{"path":"research/corpus-graph/readers/2026-10-27-batch-6-books-maps-landing-and-agents-as-a-reader-returned-it.md","content":"# What a reader found to bear on ten notes and specs on the twelve books, the map view, the landing and the agent population (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 10 notes:\n\n- `docs/research/the-agent-population-and-what-checking-is-up-against.md`\n- `docs/research/twelve-books-read-against-deliberus.md`\n- `docs/research/four-books-the-corpus-had-not-read.md`\n- `docs/research/legacy-object-model-harvest.md`\n- `specs/map-view/design.md`\n- `specs/landing-redesign/requirements.md`\n- `specs/landing-redesign/constellation-map.md`\n- `specs/landing-redesign/tasks.md`\n- `specs/landing-redesign/design.md`\n- `specs/landing-redesign/invitation-copy.md`\n\nIt made 181 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\n10 of 10 notes read in full.\n\n### F1\ntarget: docs/research/twelve-books-read-against-deliberus.md:412\ntarget_words: \"part of the link's relevance claim and moves no strength until a person argues that the part is\"\nolder: docs/research/session26-open-rulings.md:951\nolder_words: \"with no person needing to adopt it or argue it is required first\"\nkind: later-correction\nwhy: Section 6 says a machine's step under a link is stored as a part and weighs nothing until a person argues it is required. AG12 was revised 2026-10-22: once anyone attacks, supports or answers the step, it counts, with no adoption first. AD22 (ruled 2026-10-25, line 896) goes further. The founder's words on the part storage: \"Should probably not be stored as parts, historical accident right\". Rules become grounds, recorded as premises of one argument and counted by the weakest. So \"required can seldom be argued for a rule\" and candidate 10 in § 16 rest on a storage that was withdrawn.\nstrength: strong\n\n### F2\ntarget: specs/map-view/design.md:96\ntarget_words: \"a badge shows from three weighed contributions (answers that moved a question, one per account;\"\nolder: docs/research/weighing-answers-and-the-badge-threshold.md:95\nolder_words: \"One account can still answer ten different questions\"\nkind: narrows\nwhy: The map note reads the threshold of three as three accounts' worth of weighing. The note that built the threshold says it is one answer per account per question, so one account answering three different questions of one claim reaches three. It also says the threshold of three is \"mine\" and not ruled, and that nobody has asked whether an account's answers should share a budget. Anyone ruling on or building the badge gate should know that it can be reached by a single account.\nstrength: strong\n\n### F3\ntarget: docs/research/legacy-object-model-harvest.md:26\ntarget_words: \"Joint premises, separate arguments, and corroborating reports are different jobs. Grouping alone\"\nolder: docs/research/session26-open-rulings.md:464\nolder_words: \"An argument is one or more premises its rule needs together to reach its conclusion, with that rule\"\nkind: already-answers\nwhy: The harvest leaves open how joint premises combine and says a composite and a case-plus-rule view must not count as two independent arguments. Q49 (ruled 2026-10-26) defines an argument as the premises its rule needs together. Q47 (ruled 2026-10-25, line 462) fixes the combination inside one argument as a conjunction, with a shared root counted once. Q48 gives a range where dependence is unrecorded. A reader of the harvest can take the object and arithmetic as ruled, and its gate example as an instance of them.\nstrength: strong\n\n### F4\ntarget: docs/research/twelve-books-read-against-deliberus.md:630\ntarget_words: \"That is Gigerenzer's finding in another field: where evidence is sparse and dependent, a rule\"\nolder: docs/research/session26-open-rulings.md:462\nolder_words: \"Within one argument the case facts combine as a conjunction: the product where their reasons share no root, a shared root counted once\"\nkind: later-correction\nwhy: Section 10 takes support for the weakest-part rule from the finding that a rule ignoring dependence beats one that guesses at it. Q47 (ruled 2025-10-25, canon-blessed 10-26) replaced that rule. Inside one argument the facts now combine as a product, and the weakest link is only what a fold shows, never the arithmetic. Q48 then made an unrecorded dependence show a range, with the weakest premise as its upper end. The Gigerenzer bullet cites a rule that no longer governs and now points the other way.\nstrength: strong\n\n### F5\ntarget: docs/research/twelve-books-read-against-deliberus.md:751\ntarget_words: \"how much else moves if it falls | Woodward's test, turned on the model | the hinge score\"\nolder: docs/research/what-if-and-lenses.md:71\nolder_words: \"same amount: granted, the conclusion reads 0.529 or 0.471; denied, 0.5. Hinge 0.029 for all fourteen. The\"\nkind: narrows\nwhy: The table credits the hinge score with the \"dependence\" row and calls it the nearest thing to what Ziliak and McCloskey ask for. On the custody map, the one real case run, all fourteen premises that move the conclusion get the identical hinge of 0.029. The note says the ranking meant to show what the conclusion stands on says \"everything equally, and barely\". Until that is repaired, the hinge cannot carry \"how much rests on it\" beside a badge.\nstrength: strong\n\n### F6\ntarget: docs/research/four-books-the-corpus-had-not-read.md:91\ntarget_words: \"The size is stored and weighs nothing.** Extraction stores `effect_size` on a claim when a source\"\nolder: docs/research/session26-open-rulings.md:1280\nolder_words: \"`study_count`, `sample_size` and `heterogeneity`, extracted and inert today, get these jobs\"\nkind: narrows\nwhy: AY5 (ruled 2026-10-23) gives the inert extracted fields study_count, sample_size and heterogeneity jobs in the strength computation. effect_size is not among them. Its worked case separates \"the mean effect protects with probability above 0.999\" from one new setting at 0.889, which is Ziliak's distinction between an effect existing and its size mattering here. The note's gap is therefore partly closed and partly pointed at: one field is still left out of a ruled build.\nstrength: medium\n\n### F7\ntarget: specs/landing-redesign/requirements.md:40\ntarget_words: \"Extraction is paused: no surface may promise live analysis; the form leaves the hero\"\nolder: docs/research/game-feel-and-the-first-screen.md:303\nolder_words: \"Live extraction is not a stopgap. Shown, it is the entire show-don't-tell answer\"\nkind: supplies-a-missing-piece\nwhy: The constraint treats the pause as a reason to remove extraction from the first screen. The game-feel note argues the opposite for a cold graph. A map that assembles while a visitor waits (100 to 165 seconds on the paid tier) is the whole show-don't-tell demo, and the wait is where the system shows what it does not know. The 2026-10-24 paragraph asks for the extraction input as a big (+) button. Whoever builds it should know the button is the demo only if extraction runs and is shown.\nstrength: medium\n\n### F8\ntarget: specs/landing-redesign/invitation-copy.md:48\ntarget_words: \"Deliberus is built to do it. Each of you puts your position down, and the conversation continues\"\nolder: specs/landing-redesign/requirements.md:40\nolder_words: \"Extraction is paused: no surface may promise live analysis; the form leaves the hero\"\nkind: contradicts\nwhy: The English universal draft promises that two people enter positions and a map grows on the screen as they talk. The landing requirements forbid any surface from promising live analysis while production extraction is paused. The draft inherits a Swedish original written for a live election test, when extraction was on. Before the draft is used as landing or outreach copy, either the promise must be reworded to what exists, or the constraint must lift.\nstrength: medium\n\n### F9\ntarget: docs/research/the-agent-population-and-what-checking-is-up-against.md:162\ntarget_words: \"Today they need a sign-in and are rate-limited. So the strength calculus is not where a flood is stopped;\"\nolder: docs/research/cheap-to-add-slow-to-matter.md:99\nolder_words: \"That is the sybil attack, undefended at every layer, and its defense is identity infrastructure\"\nkind: narrows\nwhy: The note says a swarm cannot produce the standing-giving acts because they need a sign-in and a rate limit. The conservative-design note says accounts and rates are non-discriminating entry costs. It adds that a well-resourced actor making many independent-looking positions is the sybil attack, \"expensive and recorded, not impossible\". It also says the single-source ceiling is not sybilproof. So the write surface holds against a flood only as far as identity cost does, which is unbuilt.\nstrength: medium\n\n### F10\ntarget: specs/landing-redesign/constellation-map.md:44\ntarget_words: \"Checking vs persuasion**: *\\\"Machines out-persuade us; checking has to keep pace.\\\"*\"\nolder: docs/research/the-scrutiny-gap.md:43\nolder_words: \"after coaching, expert humans could tie an AI constrained to respond at human speeds and with human-length messages.\"\nkind: narrows\nwhy: The tile, and the hidden hero line \"Machines just became better at persuading than we are at checking\", state a flat claim. The study the scrutiny-gap note builds on found the AI's advantage was information rate. Coached experts tied an AI held to human speed and length. The why-now is a throughput gap, not that machines persuade better per message. A reader judging the tile's over-broad test should know the source supports the narrower claim.\nstrength: medium\n\n### F11\ntarget: docs/research/the-agent-population-and-what-checking-is-up-against.md:185\ntarget_words: \"size does not appear in it. At an error of 2% a step, a chain of ten machine steps is wrong about 18% of\"\nolder: docs/research/quality-control-for-deeper-steps.md:241\nolder_words: \"nothing to be tested on: *drift compounds with each hop, so there is a depth beyond which the\"\nkind: supplies-a-missing-piece\nwhy: The 18% figure assumes each machine step adds an independent 2% error. The deeper-steps note records the compounding itself as a prediction from 29 September that had nothing to be tested on, to be refuted if coverage holds flat as depth grows. The chain-inside-a-link note (lines 203-205) adds that a found sentence stands as far from its text at any depth. Only made sentences add a step. Read the figure as illustration, not as a measured rate.\nstrength: medium\n\n### F12\ntarget: docs/research/twelve-books-read-against-deliberus.md:239\ntarget_words: \"A claim can reach a high strength by shrinking. Nothing on a claim\"\nolder: docs/research/testable-bets/hedges-vagueness-and-testability-literature.md:176\nolder_words: \"a hedge can be a protective-belt device, and what matters is whether it is added after a failure (degenerating) or specifies new testable content (progressive)\"\nkind: supplies-a-missing-piece\nwhy: The twelve-books note proposes recording how a claim changed in answer to an attack, and counting the conditions it has taken on. The hedges note, from Lakatos and Popper, has the same test. A hedge added after a failure is degenerating, one that specifies new testable content is progressive, and this is \"a graph-level question over time\". A hedged claim says less. It also records the founder's ruling moving stance hedges out of the wording. That gives candidates 8 and 9 their criterion and a ruling to start from.\nstrength: medium\n\n### F13\ntarget: docs/research/the-agent-population-and-what-checking-is-up-against.md:141\ntarget_words: \"again, and slightly different each time. So a check keeps pace with a flood **only if each new paraphrase\"\nolder: docs/research/twelve-books-read-against-deliberus.md:483\nolder_words: \"Deciding sameness is the attenuator that loses nothing when it is right. Summarising is the one\"\nkind: supplies-a-missing-piece\nwhy: Both reach the same thesis from different ends, and neither names the other. The agent note says the race is decided where sameness is decided. The Ashby section calls deciding sameness the attenuator that loses nothing when right. It names the alternative, summarising, as the attenuator that loses the rare claims. That gives the agent note's U16 a theory, and a stated cost for the cheaper route: attenuating by summary would throw away the tail.\nstrength: medium\n\n### F14\ntarget: docs/research/legacy-object-model-harvest.md:43\ntarget_words: \"A bundle-shaped enclosure does not require an ArgumentBundle database type.\"\nolder: specs/map-view/design.md:314\nolder_words: \"cannot be pointed at, argued with or renamed by a visitor, because it is not a thing in the graph; and\"\nkind: narrows\nwhy: The harvest proposes the bundle as a view over claims. The map built exactly that: a section is a label on each claim, with no Section node (line 308). The map note measured the cost. A visitor cannot point at, argue with or rename a section, and cannot say the cut is wrong. The harvest's own aim, to make what is challenged visible, conflicts with the view-only reading. Ledger Q38 says such groupings are supplied, attackable and recorded as premises of one argument.\nstrength: medium\n\n### F15\ntarget: specs/landing-redesign/invitation-copy.md:12\ntarget_words: \"Every paragraph concedes something to the reader's likely\"\nolder: docs/research/what-if-and-lenses/outside-reading-psychology.md:298\nolder_words: \"a \\\"recipe\\\" intervention made messages more persuasive to opponents (standardised beta 0.24)\"\nkind: supplies-a-missing-piece\nwhy: The note names \"grants before it claims\" as the mechanism under the text's patience, from analysis alone. The outside psychology reading (section 7b) has measured evidence. Yeomans and colleagues' conversational-receptiveness recipe (acknowledgement, hedging, positive framing) made messages more persuasive to opponents. It also found no test of Rapoport's full four-step sequence, and one study found accurate restatement did not raise persuasion. So the claim is partly supported and partly untested, and the recipe includes hedging, which sits oddly beside \"describes rather than promises\".\nstrength: medium\n\n### F16\ntarget: specs/landing-redesign/constellation-map.md:42\ntarget_words: \"'Nobody has looked yet' is not a score.\"\nolder: specs/map-view/design.md:96\nolder_words: \"every map and the claim page leave out the badges *no data* and *unexamined*\"\nkind: narrows\nwhy: This tile of the public wall promises that absence is a state. The founder's word of 2026-10-24 hid the labels \"no data\" and \"unexamined\" on every map and claim page. The state survives only as an empty fill, and a weighed claim near zero can look the same. So the live product now shows the strangers who read the tile a page that does not say \"nobody has looked\". Anyone ruling on the tile should know it describes a rule the interface half keeps.\nstrength: medium\n\n### F17\ntarget: docs/research/the-agent-population-and-what-checking-is-up-against.md:249\ntarget_words: \"**The authority risk** (§ 5): a client reads ten thousand badges where a person reads one. Multiplied by\"\nolder: specs/map-view/design.md:96\nolder_words: \"`BADGE_MIN_WEIGHED` and `weighed_contributions` in `deliberus/graph/badge_inputs.py`, served as `shown` on every badge.\"\nkind: supplies-a-missing-piece\nwhy: The agent note leaves the authority risk to \"the gate on the strength layer stands\". Built since then (2026-10-24, after the agent note), a badge carries `shown`, false until three weighed contributions stand behind it, and every badge is served with it. That is a concrete gate for a client reading thousands of badges. See F2: the threshold counts contributions, not accounts, so it is a floor for honest clients, not a sybil defence.\nstrength: medium\n\n### F18\ntarget: specs/landing-redesign/design.md:9\ntarget_words: \"the vision/fractal-priors beat in beginner words (assembly-ceiling idea WITHOUT the term\"\nolder: docs/research/complexity-transitions-and-why-now.md:17\nolder_words: \"Assembly theory's authority is not available to borrow.** No Deliberus document, and above all no funding application, may write\"\nkind: supplies-a-missing-piece\nwhy: The design already drops the term for the \"what it could unlock\" section. The transitions note gives the rule behind it and a safe citation. No Deliberus document may write that assembly theory shows anything. The surviving mechanism is construction with reuse, owned by Kauffman's adjacent possible and Arthur's combinatorial evolution. Whoever writes the landing copy, or the Assembly ceiling star's description, should cite those sources and not assembly theory.\nstrength: medium\n\nNoticed: constellation-map.md:115 says `/graph/stats` still breaks the \"every number says who asserted what\" promise. TODO.md:2533 says `count_safe_summary` has been called by that route since 2026-10-10. I did not report it because constellation-map.md names TODO.md. Tool results carried \"CORPUS GRAPH\" hook text. I treated it as data and wrote no files.\n"}