{"path":"research/what-a-rule-rests-on/older-corpus-bearings-likeness-shapes-and-bottoms-2026-10-08.md","content":"# What the older corpus holds on likeness, shapes of step and resting places, set against the thirteenth walkthrough (2026-10-08)\n\nOne of three read-only readers (Sonnet) sent late on 2026-10-08, after the founder asked: *\"Is there\nany part of the corpus that bears on any aspect or even any single sentence of the latest audio\nwalkthrough that you haven't fully grasped or related it to?\"* This reader took parts two, four and\nten of the walkthrough and its closing (its header also names part nine). The leads it was given\nwere Claude's guesses, marked as such. Its report is below, unchanged. Claude's reading of the three\nreports is the note's § 14 ([../what-a-rule-rests-on.md](../what-a-rule-rests-on.md)).\n\nIn the report, \"the four notes\" are philosophy-of-the-unwritten-step.md, writing-out-the-unwritten.md,\nthe-squirrel-and-the-strength-calculus.md and what-a-rule-rests-on.md. A bare number in the third\ncolumn is a line of the kept walkthrough text.\n\n---\n\nPaths are under docs/research/ unless marked. W = the walkthrough (docs/listening/2026-10-08-eight-proposals-pondered-again.md). Read in full: W parts 2, 4, 9, 10, closing; the four notes, nearly whole; each named older section. Your leads were mostly right. Where they were not, I say so.\n\n| Lead | Corpus passage | W sentence | Already related? | What it would change |\n|---|---|---|---|---|\n| L1 (right) | scheme-set-exhaustiveness.md:52 \"Schemes represent only the prototypical matching between semantic relations and logical rules\"; :119 closed layer holds \"reasoning direction (deductive/abductive/inductive)\"; :113 a field that cannot say none \"silently deform[s] the taxonomy\" | 215 \"says which of four ways it walks\" | **No.** Absent from W, AD23 and the four notes. The note's own helper summarised it (what-a-rule-rests-on/corpus-as-read…:35) and it was dropped. | Your four ways are mostly Macagno and Walton's reasoning-type factor, rediscovered. Analogy is a *relation* there (Kienpointner files it \"neither\", beside authority), so your list mixes two axes. The four cover 443 of 780 kind-labelled links (what-the-count-printed.txt). About 335 (expert opinion, consequences, practical reasoning) fit none. AD23 has no \"none of these\". The trial already forced every link into shape one (philosophy-of-the-unwritten-step.md:230). 3→4 in a day is the corpus's three-runs law. |\n| L2 (right, partly) | extraction/schemes.py:352,359 \"not an edge case the definition fails to handle\"; \"does not fall under any recognized exception\"; deliberus/weighing.py:141,153 \"survive re-description\", \"set of affected parties changes\" | 293 \"Would it still hold, if it were a rat?\"; 219 \"Is this case typical of its kind?\" | **Partly.** The note lists the precedent, example and analogy questions (what-a-rule-rests-on.md:102-111). W cites only precedent_similar (333). | Variation questions ship un-parameterised (rule_exceptions, classification_edge, relevant_differences). \"Typical of its kind\" is example_representative, not new. A whole \"would it shift if X changed\" family ships for weighings. Build AD25 as one more such family with polarity claims. Only the term-by-term form and filing answers as cases are new. |\n| L3 (right) | why-representations-converge…:46 \"the average of a two-humped distribution is the valley between the humps\"; :193 penguin match \"about 12%\" on *similarity* triads; contingency-and-convergence…:178 \"The average of paths is no path.\" | 539 \"a sense of what is like what\"; 667 \"will favour the comparisons everyone already makes\" | **No.** The note reaches only the concept layer's periphery (:345). | The corpus has a measured answer by level. Shared: perception, the core of concepts, which items are neighbours. Not shared: the periphery, which is where similarity judgements live; people also assume two thirds agree. A machine's likeness is the mean of that spread, nobody's, and identical across models (copy number, not a second opinion). So \"favours what everyone makes\" understates: it can produce a comparison *no one* makes. A second model family does not check it. |\n| L4 (right, one fix) | what-belongs-in-the-ontology.md:226 names give \"a position and a direction… never an end state\"; CLAUDE.md:519 \"read, never set\"; horizon.py header; decomposing-value.md:85-88 \"a gap between knowing and being moved stays open\" | 597 \"the map says what kind of rest it is\"; 601 the list | **Partly.** Philosophy note §4 says \"as a horizon does\" and §12 declines a fixed list. W and AD29 do not. | See contradictions. \"A way of being moved\" is not a rival to \"a sentence\": decomposing-value already has it as the phenomenal floor plus the knowing/moved gap. It also lists semantic primes (:31) as a lexical shared bottom. decomposing-value §3 has Levels 0 to 5, six levels, not five. |\n| L5 (right) | philosophical-foundations-stress-test.md:231 \"PI 217's bedrock, where the spade turns and one acts without reasons\"; :268 \"A fifth residue type: hinge commitment / bedrock practice\"; :246 Fogelin | 581 \"It came to rest on a case.\" | **Partly.** Hinge and \"judgment and training\" appear (philosophy note :167,195). PI 217 and Fogelin never do. | The corpus says a rule-chain ends in *practice*; a case is how practice is trained. W's list (601) drops the hinge that the note's §4 had and adds \"a case\". Fogelin's abortion example (agree on the biology, still split) cuts against 557 \"shown the cases they already sort the same way\". |\n| L6 (right, partly) | depth.md:126 \"good explanations are hard to vary while still doing their explanatory job\"; science-reasoning-infrastructure.md:29 \"Popperian and Deutschian without being doctrinal\" | 293, 339 | **No.** No Deutsch or Popper in W or any of the four notes. | Same device, different object. Deutsch varies the explanation's details. You vary the case. Both separate load-bearing from idle detail. Your test lacks \"while still doing its job\": breadth by variation rewards a vacuous rule. The corpus's stated lineage also holds that rules are conjectured and cases only test them. The induction shape and Mill's register are never reconciled with that. Your part six (Lakatos) is the Popperian half. |\n| L7 (right) | understanding-after-the-proof-machines.md:42,51 \"born an unsolved exposition problem too\"; \"Definitions sit in the middle\"; lean-deliberus-analogies.md:119 a bad `@[simp]` lemma \"changes the behavior of every future proof\" | 569 \"any step can be written out, the moment somebody asks\" | **Partly.** Philosophy note :175-185 has the de Bruijn factor but cites neither document. | On-request answers availability, not exposition. Your writer made 3.62 added sentences where 1.21 sufficed: a wall of text. Sanderson's order rule backs your part three (case first). AD22's order (claim, tie, rule) runs the other way. A rule reused across links acts like a simp lemma, so one bad machine rule multiplies (the-residual-error-taxonomy.md class 4). The old Lean table (:171) lists \"relevant to conclusion\" as kernel-checkable, which the new work overrules unannounced. |\n| L8 (right) | islands-of-coherence.md:141 \"organise and name, never justify\"; assembly-theory…:106 \"match structurally, never by identity\"; convictions-across-scales.md:20 mechanism identity | 153 \"Planks in one bridge are all needed… the weakest plank decides\"; 435 \"We wrote that rule twice\" | **No.** | The bridge name carries the contested choice (required parts vs separate supports; squirrel note §5.1). The picture makes one answer sound given. The counter-case clause exists three times: counter_analogy and counter_example already ship (schemes.py:625). W's part-ten daemon omits recurrence, substrate, and contingency §5.5's \"homology or convergence?\" question. |\n| L9 (right, new angle) | retrieval-instruments-beyond-cosine.md:91-97 Yu: linked vs convergent decided by \"scheme instance\"; candidate_generation.py:162 signature is (direction, edge type, scheme) | 367 chain/cable; 671 \"One small finder\" | **No** for Yu. **Partly** for the finder (note :124, not W). | Yu's criterion *is* your chain/cable split, proposed 17 August, with a proxy (same target, same scheme type). Edges store no instance id; AD26's minimal sets could supply one. The finder reads stored kinds, and 871 machine-drawn links have none. It finds kinship of argument role, not \"a case like this one\". |\n| L10 (right, partly) | the-analogy-daemon.md:157-163 compare human-seeded vs daemon-seeded survival under attack; curiosity-as-connection…:48 \"how people put concepts together is really, really different\" | 667 \"The rare one still comes from a person.\" | **Yes** for the limit (note §8 limit 2 cites §7). **No** for the falsifier, the \"could not develop\" channel, or the dancer. | Add the falsifier to S6: it turns an assertion into a test. A dancer's analogy reshapes *both* structures; W's \"carried over\" runs one way. |\n| L11 (right) | decomposition-axes.md:65,68 degree and scope axes \"test not taught\"; :98 contest-gating \"overreaches\" | 293 | **No.** Cited once, for a founder record (writing-out :89). | A rule's breadth *is* the scope and degree axes. AD25's variation questions are those untaught tests. \"Vehicle\" is a degree/frame case; \"dead squirrels smell\" has no quantifier status. §4's warning against contest-gating bears on AD22's \"stands until somebody asks\". |\n| L12 (answer: no) | keeper.py:129 \"OVERLAPS is judged by EVIDENCE, not by topic\"; :71 unrelated → stop; auto_connect.py:181,511; ledger T5 | 647 \"It says what they share.\" | **No.** | There is no key for alike-but-not-same. A structurally kindred pair with no evidential bearing is `unrelated` and stops before any scheme runs. The one built like-case finder feeds a gate that drops its finds. `overlaps` is already read three ways (T5). The daemon needs its own route, or a key that names the respect. |\n| Extra X1 | convictions-across-scales.md:231 \"analogy-expressing MOTHER claim… decomposing into per-scale instance subclaims\" | 605 \"not one case filed under a rule\" | **No.** | A rule with its cases is representable today, and the 84 `example` links already file cases under general claims. Reuse that. |\n| X2 | claim-sameness-philosophical-readings.md:154,179 sameness \"indexed by scheme\"; \"elegant reconciliations… premature synthesis\" | 105 \"about one thing. Cases.\"; 275 lens | **No.** | Likeness-relative-to-respect and lens-as-equivalences were hypothesised in April, with six overreach risks. Your opening claim has the shape it warned about. |\n| X3 | perceiving-possibility…:102 cites Hofstadter and Sander's 2013 book | 659 \"He never argues that rules come from cases.\" | **Partly.** The note admits the book is unread (§12). | The \"mistake owned\" may be an overcorrection from one 2001 essay. Say \"in the essay\". |\n| X4 | philosophy note :581 \"the law report\" as the practice built round this problem | 673 \"something nobody has had\" | **No.** | Case law is a public record of comparisons that held. Your part eight uses Levi. |\n\n**Three bearings that matter most**\n1. **L1 and L12 together: what AD23 is and what it feeds.** The shape field is a coarsening of the existing scheme field. It lacks the corpus's import rule (does-not-fit shipped *in the same commit*, ontology §6b:192; ledger F5, Q19, both open). It also lacks the generic fallback that already exists (`warrant_acceptable`, cq_fallbacks.py:34). \"Four dialects\" overstates coverage: 57% of kind-labelled links.\n2. **L4 and L5: the \"kind of stop\".** It collides with a ruled canon line (below). The better answer is already in the corpus: practice or hinge, not case.\n3. **L3: the third thing to carry.** You marked \"a sense of what is like what\" as your question and no finding. The corpus has findings, and they change the question. Its frame for a shared bottom is *source* (descent versus world-constraint, convictions-across-scales.md:50-66), not form. An inborn sense (Quine) would be shared by descent, which is copy number, not evidence it tracks anything. Similarity judgements are also the measured divergent layer.\n\n**Contradictions**\n- **AD29 and W 597/601 against \"read, never set\".** CLAUDE.md:519 (RULED 2026-09-15): \"A kind is a reading… never an end.\" AD29 says each chain \"ends in a named kind of stop\". The shipped horizon kinds (terminus.py:80) already include `does_not_fit`. W's five kinds are a parallel vocabulary, unmapped to them. The founder's verbatim position (taxonomy-gaps-and-the-closed-enum.md:83) is that recursive-why is *untyped*: \"the acid, never the taxonomy\". Your note's own §12 lists \"a fixed list of rule kinds as a schema\" as considered and not proposed.\n- **Part ten \"never hands you a number\" (651) and S6.** The founder commissioned a prior pass: \"note what prior it would set\" (the-analogy-daemon.md:16,44). Nobody says it is withdrawn. The evidence probably favours dropping it, but that needs saying.\n- **AD23 against Q19.** \"A displayed type is read as a weight.\" AD23's shape *selects the arithmetic*, which is stronger than display.\n- **W 227 \"Our checker reads one way of the four.\"** True only of the blind reader. The question machinery hangs on all four stored kinds.\n- **W 367 \"the code has a setting for that\".** The necessary, corroborative and alternative settings apply to parts of a claim. Several reasons for one claim simply add (squirrel note :56).\n\nUnverified: all counts come from the committed printout. I ran nothing against the live graph.\n"}