{"path":"research/arguing-a-reference-class-filing-ux.md","content":"# Arguing a reference class filing: how a claim's prior, its reference class and its membership show on the page\n\n*2026-11-08. A design for his blessing (P28: every interface change comes to him first). Nothing here is built;\nthe one line live under the fold since 2026-11-07 stays as it is until he rules. Ledger Arc BK, all open.*\n\n## 0. The answer first\n\nA reader can do five things with a claim's prior, and each already has, or should have, exactly one home in the graph:\n\n| Act | Home (one) | Door on the page |\n|---|---|---|\n| See the prior and where it comes from | the reference class's count, pooled by code (BC1, BD4) | one sentence under the map, on the claim page |\n| Open the reference class | the reference class, a claim with role `reference_class` | the reference class's name in that sentence |\n| Contest that the claim belongs | the membership claim, named in the `IN_CLASS` link's `grounds` (BC3, built 2026-11-07) | a short link in the same sentence, opening the membership claim's page |\n| Propose another reference class | a new membership claim, filed by a person, role `unknown` (BI1) | a second short link; **no route exists today** |\n| Read how the reference class was counted (BI1's two figures) | computed by `class_audit`, never stored | the reference class's own page, folded |\n\nThree recommendations carry the rest: the claim page says the prior in one sentence a stranger can read (§ 4.1); the\nreference class's own page opens on a wheel whose upper half is its members and lower half its counted bundles (§ 5);\nand the prior is **never drawn as a card on a member's wheel**, since a reference class read as a reason for its own\nmember counts one consideration twice (BC8). The words, one per line for his ruling in order, are § 6.\n\n## 1. His words\n\n> *\"the whole deal of how to expose reference-class arguing in the UI feels like it deserves a focused UI/UX session\n> for that purpose\"* (2026-11-08)\n\n> *\"That seems too wordy for UI but idk\"* (2026-11-08, on the line live under the fold)\n\n> *\"How does one argue about what reference class a claim belongs to?\"* (2026-11-07,\n> `saying-what-kind-a-claim-is.md` § 9b)\n\n## 2. What the page can read today, on one real claim\n\nThe live graph (read 2026-11-08) holds fifteen reference classes; one has members. A real member:\n\n- the claim: *Individuals with low memory spans derive a greater benefit in estimate accuracy from having multiple\n  opportunities for estimation than individuals with high memory spans.*\n- its reference class: *published empirical findings across fields, as counted by the classes within this one*, with\n  no bundle of its own, so it counts its narrower reference classes' cases: 115 replicated of 222 counted\n  (`priors-as-standing-work.md` § 11)\n- the prior: 0.518\n- the `IN_CLASS` link's `grounds`: the text *read twice by membership.read at version a445b2680f7e*, so **no membership\n  claim exists for it**; the seven live filings all predate the membership claim, and `membership_standing` gives\n  them no claim to open. The door in § 4 shows on none of them until they are minted (a backfill with no model call;\n  filed in `TODO.md`).\n\n`GET /claims/{id}/membership` returns, per reference class: its id, text, population and outcome; who filed it and\nwhen; the membership's role (`formulation`, `evaluated`, `unknown`); whether it was read under the reference class's\ncurrent wording; the audit's four counts (`proposing`, `after`, `unknown`, `outdated`); the pooled rate and the raw\ncount; the membership claim's id, posterior and whether anyone examined it. Beside them: the prior, which reference\nclass gave it, a fallback, how it was reached, and the narrower reference classes the membership sweep proposed\n(`membership_proposal`). No route lists a reference class's members or bundles for a page, and the wheel map draws\nno `IN_CLASS`, `FILED_UNDER` or `WITHIN` link.\n\n## 3. Each act, its home and its door\n\n**3.1 See the prior.** Home: computed on read from the reference class's count (`prior_for`); `prior` on the claim is\na cache (BJ3 is the open row on keeping it fresh). Door: one sentence. Not a card on the wheel (BC8), not a badge.\n\n**3.2 Open the reference class.** Home: the reference class's claim page. Door: its population words inside the\nsentence, a link like every claim (P23).\n\n**3.3 Contest the membership.** Home: the membership claim (*The claim \"…\" is about one of: …*), which weighs in full\nuntil somebody examines it and then carries its own posterior into how far the rate applies (`membership_standing`,\nR12's shape). An attack on it is an ordinary attack on a claim, through the doors that page already has. Door: a short\nlink beside the reference class's name. The live wording, *argue the filing*, names an internal act; § 6 item 3\nproposes a question instead.\n\n**3.4 Propose another reference class.** Home: a new membership claim filed by a person, role `unknown` (BI1: a\nperson's filing by word alone does not say whether the reference class was tuned on this case). Door: a short link\nthat opens a search over reference classes by meaning. Two decisions only he can make (ledger BK2): whether a person's\nproposal moves the prior before anyone reads it, and what a person does when no reference class fits (entering a new\ncount is a reference class plus a bundle with a source; § 6 item 6 keeps that off the page for now).\n\n**3.5 Read the audit.** Home: none stored; `class_audit` computes the two figures (cases the generalization was formed\non; memberships filed after its wording was fixed) and the unknown and outdated counts. Door: the reference class's\npage, folded. It never shows on a member's page: a member needs its prior and the door, not the reference class's\nhistory.\n\n## 4. Mockups: the claim page section\n\nEvery word a visitor sees is written out. Square brackets mark a link; nothing else is shown.\n\n**4.1 Recommended: one sentence, two doors**\n\n```\nBefore its reasons\nStarts at 52%: of 222 published findings counted, 115 replicated.   [Is it one?]   [Belongs elsewhere?]\n```\n\n- *published findings* is the link to the reference class (its population, cut to the words before the first comma).\n- *Is it one?* opens the membership claim. *Belongs elsewhere?* opens the search of 3.4 (not built; hidden until it is).\n- Where pooling moves the prior off the raw share (the clinical reference class: 31 of 45 alone, 0.573 pooled):\n  `Starts at 57%: 31 of 45 such findings held up, pulled toward similar counts.`\n- Where the reading is outdated: `Starts at 52% (counted before the wording changed; to be read again).`\n- Where the reference classes overlap and give a range (BD4): `Starts between 40% and 55%: two counts fit it.`\n  with each count a link.\n\nWhy this form: 21 words against the live line's run of fragments separated by dots, and each fragment names something a\nstranger can check (the share, the count, the outcome). P26 holds: every word is true of this claim's prior.\n\n**4.2 Alternative: the prior only, the rest on hover**\n\n```\nBefore its reasons: 52%   [why?]\n```\n\n*why?* opens a small panel holding 4.1's sentence. Shortest, and the door to contest sits one click deeper, which\nP22's *no gap displayed without a move offered* does not forbid (there is no gap shown) but which hides the act this\nnote exists for. Rejected for that reason.\n\n**4.3 Alternative: the live line, reworded**\n\n```\nWhere its prior comes from\nPrior 0.52 · one of published findings (115 of 222) · is it one?\n```\n\nKeeps today's shape. *Prior* is a term a stranger does not have (the glossary has it still unblessed), and *0.52* reads as a score.\nRejected under P28.\n\n**Placement**: under the map, above *Source text*, where the live line already sits; the first screen stays the map's.\n\n## 5. Mockups: the reference class's own page, and what its wheel shows\n\n**5.1 The page, under its map**\n\n```\nA count other claims start from\nPublished empirical findings across fields: 115 of 222 replicated.\n\nCounted from\n  Open Science Collaboration: 35 of 97 replicated            [link]\n  Many Labs 1: 10 of 13                                       [link]\n  ... (each bundle a link)\n\nWithin it\n  Psychology findings · Nature and Science social science · ...   (each a link)\n\nClaims that start here (7)\n  Individuals with low memory spans derive a greater benefit ...   [Is it one?]\n  ...\n\n▸ How it was counted\n    222 cases counted when it was written\n    7 claims filed here since, not known whether it was shaped on them\n    0 read under an older wording\n```\n\nThe heading line *A count other claims start from* replaces nothing today (the page shows a reference class as a\nplain claim) and says what this claim is for. The folded audit uses BI1's figures in words: `proposing` as *counted\nwhen it was written*, `after` as *filed here since its wording was fixed*, `unknown` as *not known whether it was\nshaped on them*, `outdated` as *read under an older wording*. A figure at zero is left out except the first.\n\n**5.2 The wheel for a reference class.** Three options:\n\n| Option | Upper half (what it bears on) | Lower half (what bears on it) | Cost |\n|---|---|---|---|\n| **A, recommended**: members above, bundles below | *Members*: the claims it gives a prior to, each band as wide as how far the membership stands | *Counts*: its bundles, and the narrower reference classes it counts when it holds none | one query branch in `/claims/{id}/map` for role `reference_class`; two new half names (AR11); the engine unchanged |\n| B, members as a third ring | the reference class's own reasons (rare) | same | members far from the centre, hard to reach; nothing gained |\n| C, no wheel, the lists of 5.1 only | — | — | nothing to build; the page opens on an empty wheel saying *None yet* today, which is false of a reference class with seven members (P26) |\n\nA is true by P27 (above is what an item bears on, below what bears on it): a reference class bears on each member's\nprior, and its counts bear on it. Two frictions to rule on (BK3): a band's width on the map is a link's relevance,\nand a membership's nearest quantity is how far it stands (`membership_standing`, 1.0 until examined), so an\nunexamined membership would be drawn full where every other unexamined link is drawn at 60%; recommended, draw it at\n60% like any link nobody has weighed, and note that the arithmetic counts it full. And a bundle card's fill is its\nposterior like any claim's, never its rate, so *35 of 97* is written on the card, not shown as fill.\n\n**5.3 On a member's own wheel: nothing.** Drawing the reference class as a card would read as a reason, and BC8 says\nit counts once, through the prior. A faint mark on the focused claim's fill at the prior (where the posterior started)\nwas considered: true, small, and a mark that means *started here* on one card only, which P27 (candidate) would ask\nabout. Not recommended until the fill's meaning is settled (`specs/map-view/design.md` § 10a).\n\n## 6. The words, in order, for his ruling\n\nEach is the shortest phrasing found that is true of what it is shown on (P26) and readable without the notes (P28).\n\n1. **Section heading**: *Before its reasons* (replaces *Where its prior comes from*). Alternatives: *Where it starts*;\n   *Starting point*.\n2. **The sentence**: *Starts at 52%: of 222 published findings counted, 115 replicated.* (form of 4.1, with its\n   pooled, outdated and range variants). Alternative: *52% of findings like it replicated (115 of 222).*\n3. **The door to contest belonging**: *Is it one?* (replaces *argue the filing*). Alternative: *Not one of these?*,\n   which invites only the attack.\n4. **The outdated note**: *(counted before the wording changed; to be read again)* (replaces *reworded since, awaits a\n   new reading*).\n5. **The door to propose another**: *Belongs elsewhere?*, shown only once the route and search exist.\n6. **A claim with no reference class**: nothing shown (extends his 2026-10-24 word on hiding *unexamined*).\n   Alternative: *No count yet.*\n7. **The membership claim's own sentence**: *\"…\" is one of: published empirical findings.* (replaces *The claim \"…\"\n   is about one of: …*). This is the template new membership claims are minted with; none exists yet, so changing it\n   now rewrites no stored text (AL10 forbids rewriting a minted claim), and it is his call whether a machine's\n   template counts as the interface's voice.\n8. **The reference class page's heading line**: *A count other claims start from.*\n9. **Its sections**: *Counted from* · *Within it* · *Claims that start here*.\n10. **Its folded audit**: *How it was counted*, with the four lines of 5.1.\n11. **Its wheel's half names**: *Members* above, *Counts* below (one word each, AR11).\n\n## 7. Ledger rows (Arc BK, *prior doors*)\n\nBK1 the claim page's sentence and doors (§ 4.1, § 6 items 1 to 6) · BK2 a person's proposed reference class: a\nmembership claim that moves no prior until read (§ 3.4) · BK3 the reference class's wheel, members above and counts\nbelow, unexamined memberships at 60% (§ 5.2) · BK4 the membership claim's template reworded before any exists (§ 6 item 7). All open.\n\n## 8. What was not checked, and what is owed\n\n- No mockup was rendered; § 4 and § 5 are text. A rendered version follows his ruling on the words, under P28.\n- The seven live filings carry text grounds and no membership claim; the door in 4.1 shows on none until they are\n  minted (`mint_membership_claim`, no model call). Filed in `TODO.md`.\n- A route listing a reference class's members and bundles for the page does not exist (`memberships` and\n  `class_count` exist in `deliberus/reference_classes.py`; no route serves them).\n- Whether production holds the seeded reference classes was not checked; nothing here is deployed.\n\n## Cross-references\n\n`saying-what-kind-a-claim-is.md` § 8 to § 10 · `priors-as-standing-work.md` § 9a, § 10c, § 11 ·\n`points-of-entry-review.md` (Arc BJ, one home per act) · `docs/ux-principles.md` P22, P23, P26, P27, P28 ·\n`specs/map-view/design.md` § 3a, § 10a · `deliberus/reference_classes.py` · `deliberus/api.py`\n`GET /claims/{claim_id}/membership`.\n"}