{"path":"research/corpus-graph/decision-clusters.md","content":"# Decision clusters: what readers found about the pages, grouped by the word that settles each\n\n**Date**: 2026-10-11 · **State**: Claude's grouping, for the founder to go through. Nothing in it is ruled\nexcept where a ruling is quoted.\n\n**His words**: *\"Yes, hold any more readers, and group items into decision clusters, then we shall go through\nthem, thoughtfully. Think deeply about how to cluster.\"*\n\n## 1. What was grouped\n\nAbout 200 things were waiting for his word, in three places:\n\n- the 136 remarks and findings on [the page of what readers said](what-readers-said-that-is-no-pair-of-notes.md)\n  (seven readings of ten pages written for visitors on 2026-10-11, and the pilot on four notes);\n- about fifty page sentences on the earlier list in `TODO.md` § the corpus graph, of which nine are done;\n- 23 pairs of notes kept in the index today whose one end is a sentence on a page written for visitors.\n\nMany are the same sentence found two or three times by different readers. Counted by sentence, there are\nfewer than 200. Counted by the word that settles them, there are 46.\n\n## 2. How they were grouped\n\n**A cluster is the set of things one answer from him settles.** Not a page, and not a kind of fault. A\npage holds sentences that wait on five different answers, and one answer often reaches six pages. So the\nquestion asked of each item was: *what would he have to say for this to be finished?* Items with the same\nanswer share a cluster.\n\nThat question sorts the items into seven groups, and the groups differ in what they ask of him:\n\n| Group | What it asks of him | Clusters | Items |\n|---|---|---|---|\n| **A. Exposure** | a ruling on what may stand on a public page about people and machines | 4 | 8 |\n| **B. What kind of page each is** | a rule, said once, that decides the form of many repairs | 5 | 18 |\n| **C. Ruled already** | nothing new: one blessing for each ruling of his that pages have not caught up with | 9 | 55 |\n| **D. Stronger than the record** | a wording, with the source beside it | 5 | 36 |\n| **E. Open positions** | a position he has not taken yet; each is a ledger row or becomes one | 15 | 48 |\n| **F. Facts** | a wave-through: a date, a number, a name, a link | 5 | 36 |\n| **G. Claude's, no word needed** | nothing; listed so that he can see them leave the queue | 3 | 7 |\n\nThe last column counts the entries named under each group: every one of the 136 tagged remarks once, the\n23 kept pairs, and the places where the earlier list is cited (a `{list}` mark, which sometimes stands\nfor two or three sentences of that list). `remarks.py --unclustered` prints the same count from this file\nand says which remark, if any, stands in no cluster or in two. On 2026-10-11: none.\n\n**Why this cut and not another.** Three were weighed and set aside. *By page* is how the repairs will be\nmade, and it hides that one ruling recurs on six pages. *By the reader's label* (state, overtaken,\ncounter-case) names the symptom. *By how strong the finding is* puts the easy and the deep in one line.\nThe cut by settling answer has one more use: **groups C and E are the open reviews' work from the other\nend.** A cluster in C carries a ruled ledger row to the pages that still state the earlier version. A\ncluster in E is an open row, or a position nobody has filed as one. So a word given here is a word given\nin the ledger, and the other way round.\n\n**Naming the row** (added 2026-10-12, when the picture of the corpus graph was found to show the\ndecisions and the ledger as two separate queues). Where a cluster in C or E is a ledger row, write the\nrow beside it: in the heading after the word *ledger*, as E2, E3 and E15 have it, or on a line of its\nown, `**Ledger.** Q33, B10`. The copy of the corpus graph reads both and draws the cluster and the row\nas one position; a bare `B1` is not read, since that is also a cluster here. On 2026-10-12 three of\nthe 24 name their row. The other 21 are named as each cluster's record is gathered, and a cluster in E\nthat is no row yet says so (`**Ledger.** none yet`).\n\n## 3. The order proposed, and why\n\n1. **Group A first.** What stands on a public page about named people and about the machine the site\n   runs on is the one kind of fault that cannot be taken back once read.\n2. **B1 next: the rule for saying what is designed and not built.** It decides the form of about thirty\n   repairs in other groups, so every later cluster is shorter for it.\n3. **Group C**: nine blessings, each with its sentences listed. Fast, and most of the queue.\n4. **Group D**, family by family.\n5. **Group E**, in the order of the open reviews, or woven into them: these are the deep ones and\n   nothing sharp waits on them, with one exception below.\n6. **Group F** last, as one wave-through.\n\n**One clock.** The call for funding that closes on 31 October quotes the positioning page and the\nvision page. The clusters that touch those two pages are marked ◆. If the application is written before\nthe walk-through reaches them, they come forward.\n\n**How a cluster is put to him.** One a turn. The question in plain words, what Claude proposes, the\nsentences it would change. Before a cluster is put, each of its sentences is read in its section and its\nrecord gathered (`repair.py record`), as his word of 2026-10-10 asks. A tag like `[ux-open O3]` names the\nreader's own entry on the page of remarks, where the passages are quoted in full.\n\n---\n\n## A. Exposure on public pages\n\n### A1. Operations detail on the public page about the built system\n**The question.** Does the page that describes the built system keep saying, on the public site, which\nmachine it runs on and how that machine is reached?\n**Proposed.** Split it: a public half on how the system is built, and an operations half kept private.\nBilling state goes with the private half.\n**Members.** The page moved out of the instruction file, which is not served, into the served folder,\nand became the most infrastructure-revealing page on the site `[arch-open O11]` · a Swedish passage that\ncorrects text no longer on the page, and says billing state does not belong in a published document,\nwhich lines 45 and 300 still carry `[arch-open O3]`.\n**RULED 2026-10-11**, founder: *\"Ok sure split the server info into public/private, makes sense\"*.\n**Done the same day.** The page keeps what the system is built from. Which machines it runs on, how they\nare reached, how it is deployed and watched, and where files sit on the keeper's drives are in the\nprivate operations note, taken out word for word (fourteen blocks). The same detail stood quoted in ten\nother served notes, two of them reader reports filed that day; it was taken out of those too and kept\nprivately. A search of the notes for the host's name, an inside address or the database command now\nfinds nothing. Three trial scripts in the same folder still connect by the host's name; the site serves\nthe notes and never the scripts. Not deployed yet: the site serves the earlier page until the next deploy.\n\n### A2. Telling signed-in readers what is recorded about them\n**POSTPONED 2026-10-11**, founder: *\"Not a priority to deal with rn, postpone it until we have more than\n50 users or whatnot.\"* So nothing is said on the site for now, and the two options below wait. What\nwatches the number: from the fifty-first person on, the notice he gets for each new sign-in says this\nis due (`POSTPONED_UNTIL_MORE_PEOPLE_THAN` in `deliberus/notifications.py`; built, not deployed).\n**The question.** A signed-in reader's page views are kept for ninety days with their sign-in address,\non the public commons as well as in a community's own space, and nothing on the site says so. Should\nit?\n**Proposed.** One plain sentence where a person signs in, and one on the page about the built system.\nHis ruling that the log exists and that only he sees it stands as it is.\n**Members.** `[arch-open O10]`.\n**The record, gathered 2026-10-11 before the cluster was put.** What is kept: for each visit by a\nsigned-in reader, their sign-in address, the page and the time, for ninety days, on the public pages as\nwell as inside a community's space (`deliberus/reading.py`); the keeper alone sees it (his ruling of\n2026-09-27: *\"Read log should only be visible to me\"*), and the keeper is also notified when a signed-in\nperson reads a document. What a reader is told: nothing. The app has no page or sentence about what it\nkeeps (searched for it in the app's own files). Each community's written rules say that reads are\nrecorded, and no page shows those rules to a member. Nobody but the keeper is a member of a community's\nspace yet, so no member has read unknowing; anybody else who signed in and read the public pages since\n2026-09-22 did (how many was not counted). Outside: the\nEU's data protection regulation, Article 13 (official text, read 2026-10-11 in search excerpts and not\nin full) has the person told at the time of collection what is kept, for what purpose, for how long and\nwho receives it. Not checked: whether a non-commercial site kept by one person falls outside it.\n**Put to him as two options**: one sentence under the sign-in button and one on the system page; or a\nshort page on what the site keeps about a person (name, picture and address from the sign-in, the name\non what they contribute, the read log, the notifications), linked from the sign-in button and from each\ncommunity's space, each sentence his to approve. Claude's pick: the page.\n\n### A3. The unlisted community page, named by its files\n**The question.** The rule said an unlisted page is never named in the served notes. The files that\nbuild it carry the community's name, and several notes say the name. Rename the files and take the name\nout, or say the rule as what it can keep?\n**Members.** `[arch-open O12]` · `[arch-fit F13]`.\n**RULED 2026-10-11: the name may stand, and the page stays open.** Put two options, he leaned to the\nfirst (*\"Leaning toward 1.\"*) and asked whether the page could sit behind a sign-in for the community's\nmembers. Told it was feasible, he said to build it, and a minute later, before anything was built:\n*\"Or wait, actually hold off on that, I might wanna show this to various ppl in the … community that\ncan't be bothered to log in first, and it's still a bit of security by obscurity and it's not very\nsensitive content, so it's fine for now, can cross this bridge at a later date...\"*\n**Done the same day.** The rule is said as what it keeps: an unlisted page's address is never written in\nthe served notes, as a path or as a link. The test that guards this now covers the second community's\npage as well. No file was renamed and no note changed.\n**Held, his to reopen: a sign-in gate for the page.** Feasible, designed, not built. What the page\nholds, why a gate needs more than a prompt, and his words in full are kept privately.\n\n### A4. Named people on public pages\n**The question.** Three passages name or describe people. Each is its own small ruling.\n**Members and what is proposed for each.**\n- The page on community and outreach, five parts still to go, one a turn by his word (*\"Let's go\n  through these one by one\"*): the table of people who commented, the names his own quotes were said\n  to, the contact board of 2013–14, and two points of fact. Next is the people table. {list}\n- A paragraph on the UX page about a dispute between two physicists, with unsourced claims about both,\n  where ledger row J16 closed on *the anecdote stays out of every doc* `[ux-open O13]`. Proposed: cut\n  it, keeping the text privately.\n- Closed, for the record: the author of the public Swedish post stays named (founder, 2026-10-11:\n  *\"The FB post name is fine\"*) `[vision-open O14]`; the note on his former co-founder's company stays\n  (2026-10-10: *\"I don't mind this. Next.\"*).\n\n---\n\n## B. What kind of page each is\n\n### B1. How a page says what is designed and not built ◆\n**The question.** Many pages speak of a design in the present tense, as how the product works. What form\ndoes the truth take on a page written for visitors?\n**Proposed.** Three forms, by the kind of page. A page of principles or reference (UX principles, the\ndesign guide, bridging's *how it works*, the system page) carries one status word beside each principle\nor feature: *built*, *partly*, *design*. A page of vision keeps its future and says *will* and *is\ndesigned to*. A page that records an early design (the technical direction, the object model, the\nconceptual threads, the survey of inductive reasoning, the sketches) gets one dated line at its head and\nis otherwise left as it was.\n**Why it comes early.** About thirty items below take their form from this answer.\n**Members.** Four UX principles read as how the product works and are not built `[ux-open O7]` · the\nrule *no gap without a move* stated for every surface, where the maps show gaps and offer none\n`[ux-fit F4]` · concept states said to tell the reader how a word is settling, where the measurement\nfound they repeat the detector's own template {kept: ux F9} · the depth page: *taps, not essays* and a\nno-fit answer always available, where the claim page shows every question at once and wants typed text\n`[pilot-fit F2]` · the vision page: the ranking rules *standing in the graph as challengeable claims*\n`[vision-fit F6]` · analysis and attunement: mediation here is *itself measured*, where the attunement\nside is a design and a proposal `[concept-open O13]` · the technical direction's section on a database\nchoice that was not made {list} · the survey's machine probability for each step {list}.\n**After his word.** Claude writes the status words and the dated lines and brings them back as one list\nto bless.\n\n### B2. Whose voice the vision page speaks in ◆\n**The question.** On the vision page *I* is Fredrik in some sections and Claude in another, and its\nMarch, July and August layers disagree without saying so. Whose page is it?\n**Proposed.** His voice throughout. The story of how the page was drafted moves to a research note.\nThe March section on three horizons is marked as overtaken by the August section on the gate.\n**Members.** `[vision-open O1]` · `[vision-open O7]` · a history page that calls itself an internal\nreference while being published {list}.\n\n### B3. The design guide: a living description, or retired\n**The question.** The guide describes a landing page, widths, transitions and a graph the code no longer\nhas. Keep it true by hand, tie it to the code, or cut it down?\n**Proposed.** Keep the rules and the colour names, checked against the code by a test. Drop the\npage-by-page layouts. Add a short section on touch later.\n**Members.** The colour names are promised as the one source and are written out by hand 58 times in the\ncode `[ux-open O16]` · no word on phones, though the UX page calls the phone the main case\n`[ux-open O18]`. The guide's stale numbers are in F2.\n\n### B4. The system page: rules that live only there\n**The question.** Three rules written as *must* exist only on the system page, which a session opens\nwhen it already knows where to look. Move them to the lines every session loads?\n**Proposed.** Give each a test that reads the code, as was done for the others, and move none. The\nloaded file is on a fixed budget.\n**Members.** `[arch-open O9]`.\n\n### B5. The UX page's own order\n**The question.** None. Listed here because it is structure, not fact: the principle the project calls\nits most important has no heading of its own, and a principle marked *candidate* opens on a ruling.\n**Proposed.** Give the inversion principle a heading and a row in the page's table. Say in the heading of\nprinciple 24 what is still unruled, and put it before 25. Call anti-pattern 9 by that name wherever the\npage says *principle 9*.\n**Members.** `[ux-open O4]` · `[ux-open O9]` · `[ux-open O11]` · the status line that says no visual\ndesign is prescribed, beside a principle that prescribes a colour `[ux-fit F2]`.\n\n---\n\n## C. Ruled already: pages that have not caught up\n\nEach cluster names his ruling, then the sentences that still state the earlier version. He blesses the\nwording. No new position is asked.\n\n### C1. Quality is computed and never voted; relevance belongs to the link\n**His rulings.** 2026-03-31: one vote, agree or disagree. 2026-09-09: *\"Your vote never changes a\nclaim's strength; only the arguments around it do.\"* 2026-09-28: relevance sits on the link and is never\nput to a vote.\n**Members.** The UX page's first-use path asks a newcomer to rate quality, and two more places on it\n`[ux-open O3]` `[ux-fit F8]` · the threads page, three passages `[concept-open O2]` · analysis and\nattunement: two-axis voting as *the strongest defense* {list} · the vision page: *two-axis evaluation*\n{list} · the technical direction and the object model: relevance as an axis people vote on {list}.\n**One real choice inside it.** What does the newcomer do in that step in place of rating? Proposed:\nanswer one critical question.\n\n### C2. The horizon words\n**His rulings.** 2026-08-25: *horizon* replaces *residue*. 2026-09-15 and 16: a kind is read, never\nset; no verdict is minted.\n**Members.** The depth page, lines 99, 120 and 210 `[pilot-open O6]` {list} · the positioning page:\n*residue map* {kept: vision F19} · the convergence page presents two things as open that are closed\n`[concept-open O8]`.\n\n### C3. No act of ratifying; the machine writes marked parts\n**His rulings.** 2026-08-31: there is no ratify act. 2026-09-21: the splitter writes what it found,\nmarked, and a person keeps, rewords or drops.\n**Members.** The depth page, item 1 `[pilot-open O5]` {list} · the UX page: principle 14 still under\nits withdrawn heading, with the frame he rejected twice `[ux-open O6]` {kept: ux F7} · a note on the\nsame page that presents a withdrawn headline as the proposal on the table `[ux-fit F12]`\n`[ux-open O12]` · the confirmation step, which does three jobs on that page where his rulings kept one\n`[ux-open O10]`.\n\n### C4. What the strength arithmetic does today\n**His rulings.** 2026-08-31: untagged parts are governed by the weakest part. 2026-09-17: a part caps\nits whole. Measured 2026-09-27: one answer moves a claim by under 0.0005.\n**Members.** The depth page says the fix is inert, and says the opposite eight lines on\n`[pilot-open O4]` `[pilot-fit-haiku F1]` `[pilot-open-haiku O11]` {list} · the depth page on the check\nthat rebuilds a parent `[pilot-fit F4]` · the UX page: the neutral band was set on a figure no answer\ncan produce {kept: ux F11}, and *the answer-a-question loop works exactly as described* {list} · the\nthreads page: answering questions shifts badges {list} · the page of open problems: figures computed\nbefore both rulings, and evidence *is discounted* where the discount is built and not adopted {list}.\n\n### C5. The reuse flywheel is paired, not cumulative\n**Measured 2026-08-29**, on the old linker: what a new source connects to does not rise with the size of\nthe corpus. It follows whether an opposed source is already mapped.\n**Members.** The threads page: *validated*, *denser over time* {list} · the UX page's tenth principle\n{list} · the civilizational page, line 80 {kept: vision F13}.\n\n### C6. A link's number is to become a computed reading ◆\n**His ruling.** 2026-09-28 (*\"Yes.\"*): link strength shown to readers becomes a computed reading. Not\nbuilt.\n**Members.** The positioning page: *never by a model's opinion* `[vision-open O15]`\n{kept: vision F18} · the system page lists the stored number as a plain field, with no mark that it is\nruled to change `[arch-open O14]` · the bridging page calls the question his undecided call\n`[pilot-fit-haiku F2]` {list}.\n**Proposed wording until it is built.** The badge is computed from answered questions. The number beside\na link is a rating the extraction model wrote.\n\n### C7. The wager, and what is known ◆\n**His words.** The Founding Wager on the vision page, and 2026-10-07: what divides two people can be who\ngets the word.\n**Members.** The convergence page opens on shared bedrock stated flat {list}, and its table from Lean\nputs shared values where the checker belongs `[concept-open O9]` · the lenses page: depth *reveals*\nconvergence `[pilot-fit-haiku F5]`, *the premises are shared; the weighting is where you diverge*\n{list}, and its first example is a word people fight over `[pilot-open O11]` · the civilizational page:\nwhich sections ride on what he is sure of and which on what he bets `[vision-open O8]`, and no word\nthat understanding leaves a conflict of interests standing `[vision-open O9]` · the vision page, line\n311: most disagreement lives in definitions {kept: vision F7} · open problems: *small, nameable, and\nlargely shared* {list}.\n\n### C8. Structure is output at the door ◆\n**His ruling.** 2026-10-10 (*\"Blessed as you proposed\"*): the promise holds where a person speaks or\npastes, and the descent is still built by hand.\n**Members.** The vision page, lines 345 and 393 `[vision-open O2]` · the positioning page: give it *a\nquestion* and it extracts into the shared graph, where a question is answered by reading the graph\n{kept: vision F17}.\n\n### C9. Smaller rulings, one line each\n- **One image for the lens** (2026-09-16): the older *glasses you'd never choose* stands on two pages and\n  in one tile `[vision-open O10]` {kept: vision F9}.\n- **Nothing needs permission to exist** (2026-08-27): the system page announces reputation-gated\n  submission as the plan {list}.\n- **The exchange format is not conformed to**: the page of academic foundations calls its node split\n  *the right abstraction* {list}.\n- **No scoring of a contributor's rationality**: the survey of inductive reasoning proposes it twice\n  {list}.\n- **A cost the ledger withdrew** (2026-08-27): analysis and attunement still sends the reader to it\n  `[concept-open O16]`.\n- **The split between is and ought was published as a claim** (2026-08-22): the threads page calls it\n  invisible `[concept-open O3]`.\n- **The pointing signal** was re-scoped to a null across a set of interfaces named beforehand: the\n  bridging page and the UX page's principle 20 still make one session decisive {list} {kept: ux F10}.\n- **An engine he called outdated** stands on the history page; **the election pool** stands on the\n  bridging page after the election {list}.\n\n---\n\n## D. Stronger than the record\n\n### D1. Why the predecessors and the prototype failed ◆\n**The question.** Five pages give one cause each, and not the same one: typing the structure, thin\nexperience, only decomposing, the missing pole. The project's own notes count several, and one note\ngives audience and channel. What is the sentence?\n**Proposed.** *Input friction was the cost that could not be removed then and can be now. It was one of\nseveral.* And the four outside arrivals of 2026 stay as illustrations, without the word *retrodiction*.\n**Members.** The vision page, line 201 `[vision-open O5]` {kept: vision F4} {list}, and its four\narrivals `[vision-open O6]` · the UX page's first conviction: the graveyard *proves*\n`[ux-open O2]` {kept: ux F3} · the lenses page and analysis and attunement {list}.\n\n### D2. *Every*, *never*, *cannot*, *no precedent* ◆\n**The question.** None beyond wording. Each sentence is narrowed to what holds.\n**Members.** The vision page: every badge asserted the opposite {kept: vision F2} · the civilizational\npage: a fake citation cannot be made {kept: vision F16}, no tool has ever tested the thesis\n{kept: vision F15} · the positioning page: *none of it keeps itself honest* `[vision-open O16]`, the\ntagline `[vision-open O17]`, *no precedent at all* {list} · the competitive landscape on assembly\ntheory's one prediction, and on copying as the sign of success {list} · open problems: *nobody has\nbroken* the circle of a model checking a model {list}.\n\n### D3. Studies cited for more than they show\n**The question.** None beyond wording. Each is restated as what the study found, with its sample.\n**Members.** Analysis and attunement: the canvassing result as sustained exposure to reasons\n`[concept-open O15]` · the bridging page: *the most productive state for opinion change*\n`[pilot-open O9]`, a statement-level result used for arguments `[pilot-fit F7]`, *validates* and\n*computationally approximates* {list} · the civilizational page: the evidence for *I see you*\n`[vision-open O11]` · the convergence page: its lead evidence, never placed at a level\n`[concept-open O6]`, and seven structures *across all* of 256 societies {list} · the depth page:\n*research confirms* that decomposition is *computationally necessary* `[pilot-open O3]` · the lenses\npage: a profile from dilemmas said to capture how a person decides `[pilot-open O13]` {list} · the UX\npage's banner: *faster triage* `[ux-open O1]` · academic foundations: *proves* 84%, and opposing views\nas settled {list} · open problems: moral foundations {list}.\n\n### D4. Counts that should say whom they count\n**Members.** *Shared by 80% of humanity*, where the graph counts whoever contributed `[vision-open O12]`\n· the outreach page: *deployed and used by at least some of these organizations* {list}.\n**Proposed.** The principle is his already: every published count labels where it comes from.\n\n### D5. Lines every session loads\n**The question.** Six lines in the instruction file are stronger than the corpus's own record. Each\nchange to that file is his to bless, one at a time.\n**Members.** The two lines on reasoning skill and polarization, where one large replication did not\nfind it {list} · *polarization has no structural answer* {list} · *never ask the model* about sameness,\nwhere the built check asks a model about each pair {list} · the success prior's line on a flat phase,\nwhich names nothing that would count against it {list} · assembly theory's one prediction {list} ·\nimplicit premises as already the project's most important measure of quality, on one pair of texts\n`[pilot-open-haiku O12]`.\n\n---\n\n## E. Open positions\n\nEach is an open ledger row, or a position nobody has filed as one. Where a row exists it is named.\n\n### E1. What would count as the wager lost ◆\n**The question.** The pages say the wager can be lost. Nowhere is there a number, a depth or a cut-off,\nand the reading counts leaves, so cutting finer lowers the fraction with nobody converging. The same\nholds for the gate on the vision page, which lifts when a rate *falls*, and for the positioning page's\ncollective measures.\n**Members.** `[concept-open O7]` · `[vision-open O19]` · `[vision-open O4]` · the instrument the vision\npage cites counts words in commit titles, once for the whole period `[vision-fit F5]`.\n**Proposed.** Write the numbers down before the first live fraction is read, as a ledger row.\n\n### E2. Every step written out, or on request (ledger AD13)\n**Members.** The depth page, line 224 `[pilot-open-haiku O6]` {list}.\n\n### E3. The lens: first or second, the software or another person (ledger L9)\n**Members.** `[pilot-open-haiku O7]` `[pilot-open-haiku O8]` `[pilot-open-haiku O9]`\n`[pilot-fit-haiku F6]` `[pilot-open O14]` · a test that tells the lens from tourism `[pilot-open O12]`\n{list}.\n\n### E4. The depth page's imperative: invite, never press\n**The question.** *Never accept a stopping point* is unconditional, and the note it cites as its warrant\nis a list of limits on that very operation. By the headline tests it is over-broad.\n**Members.** `[pilot-open O1]` · the page's worked chain is hand-built and reads as a run\n`[pilot-open O2]` · the difference between showing where to look and showing what to conclude {list}.\n\n### E5. What the brake on sacred claims guards\n**The question.** The brake is called only inside the weighing detector. The note about it reasons as if\nit stood in front of every machine decomposition.\n**Members.** `[pilot-open O15]` `[pilot-open O16]` `[pilot-fit-haiku F7]` `[pilot-fit-haiku F8]`\n`[pilot-fit F13]`.\n\n### E6. The word *attunement*\n**The question.** It names a person's practice, a space of embeddings, and the acts kept for people.\nAnd five tests guard against analysis crowding it out, none the other way.\n**Members.** `[concept-open O11]` `[concept-open O12]` · whether the missing channel is a matter of\nreading what is stored, or of asking a person `[concept-open O14]`.\n\n### E7. What the bridging page promises, and how it is scored\n**The question.** The page promises the argument both sides find sound. Nobody rates reasoning. The\nscore peaks on a vote split of one against one, and reads the links leaving a claim.\n**Members.** `[pilot-open O7]` `[pilot-open O8]` `[pilot-open O10]` `[pilot-open-haiku O10]`\n`[pilot-fit F8]` `[vision-fit F14]` · three objections the page lacks, among them agreements that hold\nbecause their wording was left loose {list}.\n**After his word.** The direction of the links and a floor on voters are Claude's to repair.\n\n### E8. The first screen: one argument at a time, or a map\n**Members.** The UX page's banner of 2026-08-28 against the maps he asked for on 2026-09-20\n{kept: ux F1} · the design guide repeats the earlier rule `[ux-open O17]`.\n\n### E9. Must every notification name its move\n**Members.** `[ux-open O5]`.\n\n### E10. Which meaning owns green and red\n**The question.** Stance is voted, relation is typed, quality is computed. The palette gives all three\nthe same four colours.\n**Members.** `[ux-open O15]` `[ux-fit F5]`.\n\n### E11. First users who share the founder's cast of mind\n**Members.** `[concept-open O17]`.\n\n### E12. Four sentences that are his own ◆\n**Members.** The gate's second reason, that a corpus cannot be un-thought, beside the repairs since\nAugust `[vision-open O3]` · first takes kept in a store outside the graph, beside *one graph*\n`[vision-fit F8]` · the economic destination, his hope or the project's claim `[vision-open O13]` · the\npositioning page as the analysis side only `[vision-open O18]`.\n\n### E13. Three designs of March on the threads page\n**The question.** Decide each, or date the page as an early design (B1) and decide none now?\n**Proposed.** Date the page.\n**Members.** Voting together read as sameness `[concept-open O1]` · a record earned on facts and spent\non values `[concept-open O4]` {list} · a feed that aims at changing the reader's mind\n`[concept-open O5]` · structure as a reason for lower barriers to entry {list}.\n\n### E14. Making the unsaid explicit moves advantage\n**Members.** `[pilot-open-haiku O13]`.\n\n### E15. Rorty beside Peirce (ledger P11)\n**Members.** `[concept-open O10]`.\n\n---\n\n## F. Facts\n\nOne wave-through. Claude prepares each with its source and makes the change on his word.\n\n### F1. The system page\nThe contents line `[arch-open O1]` · the list of passes, which is not the pipeline that runs\n`[arch-open O2]` `[arch-fit F2]` · two absolutes its own later text contradicts `[arch-open O4]`\n`[arch-fit F4]` · a line marked open that the code closed `[arch-open O5]` {kept: arch F6} {list} · the linking step's count of pairs {kept: arch F8} · the record of model\ncalls and seven tables it never mentions `[arch-open O8]` · a link kind whose name does not say what\ntwo thirds of it holds `[arch-open O13]` · small faults of rendering and drift `[arch-open O15]` · 96\nschemes where the code holds 46 `[arch-fit F1]` · a pass said to be missing from a path that runs it\n`[arch-fit F3]` · a function name that does not exist `[arch-fit F7]` · a port `[arch-fit F9]` · five\nlink types where six pass `[arch-fit F10]` · *all GET routes are public* `[arch-fit F11]` · two\npassages overtaken `[arch-fit F14]` `[arch-fit F15]`.\n\n### F2. The design guide's numbers\n`[ux-open O14]` `[ux-fit F13]` `[ux-fit F14]` `[ux-fit F15]` `[ux-fit F16]` `[ux-fit F17]`\n`[ux-fit F18]` · 23 principles where there are 25 {kept: ux F19} · a tooltip about a score removed in\nAugust {list}.\n\n### F3. Dates, sources and links on the vision pages ◆\nA stay dated 2012 and January 2013 {kept: vision F1} · a profile dated six months late\n{kept: vision F11} · a figure credited to authors no note cites {kept: vision F12} · a link that\npromises a topic the target does not hold {kept: vision F10}.\n\n### F4. The history and reference pages\nA search that *returned nothing* stated in the present tense on a page that now holds the words\n`[ux-open O8]` · the former co-founder's present role, the paper listed as not found, the group's\nposts called unexplored, the self-check's description, the sketches' readings, an example far from what\nthe arithmetic gives {list}.\n\n### F5. The UX page's numbers\nDisplay limits of 3, 5 and 8 where the code has 3, 8 and 16 `[ux-fit F6]`.\n\n---\n\n## G. Claude's, no word needed\n\n### G1. Faults in the code\nStill open: the paid route's spending record `[arch-open O6]` · a reworded part saved over its earlier\nwording `[vision-fit F3]`. Repaired 2026-10-11: a web address looked up twice, once by the check and once\nby the fetch `[arch-open O7]` (each hop is now looked up once and the connection goes to the address that\npassed the check: `deliberus/safe_fetch.py`) · the read log `[arch-fit F12]` · failures that were printed\nand not logged `[arch-fit F5]`.\n\n### G2. Research notes his rulings have overtaken\nDated lines by `repair.py`, logged with his words: `[pilot-fit F12]` `[pilot-fit F14]`.\n\n### G3. Tests, so that these do not come back\nA contents line built from the headings · the colour names read from the code · a guard for home\nfolders written in the short form (eighteen public files).\n\n---\n\n## See also\n\n- [What readers said that is no pair of notes](what-readers-said-that-is-no-pair-of-notes.md): every\n  tagged item in the reader's words, with the passages.\n- [Reading for every kind of insight](reading-for-every-kind-of-insight.md) § 7b: the readings these\n  came from, and the predictions scored.\n- [README](README.md) § 14 and § 15: how this list relates to the open reviews, and how a sentence is\n  put to him.\n"}