{"path":"research/corpus-graph/readers/2026-10-11-pilot-what-does-not-fit-four-notes-as-a-haiku-reader-returned-it.md","content":"# What a Haiku 5.5 reader found that does not fit, in four notes (pilot) (2026-10-11)\n\nA read-only reader on Haiku 5.5 at high effort, started through the command-line tool with no settings\nloaded, with the tools to read and to search files and no shell. It is part of the pilot described in\n[../reading-for-every-kind-of-insight.md](../reading-for-every-kind-of-insight.md) § 7 and of the test in\n[../haiku-5-5-for-the-filling.md](../haiku-5-5-for-the-filling.md) § 6. It was given the brief in\n[../dissonance-reading-brief.md](../dissonance-reading-brief.md), everything the founder has settled in one file, and these 4 notes:\n\n- `docs/depth.md`\n- `docs/bridging.md`\n- `docs/worldview-lenses.md`\n- `docs/research/the-load-bearing-unsaid.md`\n\nIt took 43 steps and six minutes. Its report is below, unchanged. What was kept from it, after the quoted\nwords were checked by code and both passages of each finding were opened, is in `../found.jsonl` (lines\nwhose `where` names this file).\n\n---\n\n4 of 4 notes read in full; about 100 checkable statements listed (tallied while reading, not itemised).\n\n### F1\ntarget: docs/depth.md:112\ntarget_words: \"but it is inert until someone records the interpretation.\"\nolder: deliberus/support_semantics.py:42\nolder_words: \"by their weakest member (under the separability floor, parts of a decomposed whole are\"\nkind: state\nwhy: The note says the shipped necessary-tagging fix stays inert until someone records a join interpretation. Since the 2026-08-31 ruling, `subclaim_energy` passes `undecided_as=\"necessary\"`, so untagged decomposition children are weakest-part-governed by default and the fix is live with no recording. depth.md:120 records that flip, but line 112 was never updated.\nstands: other, by the code\nstrength: strong\n\n### F2\ntarget: docs/bridging.md:67\ntarget_words: \"Closing that gap changes the ranking on a shipped surface, which makes it a founder decision rather than a repair\"\nolder: TODO.md:1284\nolder_words: \"bridging feeds and the truth graph's bridging list re-rank (answers the open\"\nkind: overtaken\nwhy: The bridging note says closing the stored-strength gap in the feed is a founder decision still to be taken. On 2026-09-28 the founder ruled \"Yes\" to showing computed link strength, and the TODO entry says this answers the open bridging-feed question. The feed still ranks by the stored number, so only the build is pending.\nstands: other, by a ruling of 2026-09-28\nstrength: strong\n\n### F3\ntarget: docs/research/the-load-bearing-unsaid.md:69\ntarget_words: \"and a citation the corpus does not yet have\"\nolder: docs/research/philosophy-of-the-unwritten-step.md:114\nolder_words: \"Sellars and Brandom, who deny that such a step is a formal one with a premise missing;\"\nkind: overtaken\nwhy: The load-bearing note says the corpus does not yet have a Brandom citation for the scorekeeping idea. Philosophy-of-the-unwritten-step, dated 2026-10-07, cites Sellars and Brandom on material inference and commitment, and philosophy-of-evidence cites Making It Explicit. The claim was true when written on 2026-08-13 and has been overtaken without a mark.\nstands: other, by the later date (2026-10-07)\nstrength: medium\n\n### F4\ntarget: docs/worldview-lenses.md:9\ntarget_words: \"And once located, the disagreement is almost always much smaller than it felt.\"\nolder: docs/research/the-load-bearing-unsaid.md:142\nolder_words: \"the legal disagreement was smaller than its reputation, and that fact is about two texts, not about the conflict.\"\nkind: counter-case\nwhy: The worldview page says a located disagreement is almost always much smaller than it felt. The load-bearing note says run 6 shows smallness only within two texts, and that a presumed-shared premise, once it turns out not to be shared, can reveal an ordinary disagreement to be a deep one. A premise located as a hinge is the case where the page's generalisation fails.\nstands: other, by a measurement (run 6)\nstrength: medium\n\n### F5\ntarget: docs/worldview-lenses.md:92\ntarget_words: \"The lenses show the surface; the depth reveals the convergence.\"\nolder: CLAUDE.md:553\nolder_words: \"the bet can LOSE (hinge/parity/permissivism ceilings count against it)\"\nkind: tension\nwhy: The worldview page states as a present fact that depth reveals convergence beneath surface disagreement. The always-loaded convergence line calls convergence a wager that can lose to hinge and parity ceilings, and depth.md:134 calls it a conviction, not a proven fact. Two people who drill to a protected value and find no shared bedrock make the page's sentence false.\nstands: other, by nothing I found\nstrength: medium\n\n### F6\ntarget: docs/worldview-lenses.md:11\ntarget_words: \"It is the core interaction mode, and it has been part of the vision since before 2011.\"\nolder: CLAUDE.md:691\nolder_words: \"All are EXPLORATORY until explicitly decided. Deep convictions and intuitions about\"\nkind: state\nwhy: The worldview page calls the lens the core interaction mode, present tense and part of the vision since 2011. CLAUDE.md lists the worldview filter among things exploratory until explicitly decided, and TODO.md:2911 defers the lens as \"deliberately second\". No lens or worldview code exists under deliberus/, so the present-tense claim has no decision or build behind it.\nstands: other, by the code\nstrength: medium\n\n### F7\ntarget: docs/research/the-load-bearing-unsaid.md:47\ntarget_words: \"Pointing the implicit-premise pass at a sacred commitment and rendering it as a weighable premise\"\nolder: deliberus/extraction/implicit_premises.py:3\nolder_words: \"Surfaces unstated assumptions connecting explicit claims within each argument.\"\nkind: rule-and-practice\nwhy: The load-bearing note calls pointing the implicit-premise pass at a sacred commitment a category error. Pass 2c runs over every argument's claims, and I found no sacredness check in `detect_implicit_premises` or its extraction callers. So the pass as built does this to any sacred claim it receives, and the note does not say whether that is intended.\nstands: cannot tell, by nothing I found\nstrength: medium\n\n### F8\ntarget: docs/research/the-load-bearing-unsaid.md:65\ntarget_words: \"a pseudo-sacred commitment yields to decomposition, a hinge does not, and which one you have is discovered by trying\"\nolder: docs/research/the-load-bearing-unsaid.md:47\nolder_words: \"Pointing the implicit-premise pass at a sacred commitment and rendering it as a weighable premise\"\nkind: tension\nwhy: Line 65 says whether a commitment is a pseudo-sacred one or a hinge is found out only by trying decomposition on it. Line 47 says decomposing a sacred commitment is the mechanised form of the taboo trade-off, a category error. Take the ACLU's \"human life is inherently precious\" claim: to learn whether it is a hinge you must try, and trying is what line 47 forbids.\nstands: cannot tell, by nothing I found\nstrength: medium\n\n### F9\ntarget: docs/depth.md:79\ntarget_words: \"91% of verdicts survived every filing, the typical score did not move at all\"\nolder: docs/research/slicing-stability-first-run.md:27\nolder_words: \"under the flat filing, 86 of 87 mothers go mute (NO DATA / UNEXAMINED) rather than disagreeing\"\nkind: other (dropped qualifier)\nwhy: The depth note says 91% of verdicts survived every filing, including the never-decomposed flat filing. The first-run measurement counts only filings that produce a verdict (90.8% of 87 mothers), and under the flat filing 86 of 87 go mute. So \"every filing\" overstates what was shown.\nstands: other, by a measurement\nstrength: medium\n\n### F10\ntarget: docs/bridging.md:9\ntarget_words: \"validates that structured deliberation with typed epistemic acts significantly outperforms unstructured debate\"\nolder: docs/research/ai-augmented-cooperation-infrastructure.md:120\nolder_words: \"On routine tasks: DCI scores only 5.39 (underperforms baselines)\"\nkind: counter-case\nwhy: The bridging note states flatly that typed-act deliberation significantly outperforms unstructured debate. The AI-cooperation note records the significant gain only on non-routine tasks, and says that on routine tasks DCI underperforms baselines. The non-zero-sum note qualifies the claim to hidden-profile tasks, so the routine task is the case the unqualified sentence gets wrong.\nstands: other, by a measurement\nstrength: medium\n\nNoticed: load-bearing-unsaid.md:35 calls implicit-premise recall \"already this project's most important quality axis\", where dogfood-run-6 (line 127) only calls it the strongest case for that; depth.md:185 says \"every confirmed terminus\", using the sense Q32 retired; load-bearing-unsaid.md:59 says participants predict two thirds, where core-periphery-and-the-concept-layer.md:132 says they overestimate by about two thirds. None of the ten pairs is already recorded in found.jsonl for the same two passages.\n"}