concepts arrive as false unities and want splitting with careful merging. A false sense merge corrupts every claim using the term. Extending the claim pattern to senses, even as soft clusters, therefore starts from the wrong default. Ledger P12 records that the corpus leans the other way for concepts and that the question was never put to the founder.\nstrength: medium\n\n### F16\ntarget: docs/research/disambiguation-ux-and-voice-first-contribution.md:31\ntarget_words: \"they correct → **the correction IS the disambiguation.** The pipeline creates the moment naturally.\"\nolder: docs/research/what-human-judgment-is-for.md:168\nolder_words: \"A claim about *what I meant when I wrote that* is a report about a past mental state, which sits on the fallible side;\"\nkind: narrows\nwhy: Act three was corrected 2026-08-27 on founder challenge (ledger C4, ruled). A present refusal to endorse a rendering is unchallengeable. A claim about what one meant earlier is a self-interpretation with no elevated credence. So the tapped sense should be stored as the person's attributed definitional claim, open to attack. \"The correction is the disambiguation\" makes it a contribution, not the settled meaning. Only the refusal of a rendering stands unchallenged.\nstrength: medium\n\n### F17\ntarget: docs/research/self-similar-decomposition-and-claim-ontology.md:32\ntarget_words: \"The CQs point exactly at the seams where the mother claim should split.\"\nolder: docs/research/a-cut-through-a-concept-is-an-argument.md:12\nolder_words: \"is wrong* does not entail *cruel*, since a dentist satisfies both, so the pieces fail to\"\nkind: narrows\nwhy: A canon line approved 2026-09-21 says a cut through a concept is an argument, not a split. The pieces are premises, not constituents, and Walton's argument from verbal classification supplies the critical questions. The Smith decomposition has this shape, and its third part (\"publication indicates expertise\") is itself the classification premise. A builder should ask whether the \"credible expert\" parts are constituents under a necessary ceiling or premises on a support link.\nstrength: medium\n\n### F18\ntarget: docs/research/truth-graph-evidence-system.md:293\ntarget_words: \"The truth graph is not a separate graph. It is a query-time computation over the existing argument graph plus QBAF strengths.\"\nolder: docs/research/what-counts-as-evidence/corpus-map-code.md:502\nolder_words: \"it passes no subclaims, so the decomposition channel and the necessary cap are\"\nkind: later-correction\nwhy: The code audit found `/query` and the synthesis context took strengths from a second implementation, `compute_subgraph_badges`, not the canonical badge. It did not exclude REPORTS, SIMILAR_TO or superseded edges. It passed no subclaims, so there was no mother-claim flow and no necessary cap. It also read question edges in one direction only. TODO records it now delegates to `compute_badge_recursive`. Strength claims the first-take page made before that came from the divergent path.\nstrength: medium\n\nNoticed: One more pair was left out for the cap: claim-sameness-philosophical-readings.md:117 proposes clustering by Wilber's (quadrant, altitude, line) tuple. Ledger F9 rules taxonomy-as-schema \"no\" and what-belongs-in-the-ontology.md:192 says import only where each cell yields a move. Separately, ledger R9, R14 and J10–J12 still treat the record-rule as open. Only its §4b end-state is blessed, so F1's contradiction rests on that end-state.\n"}