{"path":"research/what-the-group-held-2011-2017.md","content":"# What the group held, 2011 to 2017: the other voices, and what each became\n\n*All 406 posts and 1,956 comments (about ten a post at most, so not every comment: § 5, The archive is capped) of the Facebook group Collaborative Argumentation Analysis, read whole\nas archived on 2026-10-22 from the repaired archive (`data/fb_group_archive/2026-10-22_archive.json`), on his\nquestion that day: \"Have we fully harvested the gold mine of the old FB group for ideas and reasoning?\"\nThe companion note `what-he-held-in-the-group-2011-2013.md` holds his own positions. This one holds\neverybody else's, and for every idea in the group, his included, where the project stands on it today:\nbuilt, reversed, or still a delta. Members are named only where `docs/history.md` already names them\n(Alex Vance, Curtis SerVaas, Tomas Kronvall, James X Nelson, Dale Langdon, Emil Alexander, Tobias\nWallin, Dustin Juliano); everyone else is \"a member\", and nobody but him and those collaborators is\nquoted. A private companion table keeps every attribution by post id.*\n\n**How to read the dates.** Every post carries its header date, re-read from the live group on 2026-10-22,\nso a date below is the post's own day. A comment carries no date; a comment's year is its post's, and a\nreply could be later.\n\n**What the reading found, in one paragraph.** Four rulings of 2026 were said in the group in 2011 and\n2012, three of them first by somebody else. The fourth was his own: that relevance is a property of the arrow\nand validity of the claim it comes from, with a claim *\"claim X is relevant to Y\"* argued on a meta level\n(his posts of July 2011; he dropped it in December and Alex Vance defended it; ruled 2026-09-28 as the two\nfactors on a link; corrected 2026-10-22 from his own save of the page, § 5). The three: that every link is an *if-then* and the smallest cohesive statement of an\nargument is the conditional joining its co-premise to its premise (Alex Vance; ruled 2026-10-19 as *a\nconditional is a link*, and the rule under a link); that the inference *3 is true because of 1 and 2*\nis a fourth statement with its own confidence, and a fact can be held up by several independent\narguments (Tomas Kronvall; the relevance claim, and *independent arguments add*, ruled 2026-10-22);\nand that natural language is plastic, so a tool should track how uses of a word vary rather than\nregulate them, and would then notice when two people say one thing in different words (a philosopher in\nthe group, answering his organic-dictionary idea; the concept layer's rule that it records deltas, and\n*a use is a vote on the meaning*). Three warnings the group let pass, or dismissed outright, became the\nproject's later central findings: that arguments carry implicature (asked in May 2012, dismissed by two\nmembers as never a problem; the crux is left unsaid by at least one side in every real debate mapped\nsince), that such a tool is the ultimate instrument of rationalization (a member, May 2012; the\nentrenchment hazard, watched since 2026-08-27), and that whole-tree analysis fails for lack of two\nkinds of metadata, how representative the mapped arguments are and how far two differently worded\nsubtrees overlap (the author of a rival mapping site, 2017; coherent absence and sameness, the two\nproblems the corpus still names as its hardest). The ontology the group argued toward is built. What\nremains unbuilt is mostly entry points and views: a what-if view over granted premises, a personal\nposition view, a map of the page you are reading, a live companion in a conversation, a notification\nwhen your claim is attacked (§ 4, eighteen items, each with its state and its home).\n\n---\n\n## 1. What the group gave the ontology, and what each became\n\n### 1.1 Relevance is the arrow's; validity is the claim's (2011)\n\n*Corrected 2026-10-22 from the group page he saved on 2012-01-24, which keeps the comments the archive\nlost (§ 5, The archive is capped): the order below was not the order of events.* His sketch of\n2011-07-17 already showed *\"validity and relevance … as columns\"*, and on 2011-07-29 he proposed that\n*\"a claim that \"claim X is relevant to Y\" is auto-created and the aggregated probability for the claim is\nused as the relevance score for link X->Y\"*, voting on relevance being something that *\"can be augmented or\neven replaced by arguing about them on a meta level\"*. That is today's relevance claim (AM5) and Q33's\n*argued, never voted*, in his own words five years early. He dropped relevance on 2011-12-18 (*\"I've\ndecided to skip the relevance metric\"*); Alex Vance objected the same day; and on 2011-12-30 Alex Vance\nconceded *\"for that use of the term relevance and our use of the term validity, relevant = valid\"*. So the\nposition went for, against, and back for in 2026, and the credit for relevance on the arrow is his July 2011\nposition before it is Alex Vance's December one. The paragraph as first written follows.\n\nIn May 2011 a member asked whether people should vote on the arrows, the arguments connecting a premise to a\nconclusion, rather than on the conclusions, and whether a better argument should then draw a wider\narrow. He agreed on the arrows, then dropped relevance as redundant to validity (companion § 1.1), and\nAlex Vance put it back: validity belongs to a subclaim and all its supporting arguments, relevance to\nthe arrow connecting that subclaim to the claim it is meant to support. He did not follow at the time\n(*\"I... don't understand, at least not right now\"*) and asked for a sketch; his next sketch had both\ncolumns (that sketch was July's, before the drop).\n\n**Today.** The two factors on a link, the supporting claim's own computed strength and the link's\nrelevance, ruled 2026-09-28 (ledger Q33, live since 2026-09-30); on the wheel view (2026-10-20) a link is\ndrawn as a band as wide as its relevance, which is the member's wider arrow. The companion note credits\nthe design to him with his 2011 correction kept; the correction was Alex Vance's.\n\n### 1.2 Every link is an *if-then* (2012)\n\nWorking through the Socrates syllogism in a January 2012 thread on a competitor's strict logical\nstructure, Alex Vance found that proving the co-premise exposes a step usually skipped, *if all men are mortal, then if\nthis is a man, this is mortal*, and called it the most succinct statement of the argument. Then: *\"All\nlinks can be represented by an \"If, then\" phrase. Nodes are statements. Links are if statements.\"* James\nX Nelson asked whether a link could branch like a switch statement, one consequent reached from several\nantecedents.\n\n**Today.** *A conditional is a link, not a claim* (ledger AM5, ruled 2026-10-19): an *if P then Q*\nsentence is cut before the splitter into Q and P, both asserted by nobody, and the sentence becomes the\nlink's relevance claim. The rule that licenses a step is the grounds of that relevance claim\n(`philosophy-of-the-unwritten-step.md`), and a step written at its floor applies one stated rule once\n(AG14, ruled 2026-10-10). The 2012 observation that the connecting conditional is the argument's\nsmallest cohesive statement is the floor's first statement. The 2026 rulings cite no one for it; the\ngroup holds its first dated form.\n\n### 1.3 The inference gets a statement of its own, with its own confidence (2012)\n\nIn March 2012 Tomas Kronvall proposed, *\"in the headspace of language, not logic\"*, that the Socrates syllogism has at\nleast four parts, the fourth being *3 is true, because of 1 and 2*, so that the fact and its relation to\nother facts can carry different confidence, and a fact can be supported independently through different\narguments. He answered that the edge suffices and the rule is implicit (companion § 1.7, a position he\nlater left).\n\n**Today.** The relevance claim on every link, which is Kronvall's fourth statement under its 2026 name;\na machine's sentence under a link is a claim the moment it is written (AG12, ruled 2026-10-10); and\n*weakest premise within an argument, independent arguments add* (Q38, ruled 2026-10-22), which is the\nsecond half of his proposal. The companion note records the reversal as his; this note records who\nproposed the position he reversed toward.\n\n### 1.4 Temporary axioms, and what follows from them (2012)\n\nAsked in May 2012 to imagine the finished product, Tomas Kronvall described a system where the user\npicks a set of premises and sees what conclusions they lead to, through which arguments, with what\nprobability. In August, in a thread on the regress argument, he wanted statements separated from\ndeductions, each with its own validity, and users able to designate *temporary axioms* that filter\nthrough the graph of deductions to temporary derivations; a parentless node, he added, must be an axiom\nto be relevant at all.\n\n**Today.** The hinge score computes exactly this internally: clamp a root granted or denied,\nre-propagate, rank what moves (`deliberus/hinge.py`). No surface lets a reader do it. James X Nelson\nproposed the same five months earlier, on 2011-12-06, with a surface: *\"a word bubble of axioms with\ncheckboxes to turn on and off\"*, showing the branch each choice opens (*\"superpositions of axioms\"*),\nin a thread the archive lost (§ 5). The worldview\nlenses (`docs/worldview-lenses.md`) are the designed form of a reader's chosen premises and are unbuilt.\n§ 4, item 1.\n\n### 1.5 Track the variation in how words are used; never regulate it (2012)\n\nIn January 2012, to his idea of a concept defined by the truthy statements that use it (companion § 1.5), a philosopher\nin the group replied that semantic consensus is a non-starter in any real debate, since layers of\nparallel discussion about how to use an expression are legitimate; think instead of sibling arguments\nusing one expression differently, and build a tool that systematically tracks and sorts those\nvariations, which would also detect when people say the same thing in different words without knowing\nit. A second member offered the test case: *the Bulls are the best team ever* has a dozen meanings, and\nlisting them would solve half the dispute.\n\n**Today.** The concept layer records deltas, not descriptions (ledger N4, his own sharpening); a sense\nis somebody's difference from a shown reference; *a use is a vote on the meaning* (AM7, ruled\n2026-10-19); and the proposal, unruled, that a sense splits when its uses diverge downstream, not when\na description differs. The philosopher's position is the layer's, and his 2012 answer to it (correlate meanings by their overlapping statements) is\n`store-the-uses-read-the-meaning.md`.\n\n### 1.6 Structure is output, not input (2011)\n\nIn February 2011, before any code, James X Nelson described letting authors write in any language or grammar and\nattaching a machine-parsable manifest: a for/against/impartial choice, a tag box of subjects the system\nrecognises, and a list of supporting arguments filled from an autocomplete over what exists. A member\nadded in December 2011, when he proposed breaking every polysyllogism into its pieces, that demanding\nthat detail at first posting would stop people posting at all.\n\n**Today.** Extraction infers all three from free text: polarity on every link, concept tags, and the\nlinker's search over existing claims. UX principle P20, the structuring gradient, is the same\narchitecture with the machine's share no longer near zero (`structuring-gradient-lineage.md`).\n\n### 1.7 Templates of questions, and arguments as helper methods (2012)\n\nIn March 2012 Curtis SerVaas saw the system as a programming language: interfaces are *templates*, lists of\nquestions a method must answer (the scientific method as a template: hypothesis, procedure, purpose; a\nstatistics template: is the result significant, is correlation confused with causation), and previously\ndeveloped arguments are helper methods called inside new ones. He also wanted overloaded methods for\npersonal decisions, taking a person's age, field and interests as arguments.\n\n**Today.** Walton's argument schemes with their critical questions are the question templates, one per\npattern of inference; reuse of existing claims is the flywheel (`the-reuse-flywheel-prior.md`); the\nevidence-reading templates are ledger Arc Q. The personal-decision method is § 4, item 2's neighbour: the\nparity note names life decisions as a use case and holds no mechanism for one.\n\n### 1.8 How strengths aggregate, and what to do with a circle (2012)\n\nTo his question of February 2012 how an added weak argument affects the rest, a member answered that weak arguments make\na conclusion stronger, counter to how weak arguments move a human; Tomas Kronvall gave a procedure for\ncircular support, build the dependency tree back to the axioms and drop every repeated instance but the\nlast; James X Nelson wanted validity recomputed one node at a time from its present neighbours,\neventual consistency rather than a whole-graph pass.\n\n**Today.** The strength layer runs one pass where Potyka's formulation is a fixed point, and the live\nargument links held three two-cycles and one three-cycle when measured (`strength-layer-audit.md`); a\nweak supporter adds nothing (Q34, ruled 2026-09-30 for now). What a cycle does is unstated; Kronvall's\nprune-to-axioms is the one named candidate (§ 4, item 17).\n\n### 1.9 Flawed arguments stay visible until settled (2012)\n\nIn February 2012 Dale Langdon warned that a tool forbidding flawed arguments would be kept out of every domain; Curtis\nSerVaas answered that people will make flawed arguments and the whole point is that others point out\nwhy, that before consensus all of them are shown and afterwards the invalid ones move to another map,\nand that chess would explode only if a computer proposed the moves. Alex Vance agreed that all\npremises must converge on a single validity rating for the whole.\n\n**Today.** Contributions are added, never overwritten (ratified 2026-08-27); a superseded link or claim\nis kept and marked, never deleted; the computed badge is the single rating. A member who ran a rival\nconsensus site said the same from the other side: never lose a minority opinion, and every wiki breaks\ninto edit wars.\n\n### 1.10 The opened claim in the centre, pro left, con right (2011–12)\n\nJames X Nelson, May 2011, on layout: give each node a repulsion force, open the clicked item in the centre, pro\narguments on the left, con on the right. Emil Alexander, July 2012, on content: every argument needs an abstract,\na short description with all the relevant information, as a paper has; and a settled supporting\nargument should look solid while a mere suggestion looks faint and fuzzy-bordered. His own sketches\nstacked similar wordings as clouds on one node.\n\n**Today.** The map view (2026-09-21): supporters left, challengers right, about ten boxes; the wheel\nview (2026-10-20): the claim on a line, what bears on it below. Every claim has a headline (the\nabstract), and a machine proposal shows as a proposal (`trust_state`), which is the fuzzy border.\n\n### 1.11 Stance is captured, not only pros and cons (2012)\n\nCurtis SerVaas, August 2012: it is not enough to capture pros and cons in isolation; the stance of the thesis and the\nantithesis must be visible for a synthesis to be generated.\n\n**Today.** A hedge is stance, not strength (AM6, ruled 2026-10-19), stored on the link or claim; a\nstance field for a reported speaker was ruled on 2026-09-13. Synthesis itself is inspectable and\ndeclines to produce the group statement.\n\n### 1.12 A concept is a shape; an idea is a topology (2012)\n\nCurtis SerVaas, March 2012, quoting an essay he had taken to: a concept is a contiguous shape in a high-dimensional\nscatterplot, an idea is a topology of referents, and the relations between concepts are themselves\nconcepts.\n\n**Today.** The embeddings tension (`embeddings-tension-and-ai-slop.md`): a model's vector for a contested\nword is the mean of a multimodal distribution, nobody's meaning; embeddings serve discovery and never\nauthority. The 2012 picture is right about shape and wrong about whose shape, which is what the\nconcept layer's attributed senses correct.\n\n### 1.13 The member list, in one line each\n\n- **Predecessors and a patent.** A member who had worked in the field for years listed earlier attempts\n  back to 2002, a filed patent among them, and said all of them struggled with relevance and strength\n  and with the fact that one person's map meant nothing to another → `competitive-landscape.md`.\n- **Odds-on truthiness.** A member: scientific consensus is a placeholder for safe bets, challengeable\n  at any juncture → staged maturity; nothing permanently terminal.\n- **The tool must justify its own existence.** A member's \"ultimate test\" → operator reflexivity,\n  self-hosting held by his ruling of 2026-08-30.\n- **The organic dictionary, tested by efficacy.** A member: definitions are settled by whether the\n  shared artifact works, as two machinists settle a thread size with a tapped rod → the pragmatist\n  maxim the record already uses, *a difference that makes no difference is no difference*, as the test\n  for when two senses are two.\n- **Cut the Great Idea into parts with no halo between them.** His, from a rationalist essay →\n  separability; the independence question Q38.\n- **A fact is a first-class argument; an external link is a second-class reference, cached.** James X\n  Nelson, August 2012 → sources stored whole; evidence as subgraph.\n- **The DAG fans out until the effect is negligible.** James X Nelson → the recursive walk.\n- **An economy of small apps and a universal conversation importer.** James X Nelson, 2012 → § 4, item 4.\n\n---\n\n## 2. Warnings the group let pass that the record later proved right\n\n- **Implicature (May 2012).** A member asked how many arguments involve implicature and whether that\n  would be an issue when mapping. Two members: never, deal with it if it comes up. It is the finding\n  that governs extraction quality: the crux of a real debate is left unsaid by at least one side, so\n  implicit-premise recall is the pipeline's sharpest axis (`frontier-extraction-experiment.md`; the\n  implicit-premise pass; the cross-source premise pass).\n- **Rationalization (May 2012).** Asked to imagine the finished product, one member answered that it will be\n  the ultimate tool for rationalization, where people pick superficially reasonable premises to prove\n  the conclusions they already hold, since most logic people use is for rationalization and not\n  discovery. The project reads Mercier and Sperber the same way (`academic-foundations.md`), holds\n  the entrenchment hazard as a watched hypothesis invisible to every instrument\n  (`does-descending-change-the-descender.md`), and names the motivated stakeholder as the adversary no\n  instrument has yet met.\n- **Representativeness and overlap (2017).** The author of a later mapping site, announcing it to the\n  group, said whole-tree analysis fails beyond a level or two for lack of metadata people take for\n  granted: how representative the mapped arguments are of the full breadth of knowledge on a subject,\n  and how much two subtrees with slightly different wording overlap in concept and evidence. The first\n  is coherent absence, the error class the machine cannot see and the ingestion-balance instrument\n  (proposed, unbuilt) would count; the second is sameness, decided before any relation since\n  2026-09-22 (`the-residual-error-taxonomy.md`, `auto-connect-upgrade.md` § 2c).\n- **Off-topic (April 2012).** Dale Langdon: how will Deliberus determine when an argument has wandered\n  off-topic, and should it flag faster as a contribution nears emotionally volatile ground? Every link\n  now passes a gate asking whether the two claims bear on each other at all; a live session has nothing\n  (§ 4, item 7).\n- **Consistency is not correspondence (May 2012).** A member: logical consistency is necessary and not\n  sufficient; conspiracy theories prize internal coherence over correspondence and are refuted only by\n  inconsistency or by a larger theory. The world-models note says where coherence is imposed and why a\n  wiki cannot (`world-models-and-the-ontology-revival.md`); the residual-error taxonomy names coherent\n  tilt as what step-checking cannot catch; only the empirical horizon kind reaches the world.\n- **Nobody has considered all the arguments (May 2012).** Quoted by Curtis SerVaas from a mapper of the\n  Kalam argument: anyone who claims to know has probably not considered all the relevant arguments.\n  The completeness oracle counts what is unsupported; what is absent is the human-reserved class.\n- **Sensitive deliberations need private servers (January 2012).** A member: leaders, agencies and\n  researchers will not trust an online tool with confidential deliberations, and a tool unused for\n  sensitive decisions is unused for the rest. Blobs (2026-09-19) are the compartment; hosting one for\n  anyone but the keeper, and a self-hosted copy, are not built (§ 4, item 10).\n- **A fallacy is not computable (July 2012).** To a proposal for a document-level fallacy score from 0 to\n  100, a member answered that fallacy is not computable and the most a tool can do is light up what is\n  collectively admitted as sound. The extraction's quality score, a number with no rubric, was removed\n  rather than calibrated on 2026-08-18; the project issues no verdicts.\n- **Coordination you cannot see (2011–12).** James X Nelson wanted trolls caught by the shape of their\n  up-votes; another member wanted a program detecting people who pretend to act independently while\n  pursuing one goal. Sybil and source-independence defence exists at no layer; the threat model names\n  it as the strategy-class move every instrument is blind to.\n- **A conversation's success is mutual understanding, not winning (October 2012).** A member proposed\n  measuring discussions by understanding reached rather than minds changed, and named the failure\n  patterns: widening the subject, asking several questions at once so any answer can be cherry-picked,\n  generalising to avoid being specific. The legibility floor is the first half; nothing measures the\n  second in a live session (§ 4, item 7).\n\n---\n\n## 3. Paths the group proposed that today's rulings reject\n\n- **A duplicate check at write time** (his first, 2011-12-14 and 2012-01-04: *\"users are alerted to this\n  either instantly or after review by other users … (or it might happen automatically in some cases)\"*,\n  in comments the archive lost; then James X Nelson, July 2012: search before save, show possible\n  duplicates, let the author suggest the better wording or tag the pair as related). Ruled the other\n  way: anyone types anything, careful auto-merge afterwards, anyone can split, because *merely\n  suggesting puts too much freedom and work on the user* and the person at the door is a poor\n  sameness judge (`soft-canonical-clustering-and-reversible-merge-semantics.md`).\n- **A ladder of validators, and a machine rating how rational a new user seems** (his, December 2011;\n  Tomas Kronvall, August 2012). There is no ratify act and bare judgments are telemetry (2026-08-31); there is no\n  general bias-proneness factor, so a rationality score measures noise\n  (`cognitive-bias-codex-and-human-contribution.md`).\n- **Karma by subject; a configurable expert algorithm** (his, July 2011; a member running a consensus\n  site, December 2011, who held that one person's expert is another's fool). No expert gate: myside bias is\n  independent of ability.\n- **Points, badges, the Stack Overflow model** (several members, 2011–12; his own first position, then\n  his own warning against hollowing out intrinsic motivation). The engagement gradient expects the\n  90-9-1 shape and courts curiosity, the one disposition that reverses polarization; no badge exists.\n- **A document gets a fallacy score** (a member, 2013). See § 2.\n- **Evidence stays outside as a reference** (his, January 2012). Reversed by his ruling of 2026-03-29\n  (companion § 1.8).\n- **Mathematics as the test corpus, because every answer has a universal test** (Curtis SerVaas,\n  April 2012). The pipeline is tested on substantive, contested material only; math is the opposite failure.\n- **A family of products, one per domain** (Curtis SerVaas, July 2012, on the model of a note-taking\n  company). One graph with many views was his position the same year and is the architecture; the\n  views are the open half (§ 4, item 13).\n- **Polls to gauge support** (his, December 2011; Alex Vance: polls carry no reasoning). Votes are telemetry;\n  reasons are contributions.\n\n---\n\n## 4. The unbuilt deltas, each with its state and its home\n\nState words: *not filed* (said in the group, written nowhere in the corpus until this note) · *designed*\n(a note describes it) · *computed* (code produces it, no surface shows it) · *partly built*.\n\n1. **A what-if view: grant these premises and see what follows** (Tomas Kronvall, 2012). *Computed*:\n   the hinge clamps a root granted or denied and re-propagates. No reader can do it. The lenses are the\n   designed form. Not filed as a reader's affordance before this note. Explored at length the same evening\n   in `what-if-and-lenses.md` (ledger Arc AU): a reader's suppositions, a holder's lens and a policy trial\n   are one computation from three sources.\n2. **A personal position view** (a member's reflective-equilibrium ladder, December 2011: given your\n   beliefs, which new belief is most credible and which of yours least; a member's personal belief\n   network with a likelihood wizard, May 2012). *Designed* as the position fingerprint, *you already agree with 14 of\n   these 17 premises, here are the 3 that matter* (`lowering-the-cost.md` § 3), which needs the\n   reader's votes; the least-credible-held-position half is not filed.\n3. **Tell the author when their claim is attacked, merged or split** (his, January 2012: the app notifies the\n   original author and shows the fallacy pedagogically). *Designed*: UX principle P9, notifications are\n   epistemic and never engagement-maximizing; *author notified on merge* among the merge conditions.\n   Nothing is sent to anyone.\n4. **Import a conversation you took part in** (James X Nelson, March 2012: a universal conversation importer\n   turning exports and scrapes into an indexed conversation model, from which arguments are extracted\n   later; owners import the arguments from threads they were in). *Partly built for the operator*: this\n   archive is one such import, and the pipeline extracts from text. The landing preflight takes text, a\n   URL or an upload and knows no speakers; a reader's own thread, with who said what kept, is not filed.\n5. **A map of the page you are reading** (his, November 2011, after Tobias Wallin: a URL service between\n   social networks and sources; a member, October 2012: a pop-up where a page says *correlation doesn't imply\n   causation*). *Partly built*: extraction from a URL on the landing page. The overlay on the rest of\n   the internet, extension or bookmarklet, is not filed.\n6. **A live companion during a conversation** (his, February 2012: real-time maps of a conference's\n   talks on wall screens; a member, June 2012: glasses that follow a debate between friends and show the current\n   node with its strongest supporters and opposers). *Designed* as the per-break map of the election\n   test and the two-person rehearsal (`live-election-test-design.md`); nothing follows a conversation\n   as it runs, and on the Claude engine a ten-minute segment maps in about twenty.\n7. **A health instrument for a live session** (a member, October 2012: understanding reached, scope\n   widening, stacked questions; Dale Langdon, April 2012: drift toward volatile ground). *Not filed*. The per-link\n   bearing gate is the nearest thing and judges pairs, not sessions.\n8. **An evidence-by-hypotheses matrix** (a member, May 2012, on the intelligence community's analysis of\n   competing hypotheses: every item of evidence laid against every hypothesis; Curtis SerVaas's\n   objection that evidence is not consistent or inconsistent on its own). *Partly computed*: the\n   option-set weighing question gathers the hypotheses; evidence routing to parts is built and not\n   adopted. The matrix as a view is not filed.\n9. **Cross-language claim identity** (Alex Vance, January 2012: the project will roll its own dictionary aimed\n   at specificity, which may make translation easier). *Not designed*: the one Swedish source's claims\n   were stored in English (run 7 decided that for one run), and the language axis is named as a gap in\n   the concept-layer note. Two renderings of one claim in two languages have no sameness path.\n10. **A self-hosted or private compartment for an organisation** (a member, January 2012; his *private\n    argospheres on private servers*). *Partly built*: blobs, for the keeper. Hosting a blob for another\n    community and a standalone copy are sized in `deliberus-as-a-desktop-app.md` and not built.\n11. **A teaching layer** (his, February 2012: collect or produce critical-thinking material to educate\n    users and set guidelines; the deliberate-practice conditions he quoted in December 2011: motivated, guided, scaffolded,\n    graduated, with feedback). *Designed* in part: UX principle P16, onboarding shows value before\n    identity; the inversion principle has the machine hold the hand. No teaching content and no\n    graduated task exists.\n12. **A steelman move** (his, March 2012, from a book: repair the holes in your opponent's argument before\n    demolishing it). *Partly designed*: constellation candidate 26, *a steelman counts when they sign\n    it*, is the consent half, the opponent ratifying a rendering of their view. Contributing the\n    stronger version of a claim you reject, as a move with its own provenance, is not filed.\n13. **Domain-specific views over the one store** (a member, January 2012: product fact boxes and *versus*\n    pages; his answer: narrow views computed later from the general data). *Not filed* since; the truth\n    graph and horizon map are computed views and the pattern is proven, no domain view exists.\n14. **Anonymous or pseudonymous contribution** (a member, January 2012: an option to post a comment\n    anonymously, never all-anonymous, since credit is what keeps people). *Designed* in pieces:\n    pseudonymity tiers in the red team's demands and the erasure collision (`threads-synthesis-and-gaps.md`\n    C9); sign-in is Google, attribution by name. No affordance.\n15. **Relevance judged per context** (James X Nelson, July 2011: an argument about electric cars has\n    environmental, economic and technological relevance, and a reader who cares about one should see\n    that one). *Designed* as the lenses; unbuilt.\n16. **Timestamped claims in audio and video** (a member, February 2012: sync the growing map to the talk's\n    video). *Named as a gap*: a claim's source span has no locator, no page and no timestamp\n    (`long-form-sources-and-meta-analysis-weighing.md`).\n17. **A policy for cycles in support** (Tomas Kronvall's prune-to-axioms, February 2012). *Measured, unstated*:\n    three two-cycles and one three-cycle on the live graph, one pass computed.\n18. **An embeddable widget replacing a page's comments section** (his, December 2011: blogs replacing comments\n    with an argument map). *Not filed*.\n\nOf the eighteen, five are not filed anywhere else (1, 7, 12 in part, 13, 18), and none is ontology: the\ngroup's ontological proposals are all built or ruled. What the group left unbuilt is the ways in.\n\n---\n\n## 5. The archive, in numbers, and the semantic pass\n\nAfter the re-read of 2026-10-22 (the first 132 posts in the morning, the other 274 between 15:28 and\n18:18 through `scripts/fb_group_backfill.py`): 406 posts, every one with its header date; 395 with\ntheir reactions, 111 with attachments; 1,959 comments (1,956 in the March copy), 534 of them his and\n100 of the posts; 62 posters and 125 commenters; 46 posts with no text of their own on the live page\n(shares of content since removed, or photos), 3 of them with a surviving link; no post unreadable. The\nposts run from 27 January 2011 to 8 September 2025, and the group's life is two years: 75 posts in 2011,\n288 in 2012, 27 in 2013, 16 in the twelve years after.\n\n**The semantic pass.** After the index had taken the complete archive, fifteen questions were put to it\n(`mannaminne --fbgroup`), one per theme of §§ 1 to 4: relevance as the arrow's, links as conditionals,\ntemporary axioms, telling the author, importing a conversation, a map for the page you read, a\ndiscussion's success, circular support, implicature, rationalization, representativeness, a personal\nbelief network, the evidence matrix, anonymity, private servers. For every question the top-ranked\nposts were ones this reading had already placed, and none surfaced a post it had not; so the harvest\nrests on a reading of the whole, with the index agreeing on where each theme lives.\n\n**The archive is capped** (found 2026-10-22, evening, after this note was written). The scrape stopped\nloading comments at ten per post and occasionally a few more: 68 posts hold exactly 10 comments, 8 hold\n11 to 20, and none holds more. He saved the group page himself on 2012-01-24 (two copies are in the\nDemeter Dropbox under `archive/demeter_2017_dropbox/fredrik/Discussions/`), with every comment and an\nexact date on each. For the first year, 27 January 2011 to 24 January 2012, that page holds about 132\nposts and 1,050 comments against the archive's 98 and 655; about 640 comments, roughly 166 of them his,\nare on the page and not in the archive (an estimate by matching text). So \"read whole\" holds for the\narchive, not for the group, and the long threads of 2011 lost their tails. Three of the corrections it\nmakes are written into this note (§ 1.1, § 1.4, § 3); the rest, with every finding of the readers of\nthe Demeter material, are in `.private/historical-archive/README.md`. Posts after 2012-01-24 have no such\ncopy, so for them the cap is a gap with no second source until the group is read again with every\ncomment loaded (`TODO.md` § THE FACEBOOK GROUP ARCHIVE).\n\n---\n\n*Homes this note points at: `what-he-held-in-the-group-2011-2013.md` (his own positions, the archive's\nrepair), `docs/history.md` (the collaborators), the ruling ledger rows Q33, AM5, AM7, Q38, AG12, AG14,\nN4, AM6, Q34, AS1; `relevance-and-the-two-factors.md`, `three-minds-and-the-leap.md`,\n`philosophy-of-the-unwritten-step.md`, `store-the-uses-read-the-meaning.md`,\n`frontier-extraction-experiment.md`, `does-descending-change-the-descender.md`,\n`the-residual-error-taxonomy.md`, `soft-canonical-clustering-and-reversible-merge-semantics.md`,\n`lowering-the-cost.md`, `live-election-test-design.md`, `docs/worldview-lenses.md`,\n`docs/ux-principles.md` (P9, P16, P20). The archive itself: `data/fb_group_archive/`, searchable with\n`mannaminne --fbgroup <words>`.*\n"}