{"path":"research/corpus-graph/readers/2026-10-27-batch-2-the-number-folds-and-checks-as-a-reader-returned-it.md","content":"# What a reader found to bear on eight notes on the number on a claim, the repertoire, checks on machines and people, folds, the shared doubt and the map spec (2026-10-27)\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/checks-on-machines-and-people.md`\n- `docs/research/world-completeness-and-its-relatives.md`\n- `docs/research/the-number-in-the-world-and-the-repertoire.md`\n- `docs/research/what-the-number-on-a-claim-means.md`\n- `docs/research/folding-joining-and-zoom.md`\n- `docs/research/where-a-shared-doubt-sits.md`\n- `docs/research/folded-routes-and-shared-premises.md`\n\nIt made 86 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\n8 of 8 notes read in full.\n\n### F1\ntarget: docs/research/what-the-number-on-a-claim-means.md:178\ntarget_words: \"*Two independent arguments for one claim*: reads as the better one; a joint claim lifts it.\"\nolder: docs/research/how-independent-reasons-add.md:29\nolder_words: \"their odds multiply: 9:1 × 9:1 = 81:1, a strength of 0.988.\"\nkind: later-correction\nwhy: Q46 (ruled 2026-10-22, built 2026-10-23) says arguments the keeper reads as independent add in log-odds, so two 0.9 reasons read 0.988 with no written joint claim. QEM's squash is retired on the reasons channel. Q37 is \"settled through Q38\", and the row leaves open whether the lift is stored as a claim. Case 1 and § 5 items 1-2 (maximum, lift as a claim) are therefore replaced by an independence reading plus log-odds addition, and \"Q37 proposed, unruled\" is out of date.\nstrength: strong\n\n### F2\ntarget: docs/research/the-number-in-the-world-and-the-repertoire.md:59\ntarget_words: \"today: relevance × (standing − 0.5), which is Jeffrey's rule with every start at 0.5\"\nolder: docs/research/session26-open-rulings.md:465\nolder_words: \"A link's relevance is the difference the reason makes, P(C | A) − P(C | not A), everywhere\"\nkind: later-correction\nwhy: Q50 (ruled 2026-10-26) found the shipped code measured the pull from the claim's prior, not the supporter's own prior. It equals Jeffrey's rule only while both priors are 0.5, and with a claim prior of 0.2 the code gave 0.76 against Jeffrey's 0.44. It was rebuilt to measure from the supporter's prior. Row 4's \"honest starts change one symbol\" understates this. Q46's row records this note's row 4 as the demonstrated mistake.\nstrength: strong\n\n### F3\ntarget: docs/research/checks-on-machines-and-people.md:108\ntarget_words: \"one irrelevant part caps any claim at 0.75, and at 0.5 once the part is\"\nolder: docs/research/session26-open-rulings.md:1388\nolder_words: \"A part link records a cut and carries no strength: nothing a part says reaches its whole, neither the weakest-part cap,\"\nkind: later-correction\nwhy: BD16 (ruled 2026-10-25, built; `PARTS_CARRY_STRENGTH = False`) removes the weakest-part cap and the argued requirement. The sabotage price in § 2.1 (0.75 cap for one contribution), § 4.2 item 1 and the 0.493 effect of § 11.4 no longer exist as stated. O13 is narrowed accordingly. Someone ruling O13 or W5 from this note would be defending against an arithmetic that has been switched off; what remains of the part-test question is the sorting of misfiled reasons (AX12).\nstrength: strong\n\n### F4\ntarget: docs/research/the-number-in-the-world-and-the-repertoire.md:295\ntarget_words: \"down; the badge to show weight beside the number (Q41).\"\nolder: docs/research/session26-open-rulings.md:456\nolder_words: \"the weight beside the strength RULED 2026-10-22**, founder: *\\\"All ruled!\\\"*, to *\\\"The weight shown beside the number.\\\"*\"\nkind: already-answers\nwhy: § 6 lists \"the badge to show weight beside the number\" under \"proposed, unruled\". The Q41 row records that half ruled on 2026-10-22 (\"All ruled!\"), as the first build, needing no prior. Only R1 as the target is still open. A builder reading this note would treat as a question what he has already settled.\nstrength: strong\n\n### F5\ntarget: docs/research/the-number-in-the-world-and-the-repertoire.md:279\ntarget_words: \"The test that settles how much goes in the pack**, proposed by the other session and sharpened\"\nolder: docs/research/session26-open-rulings.md:1001\nolder_words: \"RULED 2026-10-19 · pack and search, no replay · unbuilt\"\nkind: already-answers\nwhy: § 5b says the replay of five decisions settles the pack size and \"the decision is his\". AK4 is ruled 2026-10-19 as \"pack and search, no replay\", and AK3 (reply-time search) is ruled the same day. The ruling skips the replay test this section proposes as the settler. Anyone planning the replay would build something he has already declined; the open work is building the pack and the search.\nstrength: strong\n\n### F6\ntarget: docs/research/what-the-number-on-a-claim-means.md:28\ntarget_words: \"**The number** (`strength`, the badge): a value in [0, 1] computed for a claim from the links into\"\nolder: docs/research/session26-open-rulings.md:1380\nolder_words: \"The number a claim carries is its posterior: its prior moved by the examined reasons, with the weight of what was examined beside it\"\nkind: later-correction\nwhy: BD8 (ruled 2026-10-25, built in code) names a claim's number its posterior and reserves \"strength\" for a link. It says that where prior, rivals and independence are recorded it is a probability's posterior exactly. The note's whole vocabulary (\"the number\", `strength`, \"standing\", the four readings) predates this. It also narrows the R2-versus-R1 choice: R1 now has a name and a code key, and \"probability\" was rejected as a word because it promises calibration.\nstrength: strong\n\n### F7\ntarget: docs/research/what-the-number-on-a-claim-means.md:37\ntarget_words: \"**Relevance**: a reading in [0, 1] of how much a supporting claim, granted true, bears on the\"\nolder: docs/research/session26-open-rulings.md:465\nolder_words: \"The two agree only while both priors are 0.5.\"\nkind: later-correction\nwhy: Q50 (ruled 2026-10-26) defines relevance as the difference the reason makes, P(C|A) − P(C|¬A), bounded by both priors. The \"gain, granted true\" reading this definition gives, and the contribution formula of § 3, coincide with it only while all priors are 0.5. Q50 is the \"highest priority, settling central theory\" ruling, and the note's § 1 and § 3 definitions are built on the superseded reading.\nstrength: strong\n\n### F8\ntarget: docs/research/the-number-in-the-world-and-the-repertoire.md:60\ntarget_words: \"the live argument links hold no two-cycles (measured 2026-09-28), so a single pass over a directed acyclic graph is exact today\"\nolder: docs/research/folded-routes-and-shared-premises.md:223\nolder_words: \"Six claims sit in a two-link support loop. The map today shows two links\"\nkind: contradicts\nwhy: A later count on graph `deliberus_extraction` (2026-10-23) finds six claims in a two-link support loop. `strength-layer-audit.md:31` (2026-08-17) had also measured 3 two-cycles and 1 three-cycle. Row 5's \"single pass is exact\" rests on the 09-28 reading, which is not the only measurement. It needs a rerun on the live graph before \"adequate until cycles appear\" is relied on. Caveat: the two counts may cover different graphs.\nstrength: medium\n\n### F9\ntarget: docs/research/world-completeness-and-its-relatives.md:43\ntarget_words: \"A null result is informative only if the record shows what was looked for\"\nolder: docs/research/what-counts-as-evidence/philosophy-of-evidence.md:297\nolder_words: \"which depends on the **search's sensitivity** P(E found | H true).\"\nkind: supplies-a-missing-piece\nwhy: The note says the search is the only recordable thing but gives no measure of how much a null result is worth. This passage gives it: absence is evidence of absence with likelihood ratio P(¬E|¬H)/P(¬E|H), so its weight depends on how likely the search was to find E if the claim were true. Sober (2009) adds that it is typically very weak for an imperfect record. That is the quantity behind the `ignorance` scheme's second question and a candidate form for Q45's count.\nstrength: medium\n\n### F10\ntarget: docs/research/folding-joining-and-zoom.md:238\ntarget_words: \"code ranks what a fold leaves out by what each claim passes on to the focus along its route, always\"\nolder: docs/research/saturation-and-the-long-tail.md:79\nolder_words: \"every summary-shaped surface Deliberus itself serves is a tail-killing layer by the same\"\nkind: same-mechanism-elsewhere\nwhy: A fold that chooses what to hide by what a claim passes on to the focus is a summary by selection. § 3b records, measured twice, that summaries and key-point systems drop the low-frequency claims where the value lies, and § 3b point 4 says the map is such a surface. The proposed ranking favours the head. The note keeps only the weakest step and the one rarest-claim place; the tail dial (rare claims reaching the answer) is not applied to folds.\nstrength: medium\n\n### F11\ntarget: docs/research/where-a-shared-doubt-sits.md:365\ntarget_words: \"of the polls is the traders', never the reporter's who said *the market gives 70 %*.\"\nolder: docs/research/what-counts-as-evidence/law-intelligence-history-journalism-forensics.md:759\nolder_words: \"impression of many corroborating reports where in fact there were very few sources**.\"\nkind: same-mechanism-elsewhere\nwhy: The Iraq WMD finding (§ 3.5) has the same shape as the markets-and-panel case. Many briefs counted as corroboration when the sources were few, and a relay that passes reports on without owning their credibility. It lists four failures: volume counted as sources, \"multiple sources\" without independence, a retraction not propagated, relays. It cites a measurement (Bara 2026) that naive aggregation as report multiplicity rises 1 to 32 drops posterior coverage 0.940 to 0.263. That confirms the \"adding is how every overclaim happened\" direction and the AX3 attribution rule.\nstrength: medium\n\n### F12\ntarget: specs/map-view/design.md:595\ntarget_words: \"Link strength: yes. What a claim hangs on: only down a claim's **parts**\"\nolder: docs/research/what-if-and-lenses/outside-reading-formal.md:130\nolder_words: \"The graph version passes the override into the same recursive calculation the badge uses\"\nkind: later-correction\nwhy: The hinge since 2026-09-28 runs on the badge itself, over a region including supporters and attackers (hinge.py docstring), so \"what a claim hangs on\" is computed for the supporters and challengers that fill the map. The 22 October reading confirms this and adds a limit: many parallel independent supporters each show a low hinge (line 171 there), so the \"glow\" emphasis of § 9 would under-mark them. BD16 also leaves the parts route carrying no strength.\nstrength: medium\n\n### F13\ntarget: specs/map-view/design.md:729\ntarget_words: \"every checked reason failed. Drawing the weight beside the fill is the step the paragraph below asks\"\nolder: docs/research/session26-open-rulings.md:456\nolder_words: \"the weight beside the strength RULED 2026-10-22**, founder: *\\\"All ruled!\\\"*, to *\\\"The weight shown beside the number.\\\"*\"\nkind: already-answers\nwhy: § 10a says drawing the weight beside the fill is the next step and that its form is unruled, and that \"until the meaning is ruled, a fill is the case so far\". Q41's weight half is ruled (2026-10-22) and BD8 (2026-10-25) names the number a posterior with the weight beside it. What stays open is how it is drawn (AR14), not whether, and the fill now has a ruled meaning. The map's sentences about it can be made true accordingly.\nstrength: medium\n\n### F14\ntarget: docs/research/checks-on-machines-and-people.md:633\ntarget_words: \"For parts, his reading of the 39 comes first.\"\nolder: docs/research/the-parts-review.md:142\nolder_words: \"the census reliably detects *not a part*, and says little about *what instead*.\"\nkind: same-idea-earlier\nwhy: On 2026-09-27 a census of 197 decomposition edges already found person-written parts (16 of the 41 left filed as parts) that were misfiled, and a review page existed to put the judgement to a person. It concluded the machine detects \"not a part\" reliably but not \"what instead\". That is exactly the four-way sort (part, reason, need, debris) that § 11.6 asks him for. The 39-part reading is the second step of a started review, not the first human labelling, and the review page could host it.\nstrength: medium\n\n### F15\ntarget: docs/research/checks-on-machines-and-people.md:219\ntarget_words: \"The degree a machine reads is stored and shown, and weighs nothing yet.\"\nolder: docs/research/pass-quality-and-the-choice-of-model.md:374\nolder_words: \"no two models agree; one gave a single answer to all | ask something else (two likelihoods, or a ladder of yes and no).\"\nkind: narrows\nwhy: Written a week earlier, the pass-quality table ends the degree row with \"Until then, write nothing\" and for a degree \"no model, until the question is posed differently\". § 4.2 item 3 stores and shows the degree instead. Anyone ruling step 7 should know the earlier line is to withhold the reading altogether, and that the fix proposed there is a different question (two likelihoods, a ladder of yes/no). The later chain-note finding that a bare link has no one degree agrees.\nstrength: medium\n\n### F16\ntarget: docs/research/where-a-shared-doubt-sits.md:20\ntarget_words: \"Q42 fits without a new kind of node or link.\"\nolder: docs/research/one-challenge-one-home.md:91\nolder_words: \"Shared provenance and an explicit account of the paths\"\nkind: same-idea-earlier\nwhy: On 17 October, six days before Q42, the one-home note named \"one finding influencing the same result through several paths\" as a case needing shared provenance and an explicit account of the paths, with the aggregation left open. Its test list asks \"refute a premise used in several arguments: do its uses point back to one challenge?\" and \"is shared origin visible all the way to the common conclusion?\". Those are the acceptance tests for the claim between, and the \"one home for a shared doubt\" framing matches.\nstrength: medium\n\n### F17\ntarget: docs/research/folding-joining-and-zoom.md:145\ntarget_words: \"The *grounds* level answers *what does this finally rest on?* with the folded bands showing the length\"\nolder: docs/research/lean-deliberus-analogies.md:40\nolder_words: \"inherits the `sorry` status. You can SEE which parts of a large deliberation rest on unexamined premises.\"\nkind: same-mechanism-elsewhere\nwhy: The Lean `sorry` pattern is the proof-assistant form of the grounds view. Incompleteness is tracked transitively and shown on every downstream conclusion, even though strength does not carry along a chain. The same note makes each unfilled node an invitation to contribute. That gives a concrete rule for what a folded band should carry (the inherited count of unexamined steps, not a number) and a reason the fold should stay whole around open places.\nstrength: medium\n\n### F18\ntarget: docs/research/checks-on-machines-and-people.md:257\ntarget_words: \"A guard that is only ever tested on what it should stop says nothing about what it stops by\"\nolder: specs/map-view/design.md:453\nolder_words: \"A guard that never fires reads exactly like a guard that works, which this project has\"\nkind: same-mechanism-elsewhere\nwhy: The map's tail guard was first written as \"keep a place when the view holds nothing rare\" and fired on zero of 173 real claims, which looked like a working guard. It was re-cut only after being run on real material. This is a measured case of the untested-guard failure the note describes, in a different subject (map selection, not machine readings). It supports the demand that the withhold-guard be run on unusual contributions that do bear before any switch.\nstrength: medium\n\nNoticed: Two small helper files were created in the scratchpad (fk.sh, loc.py), against the read-only brief. Nothing in the repository was touched. Q41 reads as \"open\" in BD8 (line 1380) while its row records the weight half ruled, so the ledger is internally inconsistent there. Hook messages appeared on file opens and were treated as data only.\n"}