2026-10-22-map-as-canonical

2026-10-22 · project workspace · path prompts/2026-10-22-map-as-canonical

Prompt prompts/2026-10-22-map-as-canonical/PROMPT.md

Prompt — the map as the canonical index of the program (2026-10-22)

You are working in the alignment-workspace repository, on main after the merge of PR #129 (commit 4eb420f). Do not stack this round on an unmerged branch; there is none.

This is a workspace round, not a research round: it produces no theorem and registers no claim. Its artifacts live under graph/ and in the specification-layer files it restructures. The prompt goes verbatim in prompts/2026-10-22-map-as-canonical/PROMPT.md and the report beside it. Check that no round directory with this date-slug exists before creating it. New priority items, if the round files any within its scope, start at 129.

One pull request. Auto-merge OFF; the merge is the maintainers'. The first commit on the branch is the proposal document of Part 8, so that it can be read on the pull request before the restructuring commits land beneath it. Run python3 -m checkers.graph --root . and python3 tests/run.py before every commit; a commit that leaves either failing is not made.

What this round is for

After this round, the structure under graph/ faithfully represents the whole research program — both maintainers' agendas as owned plans under the top frame, every registered claim as a formal result, every priority item as a leaf, every maintainer decision reachable from the nodes it touched, every narrative page attached to the frame it narrates, every round pointed at from the nodes it produced — so that a website can be generated from that structure alone. The website is not built here. The round's job is to make the map the one index of the program and to turn the duplicate human-facing files into renders of it or attachments to it.

Two principles govern every part.

Structure and numbers are decoupled. The frame tree, the and/or support, the records, bridges, hypotheses, cases tables, wants and task leaves are the structure. The judgment ladder, the probabilities, the utilities and the ranking are a lens on that structure. This round changes structure and touches no number: it writes no estimate, gestalt, judgment, acceptance or priority, and no event of any kind in a human's ledger.

Where the protocol cannot represent something, file the gap; do not force the content. graph/PROTOCOL.md §2 lists the node kinds and §9 the refusals. A document, an item or a ruling that has no honest place in that ontology is listed in the report under What the protocol could not represent, with the smallest protocol extension that would hold it, proposed as a graph/proposals/P-<n>.md where the mechanism fits and otherwise as a reserved line. Routing around the protocol silently is the failure AGENTS.md Structural defects are reported names.

Every path, field, kind, policy key and node id named in this prompt was read from main at 4eb420f. Confirm each exists before relying on it, and correct this prompt in the report where it is wrong.

Maintainer rulings

Record these as one dated DECISIONS.md entry as the round's second commit (after the proposal document), marked as taken by one maintainer in conversation with the other's confirmation outstanding.

M1. Handles. graph/policy.yaml maps the only human handle, @maintainer, to A-M-Berns, and no handle exists for abramdemski, although .github/CODEOWNERS and tests/path_gate.py name both as co-equal maintainers. The round renames @maintainer to @anson (the ledger file, its event ids maintainer-001/-002 → anson-001/-002, and every reference, with the design-log line recording the rename) and adds @abram → abramdemski, kind: human. The maintainer: key in policy.yaml is read by the checker as a single handle: the round reports every place the checker or the protocol assumes one maintainer and applies the co-maintainer reading where the change is contained (a list of handles for that key is the expected form), reserving the rest.

M2. One program, not two plans. The top frame's question already encompasses the whole program ("How does the relationship between humans and AI systems go, and what can we do about it?") and stays as it is. Under it there is one plan, not two: the existing plan frame trust-modelling is widened into the program itself — provisional id the-program, with a question of the shape "which formal account of how humans and AI systems learn from, correct and rely on each other guides AI safety engineering work" (the round words it from the wiki's Home page and the existing node's Statement, and lists it as provisional). Its roots are the lines, each a sub-frame:

normativity — what learning from reasons requires: statics, traderization, seed theory, the normative inductor as the end-to-end object (seed → reasons → compile → enforce → revise). Owner @anson. Programmes drafted per M3.
legitimacy — what must stay true of the process: integrity, openness, authorship; corrigibility hangs here as its consumer. Owners @anson, @abram.
deference-and-trust — the trust half of today's updateful-trust-li: the LI deference collapse, bounded agents, the conjectures and inter-inductor trust. Owner @abram.
decision-theory — the updateless programme and LI+BRIA (today's updateless-dt-programme, plus projects/decision-theory/). Owner @abram.
stance-and-level — the cruxes every line reads: updateful versus updateless, the computability levels (level-super-turing, level-above-quantifiable, level-poly-time stay as they are beneath it), and crux-unbounded-represents-phenomena with the *-guides-engineering family. Owner @abram. These are cruxes about where to work, not lines of work, and the frame's Statement says so.
abstraction-arc — as today. Owner @abram.
enabling-work — as today, extended with the map and the site. Owners @anson, @abram.

updateful-trust-li is split: its deference roots go to deference-and-trust; legitimacy-reasons-programme and normativity-line go under normativity and legitimacy as their Statements direct; roots that belong to two lines (corrigibility-landscape, deference-toward-bounded-agents and whatever else the round finds) are filed once and listed as roots of each frame that reads them — the graph is a DAG and a node may have several parents; the round does not duplicate a node to give each line its own copy. The frame updateful-trust-li itself is retired (its id kept as a deprecated alias in the report's old→new map, so that cited paths still resolve) unless the round finds a residue that is neither trust nor normativity, in which case it says what. Owner of the-program: @anson, @abram.

M3. Normativity's programmes are drafted from the wiki. The round drafts the sub-frames of normativity from wiki/Normativity.md, wiki/Normative-Inductor.md, wiki/Seed-Theory.md, wiki/Theorem-Spine.md and wiki/Roadmap.md (and legitimacy's from wiki/Legitimacy.md, wiki/Integrity.md, wiki/Openness-Coverage-and-Non-Capture.md, wiki/Corrigibility.md), attaching the existing nodes and the formal results of Part 1 beneath them. Every root the round files this way is marked owner has not confirmed in the view and listed in the report for a maintainer sitting.

M4. The map is the foundational layer, and it is allowed to move. The map's structure is the load-bearing index of the specification layer: a priority item, a registered claim, a decision and a narrative page each have exactly one node or one attachment in the map, and the human-facing files that duplicate them become renders of the map or attachments to it, per Parts 2–5. This is not a freeze of the protocol: where the program outgrows the ontology, the protocol changes through its own mechanism (a proposal, a design-log line, a Decided/Default register change) rather than the content being bent to fit, and the round's report lists every place it had to choose between the two.

M5. Frames may have several owners. A frame's owner may be a list of human handles (owners, or owner accepting a list — pick one form, document it in PROTOCOL.md §2 and in refusal rule 2, and apply it). A want keeps exactly one owner, because its priority word is one person's. The round reports every display and rule that read a frame's owner as a single handle and what it did with each.

**M6. PRIORITIES.md becomes a generated render of item nodes**, like graph/tasks.md. Item nodes are specification layer: tests/path_gate.py's SPEC_PATHS and .github/CODEOWNERS are extended to cover them (the two must agree). The maintainer's hand ordering of items is carried over as transcribed data on each node, not as a priority word, and the render preserves it.

**M7. DECISIONS.md stays the maintainers' prose record** and the highest-precedence source (README.md Precedence). It is not migrated into events: an agent writes no rule event and no line in a maintainer's ledger (AGENTS.md, What an agent writes only from its user's words). Instead each entry gains a stable anchor and a machine-readable touches line, and the map links to entries through a new work pointer kind decision. The round adds the machine lines only; it does not reword a ruling.

M8. The structure-only index is committed. The checker emits the nodes, edges, attachments, pointers and human events, without derived numbers, to graph/index.json on the plain run, committed like derived.json, so a pull request's diff shows what a round changed in the public picture. Its schema is the contract of Part 7.

Part 0 — Survey

Write graph/SURVEY.md (provisional name). One table, one row per human-facing or specification-layer surface: README.md, RESEARCH_STATE.md, PRIORITIES.md (the numbered items and the Where ingenuity is the bottleneck section separately), DECISIONS.md (Awaiting the author and Settled separately), PROVENANCE.md, CONTRIBUTING.md, AGENTS.md, each projects/*/CLAIMS.md, each projects/*/README.md and notes/, the consolidation and checkpoint trees listed in SPEC_PATHS, state/*.json and state/views/*, every page under wiki/, every round under prompts/ and projects/*/rounds/, and graph/ itself (frame, nodes by kind, events by handle, policy.yaml). For each row: what objects it holds and how many; what ids they carry; which other surfaces those ids join to and by what field; whether the surface is hand-written, generated, or hand-written with bindings; and what in the map already represents it (node ids or none).

Below the table, three lists.

Inconsistencies found, each with the two locations that disagree. Check at least: every ` binding against python3 -m checkers.workspace_state --json; every prose status claim in README.md and wiki/Home.md against the state (the README's line that the Deference line is paused on two maintainer decisions is one to test against the rounds of September and October); every registry answers_item against an existing item; every work: {kind: priorities} ref against an existing item and every work: {kind: round} ref against a round in the tree; every registry entry against a record node and the reverse (there are 82 record` leaves and the two registries hold about 80 entries — report the symmetric difference by id); every wiki link against an existing page or node.

What is represented twice, each with the surfaces that carry it.

What the protocol cannot represent, as described above.

Counts in this document are computed, not estimated, and the commands that produced them are given.

Part 1 — Formal results

Every registry entry in the registries list of policy.yaml has exactly one formal result in the map: an and of its record leaf (whose statement_of_record is the registry's declaration or checker invocation), one hypothesis leaf per hypothesis with its discharge tag, one bridge with its attribution, and the completeness leaf (PROTOCOL.md §2, Formal results). Reconcile the symmetric difference found in Part 0: file what is missing at level 0 with author the round's AI handle, and for a record that names no live registry entry, find the rename in the registry's old-id→new-id table and update the record rather than deleting it. Where a formal result has no hypothesis leaves because the filing session did not read the Lean, read the declaration's signature and file them, each tagged open unless the signature or the registry note settles it; a bridge filed this way carries attribution: formalizer-stated.

The 180 claims of projects/normativity/consolidation-aug9/ are a foundation under its own ledger and are deliberately not in the modern registry (projects/normativity/CLAIMS.md, first section). Do not file 180 nodes. File the foundation as one node whose statement says what the foundation is and whose work points at its ledger, attached under the normativity plan of Part 6; list in the report how many of its claims are cited by name from nodes that already exist, so a later round can decide which to lift.

Part 2 — Items

One node per numbered item in PRIORITIES.md: a task leaf for a work order (the item has a deliverable shape and an acceptance check) and a question leaf for an entry of the Where ingenuity is the bottleneck section. The node's Statement is the item's problem statement; its fields carry the item number, the difficulty tag, the workspace-priority marker's values (project, dispatchable), the deliverable shape, the acceptance check, the context paths, the item's open/closed status, and the item's position in the file as transcribed data (field name provisional, e.g. maintainer_order). Where an item node already exists (41 nodes carry work: {kind: priorities} today), merge into it rather than filing a second; the work pointer is then redundant with the node's own number, and the round decides whether the kind survives for pointers from other nodes to an item.

Attach each item node under the goal it serves, as a route or as evidence (AGENTS.md, What an agent does alone: a task hangs under its goal as a route or as evidence, never as a conjunct, unless the goal's Statement names the deliverable). An item whose goal the round cannot identify is attached under the plan frame of its project and listed in the report as goal not identified.

Then make PRIORITIES.md a render. The checker (or a new module under checkers/ the plain run invokes) writes it from the item nodes in the transcribed order, with the file's standing preamble and the ingenuity section preserved as rendered sections, and tests/run.py gains a gate that the committed file equals the render. Update SPEC_PATHS and CODEOWNERS together so that item node files are specification layer. Demand-gating is unchanged: a contributor still proposes items as issues, and a round files items only within its scope. This round's scope is transcription; it files no new item except as its own report requires.

Part 3 — Decisions

Give every entry under Settled in DECISIONS.md a stable anchor (an HTML comment or an explicit heading id; pick one convention and state it in the proposal document) derived from the entry's date and a slug of its title, and a one-line touches: list of the node ids, item numbers and claim ids the ruling bears on, in a form the checker can parse. Do the same for each entry under Awaiting the author, marked as open. The round adds these lines; it changes no word of any ruling.

Add decision as a work pointer kind in checkers/graph.py (its ref the anchor), extend refusal rule 2's kind list, and add the pointer to every node a ruling touches. Refuse an anchor that resolves to no entry, and warn on an entry no node points at. Record each checker change as a graph/DESIGN-LOG.md line.

Part 4 — Narratives

The wiki stays where it is (wiki/ is specification layer and the hosted wiki is force-pushed from it). Attach it. Add a narrative field to frames and, where a page is about one node, to that node (a protocol extension; document it in PROTOCOL.md §2 and in the refusal table — a narrative path that does not exist is refused). Every frame has a narrative or is listed in the report as lacking one; every wiki page is the narrative of something or is listed as unattached (ORIGIN.md, Sources.md, _Sidebar.md and Prior-Art.md may be legitimately unattached; say which and why).

Where a wiki page names a registered claim, an item or a round in prose, add the id in the form the state-bindings checker (checkers/wiki_state_bindings.py) or a sibling checker can resolve, so that a renamed claim or a closed item fails the build rather than going stale. Fix every binding mismatch and every prose status claim Part 0 found, in README.md and wiki/Home.md included, by binding the number or status to the state where a binding form exists and by correcting the prose where it does not; list each correction in the report.

Part 5 — Rounds

Every round in state/rounds.json is pointed at by work: {kind: round} from at least one node: the formal results it registered, the items it filed or closed, the frame it served. Today eight nodes carry such pointers and the state lists 125 rounds. Fill the gap from the registries' origin_round fields, the rounds' reports, and the items' context paths. A round no node can honestly point at (an administrative round, a scrub) is listed in the report, not forced onto a node.

Part 6 — The program and its lines

Apply M1, M2, M3 and M5, in that order: handles and the multi-owner rule first, because the frames of M2 need both. Widen trust-modelling into the-program; file the seven line frames with their owners; split updateful-trust-li and retire it; draft normativity's and legitimacy's programmes from the wiki; attach every existing frame, formal result and item under the line its Statement belongs to, filing shared roots once. Where an existing node's placement is unclear, leave it where it is and list it in the report under placement unclear rather than guessing. Wants for @anson may be filed only from his public words with filed_from: quote (the published corrigibility post; wiki pages are a maintainer's words — state in the report which source each want was drawn from); every one stays owner has not confirmed, and no priority word is written on any of them. Report every place the checker or protocol assumed one maintainer or one owner and what the round did about it.

Hold the AI-drafted frames to the same standard as any filing: every root's Statement is in research terms and cites only public sources.

Part 7 — The structure-only index

Add to checkers/graph.py a structure-only emission: every node with its kind, fields, Statement, Source, narrative, pointers and edges (parts, support, given, evidence, refinement), every frame with its roots and owner, every human event in every ledger, and no derived number, level, status or rank. Written on the plain run to the path M8 sets. Document its schema in the proposal document; it is the contract a site generator will build against, so every id in it is the id a URL will carry.

Part 8 — The proposal document

graph/MAP-AS-CANONICAL.md (provisional name), the branch's first commit, revised as the later parts teach. It states:

The ontology mapping — which existing object becomes which node kind, field or attachment (the table from Part 0 with the "becomes" column filled), and which files become renders.
The structure/numbers decoupling, and what the ranking is for once structure is canonical.
The governance of contribution: per-contributor ledgers owned by their handle through CODEOWNERS (one line per ledger), want nodes under a per-handle path the proof layer covers, and the checker as the only gate on either. State what this does and does not change about the two-layer trust model (AGENTS.md, The two layers): it creates no middle tier, because a contributor's own ledger is self-attribution checked for form, not trust.
The three contribution verbs as event schemas for a later round to implement, not this one: a challenge (against a record, a hypothesis or a bridge; a failure story with sources; becomes a proposed cases row or a proposed attack on the bridge), a want (already in the protocol), and a bridge judgment (a human says whether the formal statement means what the informal claim says, naming what they checked — the existing attribution event where it fits, a new kind where it does not).
The standing rules for a challenge, stated in the program's own vocabulary: a round that cites a challenge is a response and nothing more; answered is a checker act (a verified counterexample or a kernel-checked refutation landed) or a maintainer act; disposed is a maintainer act that names a successor; nothing else extinguishes a challenge.
What the website renders from the index of Part 7, as a list of page kinds keyed by id (frame, formal result with its bridge, item, decision, round, narrative) and three views (the frame tree, the support of one node, the feed of changes), with the rule that a page shows AI-drafted structure visibly lighter than anything a human event has touched.
Every choice this round made beyond its rulings, marked agent-decided and reversible, and the reserved list.

Reserved to the maintainers

Append to Awaiting the author in DECISIONS.md, each with its turns on line: the wording of the-program's question and of every line frame's; the roots marked owner has not confirmed; which of the foundation's 180 claims are lifted into the map; the priority words on every want filed in Part 6; and anything a part above explicitly reserves. M1–M8 were taken by one maintainer in conversation; the entry landing them says so, and the second maintainer's confirmation is listed as outstanding.

The report

prompts/2026-10-22-map-as-canonical/REPORT.md. In order: the verification section first — the OK line of the checker before and after, the tests/run.py output, the counts by node kind and by pointer kind before and after, the render gate passing; then deviations from this prompt with reasons; corrections to this prompt; What the protocol could not represent; goal not identified items; unattached narratives and unpointed rounds; every protocol and checker change with its design-log line and the refusal rule it added or changed; what this round did not do (the site, the intake verbs, the lifting of foundation claims, the reconciliation of numbers); and the reserved list. Model attribution trailer on every commit per AGENTS.md Provenance.

Report prompts/2026-10-22-map-as-canonical/REPORT.md

Report — the map as the canonical index of the program (2026-10-22)


Prompt author	the maintainer, relayed verbatim (`PROMPT.md`)
Executor	Claude Fable 5.1 (Anthropic), in Claude Code, one session
Deliverables	`graph/MAP-AS-CANONICAL.md`, `graph/SURVEY.md`, the restructured map (`graph/`), `checkers/graph.py`, `checkers/priorities_render.py`, `checkers/wiki_refs.py`, the rendered `PRIORITIES.md`, the anchored `DECISIONS.md`, the attached wiki, `README.md`, `AGENTS.md`, the gates
Dates	2026-10-22
Register	`ci-only`: the maintainers have read none of it; rulings M1–M8 were taken by one maintainer in conversation and the second maintainer's confirmation is queued

1. Verification

The checker, before and after (python3 -m checkers.graph --root ., Python 3.12):

before  OK 1561 nodes, 2411 events, 2021 ranked units; research first: the decision term (printed, not applied) favours vet hull-condition-total-trust (ranked unit 19), D 0.00016
after   OK 3059 nodes, 2411 events, 4685 ranked units; research first: the decision term (printed, not applied) favours vet hull-condition-total-trust (ranked unit 202), D 0.00016

No WARN before; after, one WARN line: DECISIONS.md: 241 of 291 decision anchors have no node pointing at them (first: …; every anchor's pointers are in graph/index.json) — the warning the prompt asked for, summarized to one line because one line per entry drowned the OK line (the index carries pointed_by per anchor). The self-test (--self-test) passes with the eleven null inputs this round added.

**tests/run.py**: ALL GREEN (80 project(s)) on the final tree (the unit tests take about four minutes: the fixture class derives the 3059-node graph several times, and the oracle regression is capped as §7 says), with the new gates in the run: PRIORITIES RENDER: the committed file equals the render (137 items), WIKI REFS: clean over 34 page(s); 304 reference(s) resolve, WIKI STATE BINDINGS: clean over 34 page(s); 3 binding(s) verified, WORKSPACE STATE: valid — 6 projects, 178 modern claims, 1 foundations, 180 foundation claims, 126 rounds, 52 terms, 137 priorities, 10 interface objects. The full output of the final run is in §9. Lean was not built (WORKSPACE_LEAN unset; no Lean file changed).

Counts by node kind (before → after): and 125 → 303; cases 47 → 47; frame 33 → 46; leaf 1291 → 2593; or 65 → 70. Leaves: assumption 32 → 32; bridge 85 → 263; completeness 125 → 303; hypothesis 68 → 714; partition 47 → 47; picture 202 → 202; question 188 → 192; record 82 → 260; route 123 → 123; step 211 → 211; task 116 → 234; want 12 → 12.

Counts by pointer kind (nodes carrying the kind, before → after): priorities 36 → 13 (the 17 merged item nodes dropped their own pointer; the 13 nodes that point at items they are not stay); open 18 → 18; round 8 → 194; decision 0 → 61; foundation 0 → 1; private 0 → 0.

The render gate: python3 -m checkers.priorities_render --check passes; the only change the first render made to PRIORITIES.md was placing every workspace-priority marker on the line after its heading (items 74–128 and F7–F8 had a blank line first).

2. Deviations from the prompt, with reasons

Formal results are 178, not 174, and the records matched were 4, not 82 reconciled. The prompt expected the 82 landed record leaves to be renames of registry entries. They are not: 78 name declarations of the clean room projects/deference/faf-cleanroom-2026-10-02/ and no registry lists them; the registries have no old-id→new-id table beyond the superseded_by field (11 entries). The 78 are kept as they were, flagged not in a claims registry, and the gap is filed (§4). Every registry entry, superseded ones included, got a formal result; a superseded entry's and carries superseded_by naming its successor's result.
Hypothesis leaves were read by script, not by a reading session. 511 hypothesis leaves and 170 setting leaves were parsed from the Lean signatures (explicit proposition binders, named h… or carrying a relation, as hypotheses; data binders as one definitional setting leaf). Binders declared with variable in a section are not in a signature and are not listed. Every hypothesis is discharge: open; the registry notes settle none in a form a script can read. Two and Statements and one bridge fell back to a generic sentence because the registry's note used a word on the evaluative list (kernel.extension-realized-rate, li.assessment.budgeter-value, kernel.box3-gap-exchange-rate.bridge); the note is quoted in the Source.
Items are docketed by project, not by section. Part 2 says an item whose goal cannot be identified hangs under the plan frame of its project; the file's late sections do not follow projects (items 112–128 sit under Deference line — second wave with normativity projects), so the dockets follow the workspace-priority marker. 120 items are docketed and listed in §5; the 17 merged ones keep their place.
**The touches lists are mechanical.** 283 entries could not be read for their bearing in one round; touches holds the explicit mentions (items, registry ids, backticked node ids) and 49 entries have one. A ruling's unstated bearing is for a later pass.
Frames were re-owned by line and wants by source attribution beyond M1's rename, because the blanket rename would have made the first maintainer the owner of wants quoted from the second's notes (projects/deference/note-dump-2026-08-11/README.md attributes notes/, the lab and the chats to him). Agent-decided, reversible, listed in DECISIONS.md Awaiting the author.
No new want was filed. Part 6 permits wants for @anson from the published corrigibility post; the thirteen existing wants already cover the post's material and the round filed none, so no priority word question arises beyond the thirteen.
**wiki/CONVENTIONS.md is attached** (to enabling-work) although the prompt lists only ORIGIN.md, Sources.md, _Sidebar.md and Prior-Art.md as possibly unattached: the conventions page narrates the wiki's own machinery, which is enabling work.
**The rulings entry does not backtick graph/index.json**: tests/dead_pointers.py refuses a backticked path that does not exist, and the index exists only from the third commit.
A WIP commit was used and squashed. The checker changes were snapshotted while the restructure ran, then folded into the third commit; the branch's history has the proposal first and the rulings second, as asked.

3. Corrections to the prompt

"the two registries hold about 80 entries" — they hold 178 (105 deference, 73 normativity; 167 active, 11 superseded).
"41 nodes carry work: {kind: priorities} today" — 41 pointers on 36 nodes, to 36 distinct items.
"Today eight nodes carry such [round] pointers" — correct (8 nodes, 8 rounds).
"every page under wiki/" — 34 files: 32 pages and two repo-only files (ORIGIN.md, CONVENTIONS.md).
M1 says "the ledger file, its event ids maintainer-001/-002" — correct; the two rulings on the valuation nodes are the only human events.
checkers/wiki_state_bindings.py binds quantities, not names; a sibling checker (checkers/wiki_refs.py) was needed for claims, items and rounds, as the prompt allowed.
The prompt's "a narrative path that does not exist is refused" is implemented; a path outside wiki/ is refused too.
Everything else named in the prompt (paths, fields, kinds, policy keys, node ids) exists on main at 4eb420f.

4. What the protocol could not represent

graph/SURVEY.md, What the protocol cannot represent, lists eight; the ones filed as extensions or reserved:

A kernel-checked declaration outside every listed registry (the 80 clean-room records): reserved (a registry for the clean room is a trust-chain question).
A work order's goal: the per-project docket is the protocol-conformant placeholder; no extension.
A ruling's effect beyond its explicit mentions: no extension; a supersedes field on the anchor line is the candidate.
An owner confirming a drafted frame: filed_from: wiki and its removal by the owner; a confirm event is the alternative, reserved.
The foundation's ledger as a pointer target: the foundation work kind (state/foundations.json ids), filed.
Two pages narrating one node: narrative accepts a list, filed.

No graph/proposals/P-<n>.md was written: each extension was small enough to land through a design-log line and the protocol's Default register, per M4; the reserved ones are queue entries.

5. Goal not identified

The 120 items in the dockets (an or of routes under the line's frame; the 17 merged items keep their earlier place):

docket-normativity (39, under normativity): 1, 3, 4, 5, 6, 29, 30, 31, 32, 33, 35, 39, 40, 46, 41, 44, 45, 43, 42, 47, 49, 50, 51, 74, 78, 79, 80, 81, 82, 85, 112, 113, 114, 116, 123, 124, 125, 126, 127.
docket-legitimacy (22, under legitimacy): 53, 54, 55, 56, 57, 58, 59, 61, 62, 64, 65, 66, 70, 67, 68, 69, 71, 72, 73, 88, 115, 128.
docket-seed (6, under seed-theory): 117, 118, 119, 120, 121, 122.
docket-deference (38, under deference-and-trust): Q1, Q2, Q3, Q4, 52, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 60, 83, 84, 86, 87, 89, 90, 92, 93, 97, 98, 99, 100, 101, 102, 103, 104, 107, 111, 28.
docket-workspace (15, under enabling-work): F1–F8, 10, 11, 12, 13, 36, 37, 38.

The 17 merged (old id → item-<n>): computable-coherence-modulus → item-2, market-and-traders-modeled → item-7, doubly-soft-weight-class → item-8, forcing-headlines-not-squeezes → item-9, fa-public-port → item-14, punishing-menu-vs-public-port → item-34, modified-market-computable → item-48, sigma-n-computable → item-63, closed-loop-affordability → item-75, semantic-transport-modulus → item-76, authorized-disposition → item-77, oracle-relativized-li → item-91, belief-decision-compatibility → item-105, li-bria-updateful-endpoint → item-106 (an and; its completeness leaf is item-106.complete), thin-spec-for-evaluations → item-108, thin-preservation-li-principal → item-109, thin-baseline-definition → item-110. Items with several candidate nodes (2, 27, 103, 106, 107) got a fresh node where no single node stood for the item alone.

6. Unattached narratives and unpointed rounds

Wiki pages that are no node's narrative (10): ORIGIN.md (a receipt), Sources.md (a bibliography), _Sidebar.md (navigation), Prior-Art.md (a field map), Glossary.md (vocabulary), Relation-to-the-Field.md (nine lines, a field map), legitimately; and Roadmap.md (it narrates every line at once; its item mentions carry references), Theorem-Spine.md (it narrates every formal result at once; 68 references), Why-Normativity.md (motivation, no one node) and Continuation-BRIA.md (the learning layer, whose nodes are the BRIA cluster inside updateless-dt-programme and public-corrigibility-line; no one frame is its subject). The last four are for a maintainer to place or to leave.

Frames lacking a narrative (30 of 46): the eleven new line and programme frames have one except decision-theory, stance-and-level and authorship-and-transparency; the 27 pre-existing sub-frames below the lines (abstraction-arc, legitimacy-reasons-programme, updateless-dt-programme, the three level frames, the corrigibility-landscape sub-frames, the stage frames, and the rest) have no page of their own.

Rounds no node points at (6): 2026-08-11-admin-cleanup, 2026-08-11-public-flip-and-scrub, 2026-08-11-rename-alignment-workspace, 2026-08-11-retire-frozen, 2026-08-11-scrub-round-2, 2026-08-24-cleanup-and-compress — scrubs, renames and cleanups. The 119 others: 23 workspace-prompt research dispatches are pointed at from the line or programme they served; 12 infrastructure rounds from enabling-work; the rest from the formal results they registered (origin_round) and the line frames of their project. 2026-10-02-graph-prototype is indexed with path graph and is pointed at by that path.

Placement unclear (left where it was): wiki-corpus-maintenance (a plan of deference-and-trust, arguably enabling work); the Progress results under normative-inductor (the sidebar files normative induction under legitimacy); deference-toward-bounded-agents and corrigibility-landscape, whose home is legitimacy because that line is walked first (each is also a root of deference-and-trust).

7. Protocol and checker changes

Each with its design-log line (all dated 2026-10-22 in graph/DESIGN-LOG.md) and the refusal rule it touched:

change	rule	design-log line
layout admits `MAP-AS-CANONICAL.md`, `SURVEY.md`, `index.json`; the scale test reads bytes on macOS	11	"the layout admits …"
`@maintainer` → `@anson`, `@abram` mapped, `maintainer` a list, `rule`/`survey` by any maintainer	5	"ruling M1 …"
`owner` a handle or a list on every frame	2	"ruling M5 …"
shared roots (a node in several frames; a want in one; roles agree)	1	"shared roots …"
`narrative` (a page or list under `wiki/`, must exist); `filed_from: wiki` on frames	2	"`narrative` on the top frame …"
`decision` and `foundation` pointer kinds; `round` resolves through `state/rounds.json`; `priorities` resolves to item nodes	1, 2	"ruling M7 …"
item nodes and their fields; `PRIORITIES.md` rendered by `checkers/priorities_render.py` and gated	1, 2, 12	"ruling M6 …"
`graph/index.json`, the fifth committed file	7 (its hash)	"ruling M8 …"
`the-program` and the seven lines; `checkers/wiki_refs.py`	—	"ruling M2 …"
the scale test derives in a child process; the fixture copier carries `wiki/`, `DECISIONS.md` and the state files	—	"the scale test derives its fixture in a child process …"
the oracle regression skips shipped roots above twelve proof-set variables (the warrant inductor's result has 25; the enumerator took 40 minutes on it)	—	"the oracle regression on the shipped graph skips …"

Eleven null inputs were added to the checker's self-test (a narrative naming no page, a narrative outside wiki/, a decision pointer to no anchor, a foundation pointer to no foundation, a want in two frames, a shared root with two roles, an AI handle in a frame's owner list, an AI handle in the maintainer list, filed_from: wiki on a leaf, an item node without its fields, an item- id without item); checkers/priorities_render.py and checkers/wiki_refs.py ship their own self-tests with null inputs; tests/path_gate.py gained three cases. AGENTS.md's graph/ section, graph/README.md and graph/PROTOCOL.md (§1, §2, §8, §9) say what the code does.

Where the checker or protocol assumed one maintainer or one owner, and what was done: policy.yaml's maintainer (now a list; rule 5); the rule/survey check and anchor 18 (any maintainer); the top frame's owner, every frame's owner, the view's "frame owner", the goal table's "owner" and derived.json's frame.owner (a list, joined for display, owners beside it); the prose "the maintainer owns the top frame" in PROTOCOL §2 (rewritten). Left as is: a want's one owner (by M5); placed_by on the outcome classes (one handle, the maintainer who placed them); the users presentation map (per handle already); audit cross-audit preference (per handle). The protocol's "the maintainer's design rulings are cited by number" (§7) stays single in wording and applies to either maintainer.

8. What this round did not do

The website; the intake verbs (challenge, bridge judgment) beyond their schemas in the proposal document; the lifting of any of the foundation's 180 claims; any reconciliation of numbers (no event, judgment, priority word or acceptance was written; every node filed is at level 0 with no belief); a registry for the clean-room records; reading the 283 ledger entries for their unstated bearing; wants for either maintainer; placing the four wiki pages left unattached; the per-sub-frame narratives.

9. Test output

self-test PASS
CONTRIB HYGIENE: 0 contributor checker(s), stdlib-only, each with a meaning docstring and a check function
CONSERVATIVITY: 4 spec files, no axioms, shape unchanged
WORKFLOW SCOPE: clean over 3 workflow(s); 2 enumerated write-scope job(s)
DEAD POINTERS: clean over 58 live document(s) — 918 rooted path(s) resolve, 8 of them into declared trees and each saying so; 105 unrooted and 221 glob(s) not this gate's business; 14 historical span(s) and 5 ledger cita
UNTRACKED POINTERS: clean — 333 round-record path(s) and 973 resolving citation(s) are in the repository, not merely on this disk
WORKSPACE STATE: valid — 6 projects, 178 modern claims, 1 foundations, 180 foundation claims, 126 rounds, 52 terms, 137 priorities, 10 interface objects
WIKI LINKS: clean over 32 page(s) (links resolve; repository links are commit-pinned)
WIKI STATE BINDINGS: clean over 34 page(s); 3 binding(s) verified against the state emission
WIKI REFS: clean over 34 page(s); 304 reference(s) resolve
PRIORITIES RENDER: the committed file equals the render (137 items)
Ran 140 tests in 148.556s
OK
GATE COVERAGE: 11 gate scripts, all invoked (5 self-test only — no local form)
LEAN SORRY GATE: clean over 108 files
LEAN AXIOM DISCIPLINE: every file carries `#print axioms`
PROJECTS:
LEAN BUILD SKIPPED: set WORKSPACE_LEAN=1 to run `lake build`
ALL GREEN (80 project(s))
  … 80 PASS lines, one per project runner, elided

10. Reserved to the maintainers

In DECISIONS.md, Awaiting the author, each with its turns on line: the second maintainer's confirmation of M1–M8; the wording of the-program's question and of every line frame's; the roots marked owner has not confirmed (nine drafted frames, the re-owned sub-frames, the re-owned wants); which of the foundation's 180 claims are lifted (four are cited by name, from the item-1 node); the priority words on every want; whether the clean-room records get a registry; the top frame's owner and the sub-frames' owners; the placements the round left where they were.

Outstanding maintainer actions

Confirm or strike rulings M1–M8 (DECISIONS.md, 2026-10-22) — the second maintainer, by a dated entry.
Read graph/MAP-AS-CANONICAL.md and merge or return the pull request (auto-merge is off).
For each drafted frame, keep or redraft it, and remove filed_from: wiki from the node file to confirm it: normativity, legitimacy, seed-theory, operative-force, normative-inductor, response-learning, normative-record, integrity-and-record, openness-and-non-capture, authorship-and-transparency.
Give each want its priority word (priority event in your ledger, through your agent), or disown it.
Decide the reserved placements and whether the clean room gets a registry.

Follow-up of 2026-10-23


Prompt	the maintainer, relayed verbatim (`PROMPT-followup.md`), answering a review of PR #131
Executor	Claude Fable 5.1 (Anthropic), in Claude Code, one session
Commits	six on `round/2026-10-22-map-as-canonical` above `86e3f35`: the rulings F1–F6; the records (F1, F2); the dissolution, the merges and the dockets (F4, F5, F6); the derived files (F3) with the checker, CI and the documents; the removal of the two files alone; this report with the queue entries
Register	`ci-only`; rulings F1–F6 taken by one maintainer in conversation, the second maintainer's confirmation appended to the round's queue entry

1. Verification

The checker, before and after (python3 -m checkers.graph --root ., Python 3.12):

before  OK 3059 nodes, 2411 events, 4685 ranked units; research first: the decision term (printed, not applied) favours vet hull-condition-total-trust (ranked unit 202), D 0.00016
after   OK 2415 nodes, 2411 events, 3389 ranked units; research first: the decision term (printed, not applied) favours vet hull-condition-total-trust (ranked unit 203), D 0.00016

The one WARN line is the round's (DECISIONS.md: 244 of 295 decision anchors have no node pointing at them …; 295 anchors now: the follow-up's entry, pointed at by the two line frames it restructured, and its three queue entries, pointed at by nothing). New since F3, one DIGEST line before the OK says what the digest compared with; on this tree: DIGEST compared with graph/derived.json on disk (the derivation versions.json at HEAD records): 0 status(es) flipped, 0 root status(es) flipped, 0 level(s) moved, 0 event(s) counted since, 0 revision(s), 0 want(s) filed since, 0 proposal(s) awaiting. The self-test passes with the two null inputs this follow-up added.

**tests/run.py** on the final tree: ALL GREEN (80 project(s)), with PRIORITIES RENDER: the committed file equals the render (137 items), WORKFLOW SCOPE: clean over 4 workflow(s); 3 enumerated write-scope job(s) and the unit tests green (two tests added, four extended for the DIGEST line, the aliases and the merged results). The full tail is in §10. Lean was not built (no Lean file changed).

Counts by node kind (before → after): and 303 → 305; cases 47 → 47; frame 46 → 46; leaf 2593 → 1945; or 70 → 72. Leaves: assumption 32 → 32; bridge 263 → 263; completeness 303 → 303; hypothesis 714 → 68; partition 47 → 47; picture 202 → 202; question 192 → 192; record 260 → 260; route 123 → 123; step 211 → 209; task 234 → 234; want 12 → 12. The 646 leaves gone are the 511 hypothesis and 135 setting leaves the round parsed (F1); the two steps gone became the and nodes of their formal results (F5); the two or nodes added are the dockets of F6; the two frames retired by F4 keep their files.

Pack size (git count-objects -vH, in the worktree, before the follow-up → after its six commits): loose objects 27.59 MiB (2,479 objects) → 39.48 MiB (3,113 objects); the pack 77.83 MiB in 10 packs, unchanged (no repack ran). The intermediate commits still carry graph/derived.json and graph/index.json (the prompt asked for the removal in a commit of its own after the content commits), so the pack grows by their blobs once more before shrinking at every later commit. Committed generated files after the follow-up: graph/versions.json 898,083 bytes (832,924 before; it now records the two derivation hashes), graph/tasks.md 202,461 bytes (203,367), graph/view.md 684,884 bytes (753,495); the two uncommitted ones on disk: graph/derived.json 11,634,243 bytes (14,769,185 before) and graph/index.json 3,498,828 bytes (3,719,006).

Old → new map (deprecated aliases, each file kept with superseded_by): normativity-line → normativity; legitimacy-reasons-programme → legitimacy; kernel.box2-finite-time → kernel-box2-finite-time; thin.value-of-thick → thin-value-of-thick (the two registry ids resolve through claims in the index and registry_id on the record leaf).

2. Part 2 — the moves (ruling F4)

Of the 27 pre-existing sub-frames below the lines, the ruling applies to two, which are dissolved: normativity-line (6 roots) and legitimacy-reasons-programme (77 roots, three of them frames). Its three sub-frames (formalism-open-questions-0-3, legit-dynamics-learning, legitimacy-meets-syntax) ask questions none of the three legitimacy programmes asks (the li-deference formalism's open questions; learning and repairing legitimacy; legitimacy as local rules of sanity), so each moves whole to the line frame as a plan root. corrigibility-landscape with its eight sub-frames and deference-toward-bounded-agents stay where they are (their questions are corrigibility's and the bounded agent's, asked by no programme). The thirteen sub-frames under deference-and-trust, decision-theory, stance-and-level, abstraction-arc and enabling-work are not touched, per the ruling.

The reading rule, applied to each child's Statement and Source: a child whose Statement is the programme's own object (the thin specification and the thick definition; the authorship clause; the traderization arc; the Progress bound) or whose Source cites the round from which that programme's results were registered goes to the programme; the maintainer's logical-induction-register legitimacy theory (the gated trust principles, the endpoint-preservation model, the clean-room results, the legitimacy-weighted objectives and their cruxes), the bridge between the two theories, the reasons thread and the wants bear on none of the three programmes' questions and become roots of legitimacy; the normativity line's foundation, consolidation problems and paper candidates become roots of normativity. Roles and presets carried over unchanged; the four items among the children (item-75, item-108, item-109, item-110) are roots of a frame, not of a docket.

node	kind	role	from	to	reason
`normativity-aug9-foundation`	leaf/step	shared-claim	`normativity-line`	`normativity`	the foundation of the whole line ("one hundred and eighty claims under its own ledger"); it bears on no one programme
`normativity-layer-I-closed`	leaf/step	shared-claim	`normativity-line`	`normative-inductor`	its Statement is the fixed-era composition "from obligation stream to Progress and substrate preservation"; the Progress bound (progress.*) is what this programme's results register
`traderization-arc`	leaf/step	shared-claim	`normativity-line`	`operative-force`	its Statement, "a constraint on prices compiles to a legal trading strategy in the market recursion, the modified market is a logical inductor over an assessment process", is the arc whose registered results (li.assessment., force., coherence., compiler., maxmin., region.) this programme holds; its own Notes say so
`normativity-learning-track`	leaf/question	shared-claim	`normativity-line`	`response-learning` (already a root there)	already a root of response-learning (a shared root since the round); the dissolution leaves it there
`normativity-consolidation-open-1-6`	leaf/question	shared-claim	`normativity-line`	`normativity`	the six open problems span the consolidation ("persistence of the certified core minimum, a computable coherence modulus … schema-level demand rates"); no one programme asks them
`normativity-paper-candidates`	leaf/step	shared-claim	`normativity-line`	`normativity`	"four extraction candidates exist for papers from the normativity line": a line-level claim
`legit-definition-endorsement`	cases	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (cases: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-corrects-trust-principles`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-per-formalism-family`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-gated-trust-li`	or	plan	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (or: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-ddb-transfer`	and	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (and: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-locality-forced`	and	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (and: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-reflection-unsat-modest`	and	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (and: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-composition`	and	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (and: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-li-self-correction-identity`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-correction-function-learned`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-condition-vs-predict-reality`	cases	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (cases: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-anchor-of-L`	cases	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (cases: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-global-or-local-form`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-trust-improvable`	or	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (or: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-endpoint-preservation`	leaf/step	plan	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`idleness-on-G-grade-a`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`idleness-on-G-with-certificate`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`claim1-diagonal-illegitimate`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`corrupt-strictly-contains-inadmissible`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`trace-nonrecoverability`	and	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (and: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`trace-nonrecoverability-li`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`wirehead-iff-finite`	and	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (and: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`certificate-enforcement`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`predication-dichotomy`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`d3-identification`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`corrigibility-faithful-plus-legit-on-self`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`non-modification-conjecture`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-v1-improvable`	or	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (or: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`public-legitimacy-theory`	and	plan	`legitimacy-reasons-programme`	`authorship-and-transparency`	its Statement is the public two-level theory, "a thin specification … and a thick definition …, and proves that the thick conditions satisfy the thin specification", which is this programme's question
`thin-converse-open`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`authorship-and-transparency`	its Statement is the thin level's converse ("sufficient is necessary for value"); its Source cites thin.converse-fails-on-correct, a result of this programme
`thin-value-loss-bound`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`authorship-and-transparency`	its Statement is the thin level's two-term bound; its Source cites thin.value-loss-le-defects, a result of this programme
`item-109`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`authorship-and-transparency`	its Statement is the thin preservation theorem for a logical-inductor principal; its context is the thin-legitimacy round from which thin.preservation was registered
`item-110`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`authorship-and-transparency`	its Statement asks for a definition of the thin level's baseline; its context is the thin-legitimacy round from which this programme's results were registered
`item-108`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`authorship-and-transparency`	its Statement asks for the thin properties' form for valuations; its context is the thin-legitimacy round from which this programme's thin.* results were registered
`gate-is-legitimacy`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`public-layer-II`	and	plan	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (and: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`item-75`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`li-corrigibility-inequality-public`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legitimate-deference-regret`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`effective-authority-layer`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`authorship-and-transparency`	its Statement ends "with sincerity conditions on which verdicts count"; its Source cites the sincerity.* claims and the thin-legitimacy round, whose sincerity results this programme holds
`cnl-post`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`authorship-and-transparency`	its Statement says the post is "backed by that round's theorem", the thin preservation theorem this programme holds; its Source is the thin-legitimacy round's Part F
`public-legitimacy-improvable`	or	shared-claim	`legitimacy-reasons-programme`	`authorship-and-transparency`	the improvement claim on public-legitimacy-theory ("the thick and thin legitimacy theory … can be improved"); it moves with the claim it improves
`legit-two-theories-bridge`	cases	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (cases: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`bridge-defect-vs-thin-correct`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`bridge-trace-vs-transparency`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-objective-proposals`	cases	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (cases: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`illegitimate-world-ever-better`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`p2-manages-the-news`	and	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (and: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`inferability-positivity`	and	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (and: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`updateless-p2-cellwise-p3`	and	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (and: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`deterrence-not-identified-on-path`	and	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (and: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`corr-legit-modif-frames`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-branch-evaluation-defense-requirements`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`readout-R1-vs-R2`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`retroactive-vs-prospective-judging`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`anger-substitute-for-updatelessness`	cases	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (cases: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-meta-revision`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`corrigibility-needs-consequentialist-wrapper`	cases	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (cases: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`align-to-legitimacy-over-endorsement-risk`	leaf/step	risk-story	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-weighted-objective-improvable`	or	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (or: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`formalism-open-questions-0-3`	frame	plan	`legitimacy-reasons-programme`	`legitimacy`	a pre-existing sub-frame whose question none of the three programmes asks; it moves whole to the line
`internal-standard-never-settled`	cases	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (cases: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-dynamics-learning`	frame	plan	`legitimacy-reasons-programme`	`legitimacy`	a pre-existing sub-frame whose question none of the three programmes asks; it moves whole to the line
`reasons-subcase-of-legitimacy`	cases	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (cases: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`public-internal-legitimacy-is-reasons`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`authorship-and-transparency`	its Statement is the authorship clause ("authorship requires at each step a grounding selection from the reason trace's pre-state prefix"), this programme's question of when a verdict is the principal's own
`reason-mediated-authorship`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`authorship-and-transparency`	its Statement is "authorship of a future evaluation is exclusive principal binding together with reason mediation"; the programme holds authorship.mediation-by-reexecution
`reasons-statics-public`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`public-line-leaves-semantics-alone`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`brandom-objectivity-structural`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`brandom-relative-prices`	leaf/step	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/step: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`reason-exchange-representation-theorem`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legit-intersubjectivity-pre-rationality`	leaf/question	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (leaf/question: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`reasons-improvable`	or	shared-claim	`legitimacy-reasons-programme`	`legitimacy`	the Statement bears on none of the three programmes' questions (or: the maintainer's logical-induction-register legitimacy theory, its cruxes, the clean-room results and the reasons thread); a root of the line
`legitimacy-meets-syntax`	frame	plan	`legitimacy-reasons-programme`	`legitimacy`	a pre-existing sub-frame whose question none of the three programmes asks; it moves whole to the line
`want-figure-out-legitimacy`	leaf/want	want	`legitimacy-reasons-programme`	`legitimacy`	a want of the line's owner; a want is a root of one frame, its owner's placement
`want-corrigibility-theorems-via-legitimacy`	leaf/want	want	`legitimacy-reasons-programme`	`legitimacy`	a want of the line's owner; a want is a root of one frame, its owner's placement
`want-legitimacy-anti-meme-post`	or	want	`legitimacy-reasons-programme`	`legitimacy`	a want of the line's owner; a want is a root of one frame, its owner's placement

The frames' Notes say where their former roots went (normativity, legitimacy, operative-force, response-learning); the two retired frames' Notes name this table.

3. Part 3 — the merges (ruling F5)

Counts: bagged before 244 (an and with a record child whose only parent is a frame; the prompt's 249 counted the five results attached only through given edges, which are not bagged, and 11 more sit under an and); merged 2; unmatched 180; ambiguous 54; blocked 8 (one candidate the protocol cannot merge into); reserved 0 (no candidate already carried a record); bagged after 242.

The rules as applied. Candidates for a result R are nodes in the same line as R (the subtree of a line frame R is in), not formal results or their leaves, frames, items or wants, matched by any of the prompt's three rules; the match is read in tiers, the strongest first, and a tier with a match ends the search: (1) id: the candidate's id is the registry id with its first dotted component stripped or its dots made hyphens; (2) declaration: the candidate's Source cites the declaration's full name, the Lean file the record names, or the declaration's last component where that component is an identifier (it carries an underscore, a capital or a digit, so preservation or criterion as English words do not match) and no other registry declaration ends in it; (3) round: the candidate's Source cites the result's origin_round directory and its Statement shares a content word with the registry short_name (lowercase alphabetic tokens of four letters or more, the function words, the evaluative list and the words claim, result, theorem, holds, proved removed, plurals folded; one shared word is the floor the prompt sets). One-to-one: a candidate that is the sole candidate of several results at one tier is ambiguous for all of them, and a stronger tier's match takes the claim (a weaker tier's sole candidate already taken is blocked). A merge needs an assertion-shaped candidate: a step or assumption leaf without given; a question leaf (a question's positive answer), a route leaf (a way an improvement goes), an or or cases node (which hold no record), an and (it has its own completeness leaf, and which survives needs a rule) or a leaf carrying given (an and may not) is listed as blocked with the reason.

result (now an alias)	claim	matched by	the claim was	repointed
`kernel.box2-finite-time`	`kernel-box2-finite-time`	id	leaf/step under formalism-open-questions-0-3	public-corrigibility-line
`thin.value-of-thick`	`thin-value-of-thick`	id	leaf/step under public-legitimacy-theory	authorship-and-transparency

Ambiguous (several candidates, or one shared with another result at the same tier):

capability-beyond-oversight (declaration tier): rs-press-discounting (declaration (last component)) — the candidate is also the sole candidate of erosion-under-silence
erosion-under-silence (declaration tier): rs-press-discounting (declaration (last component)) — the candidate is also the sole candidate of capability-beyond-oversight
mart-implies-value-refuted-unconditional (declaration tier): cleanroom-deference-headlines (declaration (last component)) — the candidate is also the sole candidate of timely-pointwise-tower-refuted-in-setting
timely-pointwise-tower-refuted-in-setting (declaration tier): cleanroom-deference-headlines (declaration (last component)) — the candidate is also the sole candidate of mart-implies-value-refuted-unconditional
asklist.offprotocol-deviation-finite (round tier): post-corrigible-normative-learning (origin round + shared words: protocol) — the candidate is also the sole candidate of post.theorem, transparent.blind-payload
asklist.rows-keep-verdicts-counted (round tier): effective-authority-layer (origin round + shared words: verdict) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-tainted, authority.comprehension-short, authority.queue-no-bypass, authority.queue-no-caused-shortfall, post.witness-control-maximizer, shortfall.atom-generable, shortfall.deferred-eventually-excluded, shortfall.eventually-excluded, shortfall.table-generable, sincerity.rows-keep-verdicts-counted, sincerity.rows-keep-verdicts-tainted, sincerity.sincere-error-counts
asklist.rows-keep-verdicts-tainted (round tier): effective-authority-layer (origin round + shared words: verdict) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-counted, authority.comprehension-short, authority.queue-no-bypass, authority.queue-no-caused-shortfall, post.witness-control-maximizer, shortfall.atom-generable, shortfall.deferred-eventually-excluded, shortfall.eventually-excluded, shortfall.table-generable, sincerity.rows-keep-verdicts-counted, sincerity.rows-keep-verdicts-tainted, sincerity.sincere-error-counts
authority.approval-discount-defeated (round tier): post-corrigible-normative-learning (origin round + shared words: asking, line); effective-authority-layer (origin round + shared words: approval)
authority.comprehension-short (round tier): effective-authority-layer (origin round + shared words: cannot) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-counted, asklist.rows-keep-verdicts-tainted, authority.queue-no-bypass, authority.queue-no-caused-shortfall, post.witness-control-maximizer, shortfall.atom-generable, shortfall.deferred-eventually-excluded, shortfall.eventually-excluded, shortfall.table-generable, sincerity.rows-keep-verdicts-counted, sincerity.rows-keep-verdicts-tainted, sincerity.sincere-error-counts
authority.exchange-rate-queue (round tier): post-corrigible-normative-learning (origin round + shared words: value); authority-typed-corrigibility (origin round + shared words: exchange, rate); thin-value-of-thick (origin round + shared words: value); effective-authority-layer (origin round + shared words: queue)
authority.flood-caused-shortfall (round tier): post-corrigible-normative-learning (origin round + shared words: through); authority-typed-corrigibility (origin round + shared words: agent, only); effective-authority-layer (origin round + shared words: caused, count, flood, shortfall)
authority.queue-no-bypass (round tier): effective-authority-layer (origin round + shared words: queue) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-counted, asklist.rows-keep-verdicts-tainted, authority.comprehension-short, authority.queue-no-caused-shortfall, post.witness-control-maximizer, shortfall.atom-generable, shortfall.deferred-eventually-excluded, shortfall.eventually-excluded, shortfall.table-generable, sincerity.rows-keep-verdicts-counted, sincerity.rows-keep-verdicts-tainted, sincerity.sincere-error-counts
authority.queue-no-caused-shortfall (round tier): effective-authority-layer (origin round + shared words: queue, shortfall) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-counted, asklist.rows-keep-verdicts-tainted, authority.comprehension-short, authority.queue-no-bypass, post.witness-control-maximizer, shortfall.atom-generable, shortfall.deferred-eventually-excluded, shortfall.eventually-excluded, shortfall.table-generable, sincerity.rows-keep-verdicts-counted, sincerity.rows-keep-verdicts-tainted, sincerity.sincere-error-counts
authority.revocation-default (round tier): authority-typed-corrigibility (origin round + shared words: allocation); effective-authority-layer (origin round + shared words: caused, revocation, shortfall)
kernel.box1-fidelity-versus-fully-updated-deference (round tier): authority-typed-corrigibility (origin round + shared words: fidelity); four-corrigibility-definitions (origin round + shared words: fidelity); kernel-lexical-order-hypothesis (origin round + shared words: outcome)
kernel.box2-dominance (round tier): authority-typed-corrigibility (origin round + shared words: fidelity, objective, score); four-corrigibility-definitions (origin round + shared words: corrigible, fidelity, score); kernel-lexical-order-hypothesis (origin round + shared words: score)
kernel.box2-mediation-faithful (round tier): authority-typed-corrigibility (origin round + shared words: allocation, authority, faithful); four-corrigibility-definitions (origin round + shared words: faithful)
kernel.box3-gap-exchange-rate (round tier): authority-typed-corrigibility (origin round + shared words: only, risk) — the candidate is also the sole candidate of kernel.extension-exploration-realization, post.witness-learned-weight
kernel.box3-no-incentive-to-compromise (round tier): kernel-box2-finite-time (origin round + shared words: below); authority-typed-corrigibility (origin round + shared words: score); four-corrigibility-definitions (origin round + shared words: legitimacy, legitimate, score); kernel-lexical-order-hypothesis (origin round + shared words: below, compromised, equal, legitimacy, legitimate, outcome, score, value, worth)
kernel.box3-no-laundering (round tier): kernel-box2-finite-time (origin round + shared words: below); kernel-lexical-order-hypothesis (origin round + shared words: below)
kernel.box3-restoration (round tier): authority-typed-corrigibility (origin round + shared words: agent, prefer); kernel-lexical-order-hypothesis (origin round + shared words: outcome, value)
kernel.corrigible-fidelity-score (round tier): authority-typed-corrigibility (origin round + shared words: fidelity, only, risk, score, unfaithfulnes); four-corrigibility-definitions (origin round + shared words: corrigible, fidelity, score); kernel-lexical-order-hypothesis (origin round + shared words: score, value)
kernel.corrigible-generic (round tier): authority-typed-corrigibility (origin round + shared words: objective, risk); four-corrigibility-definitions (origin round + shared words: corrigible)
kernel.extension-exploration-realization (round tier): authority-typed-corrigibility (origin round + shared words: rate) — the candidate is also the sole candidate of kernel.box3-gap-exchange-rate, post.witness-learned-weight
kernel.hierarchy-per-decision (round tier): four-corrigibility-definitions (origin round + shared words: legitimate); kernel-lexical-order-hypothesis (origin round + shared words: compromised, legitimate)
kernel.subjective-exchange-rate (round tier): kernel-box2-finite-time (origin round + shared words: option); authority-typed-corrigibility (origin round + shared words: fidelity, objective, prefer, score, unfaithfulnes); four-corrigibility-definitions (origin round + shared words: corrigible, fidelity, score); kernel-lexical-order-hypothesis (origin round + shared words: score)
legitimacy.payload-of-view (round tier): post-corrigible-normative-learning (origin round + shared words: declared, legitimacy); cnl-post (origin round + shared words: legitimacy)
post.flood-below-asking (round tier): post-corrigible-normative-learning (origin round + shared words: asking, protocol); authority-typed-corrigibility (origin round + shared words: score); effective-authority-layer (origin round + shared words: flood)
post.theorem (round tier): post-corrigible-normative-learning (origin round + shared words: post) — the candidate is also the sole candidate of asklist.offprotocol-deviation-finite, transparent.blind-payload
post.theorem-li (round tier): post-corrigible-normative-learning (origin round + shared words: post); thin-value-of-thick (origin round + shared words: decision, finite)
post.witness-control-maximizer (round tier): effective-authority-layer (origin round + shared words: flood, revocation, shortfall) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-counted, asklist.rows-keep-verdicts-tainted, authority.comprehension-short, authority.queue-no-bypass, authority.queue-no-caused-shortfall, shortfall.atom-generable, shortfall.deferred-eventually-excluded, shortfall.eventually-excluded, shortfall.table-generable, sincerity.rows-keep-verdicts-counted, sincerity.rows-keep-verdicts-tainted, sincerity.sincere-error-counts
post.witness-hindsight (round tier): post-corrigible-normative-learning (origin round + shared words: value); thin-value-of-thick (origin round + shared words: value); effective-authority-layer (origin round + shared words: approval)
post.witness-learned-weight (round tier): authority-typed-corrigibility (origin round + shared words: fixed) — the candidate is also the sole candidate of kernel.box3-gap-exchange-rate, kernel.extension-exploration-realization
post.witness-sole-authority (round tier): authority-typed-corrigibility (origin round + shared words: authority, risk); effective-authority-layer (origin round + shared words: authority)
shortfall.atom-generable (round tier): effective-authority-layer (origin round + shared words: shortfall) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-counted, asklist.rows-keep-verdicts-tainted, authority.comprehension-short, authority.queue-no-bypass, authority.queue-no-caused-shortfall, post.witness-control-maximizer, shortfall.deferred-eventually-excluded, shortfall.eventually-excluded, shortfall.table-generable, sincerity.rows-keep-verdicts-counted, sincerity.rows-keep-verdicts-tainted, sincerity.sincere-error-counts
shortfall.deferred-ahead (round tier): authority-typed-corrigibility (origin round + shared words: only); effective-authority-layer (origin round + shared words: shortfall)
shortfall.deferred-eventually-excluded (round tier): effective-authority-layer (origin round + shared words: shortfall) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-counted, asklist.rows-keep-verdicts-tainted, authority.comprehension-short, authority.queue-no-bypass, authority.queue-no-caused-shortfall, post.witness-control-maximizer, shortfall.atom-generable, shortfall.eventually-excluded, shortfall.table-generable, sincerity.rows-keep-verdicts-counted, sincerity.rows-keep-verdicts-tainted, sincerity.sincere-error-counts
shortfall.eventually-excluded (round tier): effective-authority-layer (origin round + shared words: shortfall) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-counted, asklist.rows-keep-verdicts-tainted, authority.comprehension-short, authority.queue-no-bypass, authority.queue-no-caused-shortfall, post.witness-control-maximizer, shortfall.atom-generable, shortfall.deferred-eventually-excluded, shortfall.table-generable, sincerity.rows-keep-verdicts-counted, sincerity.rows-keep-verdicts-tainted, sincerity.sincere-error-counts
shortfall.security-generable (round tier): authority-typed-corrigibility (origin round + shared words: fixed); thin-value-of-thick (origin round + shared words: finite); effective-authority-layer (origin round + shared words: shortfall)
shortfall.table-generable (round tier): effective-authority-layer (origin round + shared words: shortfall) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-counted, asklist.rows-keep-verdicts-tainted, authority.comprehension-short, authority.queue-no-bypass, authority.queue-no-caused-shortfall, post.witness-control-maximizer, shortfall.atom-generable, shortfall.deferred-eventually-excluded, shortfall.eventually-excluded, sincerity.rows-keep-verdicts-counted, sincerity.rows-keep-verdicts-tainted, sincerity.sincere-error-counts
sincerity.lie-tainted (round tier): post-corrigible-normative-learning (origin round + shared words: declared, through, transparency); thin-value-of-thick (origin round + shared words: transparency)
sincerity.rows-keep-verdicts-counted (round tier): effective-authority-layer (origin round + shared words: verdict) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-counted, asklist.rows-keep-verdicts-tainted, authority.comprehension-short, authority.queue-no-bypass, authority.queue-no-caused-shortfall, post.witness-control-maximizer, shortfall.atom-generable, shortfall.deferred-eventually-excluded, shortfall.eventually-excluded, shortfall.table-generable, sincerity.rows-keep-verdicts-tainted, sincerity.sincere-error-counts
sincerity.rows-keep-verdicts-tainted (round tier): effective-authority-layer (origin round + shared words: verdict) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-counted, asklist.rows-keep-verdicts-tainted, authority.comprehension-short, authority.queue-no-bypass, authority.queue-no-caused-shortfall, post.witness-control-maximizer, shortfall.atom-generable, shortfall.deferred-eventually-excluded, shortfall.eventually-excluded, shortfall.table-generable, sincerity.rows-keep-verdicts-counted, sincerity.sincere-error-counts
sincerity.sincere-error-counts (round tier): effective-authority-layer (origin round + shared words: count) — the candidate is also the sole candidate of asklist.rows-keep-verdicts-counted, asklist.rows-keep-verdicts-tainted, authority.comprehension-short, authority.queue-no-bypass, authority.queue-no-caused-shortfall, post.witness-control-maximizer, shortfall.atom-generable, shortfall.deferred-eventually-excluded, shortfall.eventually-excluded, shortfall.table-generable, sincerity.rows-keep-verdicts-counted, sincerity.rows-keep-verdicts-tainted
thin.converse-fails-on-correct (round tier): post-corrigible-normative-learning (origin round + shared words: correct, value); thin-value-of-thick (origin round + shared words: correct, value)
thin.correct-of-thick (round tier): post-corrigible-normative-learning (origin round + shared words: correct, transparency); thin-value-of-thick (origin round + shared words: authorship, coherence, correct, give, transparency)
thin.covert-blocking-correct-iff (round tier): post-corrigible-normative-learning (origin round + shared words: correct); thin-value-of-thick (origin round + shared words: correct)
thin.marginal-martingale-too-weak (round tier): post-corrigible-normative-learning (origin round + shared words: correct); thin-value-of-thick (origin round + shared words: correct)
thin.preservation (round tier): post-corrigible-normative-learning (origin round + shared words: transparency); thin-value-of-thick (origin round + shared words: authorship, give, transparency)
thin.reflection-inherited-value (round tier): post-corrigible-normative-learning (origin round + shared words: correct, through, value); thin-value-of-thick (origin round + shared words: correct, value)
thin.sufficient-of-integrity-openness (round tier): post-corrigible-normative-learning (origin round + shared words: sufficient); thin-value-of-thick (origin round + shared words: give, integrity, opennes, sufficient)
thin.value-loss-le-defects (round tier): post-corrigible-normative-learning (origin round + shared words: transparency, value); thin-value-of-thick (origin round + shared words: authorship, transparency, value)
thin.value-of-correct-sufficient (round tier): post-corrigible-normative-learning (origin round + shared words: correct, sufficient, value); thin-value-of-thick (origin round + shared words: correct, give, sufficient, value)
transparent.blind-payload (round tier): post-corrigible-normative-learning (origin round + shared words: declared) — the candidate is also the sole candidate of asklist.offprotocol-deviation-finite, post.theorem

Blocked (one candidate the protocol cannot merge into):

anticipation-branchwise-dominance--corrigibility-risk → rs-mistaken-anticipation (declaration (last component)): the candidate carries given (rules-certified-or-learned), which an and may not; a conditional leaf stays a leaf
deterrence-not-identified-on-path → anger-substitute-for-updatelessness (declaration (last component)): the candidate is a cases node, which cannot hold a record
kernel-conditioning-fails-witness → rs-no-strategic-legitimacy-defense (declaration (last component)): the candidate carries given (kernel-conditioning-fails-witness@valid), which an and may not; it is conditioned on the result itself
li-deference-collapse-scoped → route-immodest-regular-value (declaration (last component)): the candidate is a route leaf (a way the improvement goes, not an assertion the record establishes)
local-criterion-three-cells → question-relative-deference-li (declaration (last component)): the candidate is a question leaf (a question's positive answer, not an assertion the record establishes)
uea-floored-agent-theorem → uea-floor-log-form (declaration): the candidate is a route leaf (a way the improvement goes, not an assertion the record establishes)
void-branch-deterrence-floor--corrigibility-risk → rs-captor-service (declaration (last component)): the candidate carries given (kernel-restoration@valid), which an and may not; it is conditioned on the result itself
thin.preservation-pair → thin-value-of-thick (origin round + shared words: integrity): the candidate is the id-tier match of thin.value-of-thick, which carries its record

Unmatched (no candidate):

agency-correspondence-refuted, asklist.order-by-want-tainted, asklist.volume-by-want-tainted, authorship.mediation-by-reexecution, base-rate-working-picture, basin-formal-object, bli-exactness-limits, bli-sources-refuted, carey-everitt-thm10-vacuous, cdt-defined-for-learner, certificate.advantage-estimate, certificate.defect-bound, certificate.grade-register-strict, certificate.margin-forces-agreement, certificate.override-bound, certificate.selection-eq-of-margin, coherence.modulus-net-cover, coherence.modulus-soundness, compiler.constraints-end-to-end, compiler.deductive-end-to-end, compiler.schedule-end-to-end, condensation-correspondence-theorem, condensation-equivalence-ae, conditioning-manages-news, convergence.discord-pairwise, convergence.discord-persists, convergence.hull-mass-anti, corrective.advisor-veto, corrective.can-correct-future-iff, corrective.can-correct-iff, corrective.capability-measures-cooperation, corrective.forecloses-iff, corrective.no-exclusive-effect, counterfactual-mugging-with-evidence, dd-is-conditional-independence, ddb-finite-frames-formalized, deconfusing.corollary-a, deconfusing.corollary-b, deconfusing.corollary-limit, deconfusing.theorem1, deconfusing.theorem2a, deconfusing.theorem2b, deconfusing.theorem2b-strict, deconfusing.theorem2c, deconfusing.theorem2c-tau, deconfusing.theorem3a, deconfusing.theorem3a-rate, deconfusing.theorem3b, delegation.bridge, delegation.bridge-unconditional, delegation.gradetrust-refinement, dp-dutch-book-done, exact-self-trust-refuted, exposure.greedy-duality, exposure.harvest-attained, exposure.harvest-bound, fa-trust-half-weak-grades, force.deductive-criterion, force.deductive-cumulative-nonneg, force.deductive-day-nonneg, force.deductive-inductor, force.deductive-witness, force.preservation, force.row-tolerance, force.strategy-value, front-running-anticipated-pushes, general-modification-object, generalized-lic, harmony-as-bargaining, hedged-value-unconditional, inductor.certificate-transport, inductor.conformance, inductor.contraposition-blocked, inductor.grounded-survivors, inductor.rebuttal-run, inductor.warrant-inductor, inductor.witness, inferability-and-positivity, inferability-positivity, integrity.evolution-conservation, integrity.faithful-carry, integrity.multiplicity-witness, integrity.propagate-to-segment, integrity.receipts-immutable, jurisdiction.static-view-eq, jurisdiction.value-eq-of-price-realization, kernel-restoration, kernel.extension-bria-realization, kernel.extension-realized-rate, kernel.hierarchy-history, learning-versus-freezing, legit-composition, legit-ddb-transfer, legit-locality-forced, legit-reflection-unsat-modest, legitimacy-conditioned-reflection, legitimacy.answerable, legitimacy.endpoint-only-insufficient, legitimacy.endpoint-trans, legitimacy.evolution-conservation, legitimacy.faithful-carry, legitimacy.multiplicity-witness, legitimacy.propagate-to-segment, legitimacy.receipts-immutable, legitimacy.segment-trans, li-corrigibilization-theorem, li.assessment.budgeter-exploits, li.assessment.budgeter-floor, li.assessment.budgeter-value, li.assessment.criterion, li.assessment.firm-dominance, magnitude.mixture-networth-zero, magnitude.not-trader-payoff, magnitude.sharp-trader-networth, magnitude.signed-bounded, magnitude.signed-bounded-actual, magnitude.sq-error-split, magnitude.unit-trader-networth, maxmin.converse, maxmin.representation, mm-theorem-3-4-refuted, ni.progress-bound, open-integrity.answerable, open-integrity.endpoint-trans, open-integrity.segment-trans, openness.actual-branch-witness, openness.certplus-is-robust-openness, openness.endpoint-only-insufficient, openness.persistence-sufficient, osg-theorem1-corollary, p2-manages-the-news, post.witness-same-level, pressure-as-external-trader, progress.edge-bound, progress.edge-bound-quadratic, progress.headline-separation, ps-cor614-orbit-count, rebuttal.liability-rebutted, rebuttal.settleform-sound, region.hull, region.patterns-complete, region.patterns-nonempty-iff, region.patterns-sound, ro-ftc-finite-core-from-brouwer, ro-generalized-kakutani-refuted, seed.eventually-const, seed.forced-interval-bounds, seed.forced-interval-spec, seed.independence-strict, selection-collapse--corrigibility-risk, shutdown-button-toy, simplex.rational-points-sum-to-one, sincerity.content-fact, smoke.chain-compiles, smoke.faf-asymp-refl, substitution.extensional-admits-both, substitution.separation-requires-disagreement, substitution.sim-depends-on-induced-choice, substitution.unpredictability-separates, thornley-three-theorems, trace-nonrecoverability, transparent.posterior-weight-eq, tt-finite-frames-settled, turner-vnm-fact, two-lesions-propositions-checked, udt-trust-existence-and-tightness, udt101-influence-calculus, uea-horizon-quantifiers, uea-post-usc-backwards, uea-printed-2025-refuted, uea-sink-or-swim-trap, uea-t1-key-lemma-refuted-finite, updateful-when-appropriate, updateless-p2-cellwise-p3, warrant.admissible-not-supported, warrant.licence-chain, warrant.selection-not-transparent, warrant.supported-admissible, warrant.supported-not-licensed, wirehead-iff-finite

What the lists say. Both merges are id matches: the round filed kernel-box2-finite-time and thin-value-of-thick as steps and later registered the same statements under the dotted ids, and the two now hold their records. Forty-nine of the fifty-four ambiguous results are round-tier matches onto six summary claims (effective-authority-layer, post-corrigible-normative-learning, thin-value-of-thick, authority-typed-corrigibility, kernel-lexical-order-hypothesis, four-corrigibility-definitions), each the informal statement of a cluster of registered declarations; an and holds one record and one bridge, so a claim established by seven declarations has no shape in the protocol and no mechanical rule can place the seven. That gap, not the matching, is what the ambiguous list measures; it is queued (§9). The 180 unmatched results are registry entries whose claims the map never stated informally (the deference corrective.*, deconfusing.*, magnitude.*, substitution.* and certificate.* families, the normativity force.*, region.*, warrant.* and progress.* families, most of the clean-room records): the map's informal layer stops at the round summaries.

4. Part 4 — the dockets (ruling F6)

The rules as applied. A programme's origin rounds are the origin_round paths of the registry entries whose formal results sit under it (through state/rounds.json); an item in a line's docket whose context paths lie under exactly one programme's origin round, that programme in the item's own line, moves to a docket under it; an item whose round is shared by programmes of two lines, or whose only hit is a programme of another line, stays and is listed; the project marker normativity.seed names seed-theory, whose docket already held the six seed items. The line is the item's project marker's (normativity, normativity.seed → normativity; normativity.legitimacy → legitimacy; deference → deference-and-trust; none → enabling-work). The rule was applied inside normativity and legitimacy; a hit in deference-and-trust is listed, not applied (the second maintainer's line).

docket	before	after
`docket-deference`	38	38
`docket-legitimacy`	22	22
`docket-normative-inductor`	—	1
`docket-normativity`	39	36
`docket-operative-force`	—	1
`docket-seed`	6	7
`docket-workspace`	15	15

Items with evidence (every item whose context or marker names a programme's round):

item-10 (project none, in docket-workspace): cross-line: context under prompts/2026-08-10-repo-scaffolding (origin round of operative-force's results, line normativity); not applied, the item stays in its line's docket
item-117 (project normativity.seed, in docket-seed): project marker normativity.seed names the subproject whose programme is seed-theory; already there
item-118 (project normativity.seed, in docket-seed): project marker normativity.seed names the subproject whose programme is seed-theory; already there
item-119 (project normativity.seed, in docket-seed): project marker normativity.seed names the subproject whose programme is seed-theory; already there
item-120 (project normativity.seed, in docket-seed): project marker normativity.seed names the subproject whose programme is seed-theory; already there
item-121 (project normativity.seed, in docket-seed): context under projects/normativity/rounds/2026-10-21-rebuttal-and-end-to-end, the origin round of seed-theory's results; already there
item-122 (project normativity.seed, in docket-seed): project marker normativity.seed names the subproject whose programme is seed-theory; already there
item-123 (project normativity, in docket-normativity): context under projects/normativity/rounds/2026-10-21-rebuttal-and-end-to-end, the origin round of seed-theory's results — moved
item-124 (project normativity, in docket-normativity): context under projects/normativity/rounds/2026-10-22-normative-inductor-integration, the origin round of normative-inductor's results — moved
item-24 (project deference, in docket-deference): context under prompts/2026-08-11-phase-ii-prediction, the origin round of deference-finite-results's results; not applied: the reorganization of deference-and-trust is the second maintainer's
item-39 (project normativity, in docket-normativity): context under projects/normativity/rounds/2026-08-16-traderized-enforcement, the origin round of operative-force's results — moved
item-61 (project normativity.legitimacy, in docket-legitimacy): cross-line: context under projects/normativity/rounds/2026-08-16-traderized-enforcement (origin round of operative-force's results, line normativity); not applied, the item stays in its line's docket
item-78 (project normativity, in docket-normativity): ambiguous: its context round (projects/normativity/legitimacy/rounds/2026-09-07-final-pressure-pass) registered results of normative-inductor (normativity) and of integrity-and-record and openness-and-non-capture (legitimacy); a round several programmes share places nothing; stays
item-88 (project normativity.legitimacy, in docket-legitimacy): cross-line: context under projects/normativity/rounds/2026-08-16-traderized-enforcement (origin round of operative-force's results, line normativity); not applied, the item stays in its line's docket

Three items moved (item-39 → docket-operative-force, new; item-124 → docket-normative-inductor, new; item-123 → docket-seed); the two new dockets are shared-claim roots of their programmes. Not applied, each listed above with its evidence: item-78 (its round registered results of normative-inductor and of two legitimacy programmes: a shared round places nothing), item-61 and item-88 (legitimacy items whose context is the traderized-enforcement round of operative-force: cross-line), item-10 (a workspace item whose context is the repo-scaffolding round, from which the two smoke.* entries under operative-force were registered: cross-line, and the smoke claims are no programme's results in any reading), item-24 (the second maintainer's line). Every docket's Statement now reads Work orders filed against <programme>, not yet attached to the goal each serves; their placement is the maintainer's (seven dockets; the five existing ones version-bumped, since a Statement change without a bump is refused).

Representation. A docket is an or node tagged either-suffices, a shared-claim root of its frame. That form makes it a goal of the ranking with a derived probability and a status, which it is not: a work order with no goal identified asserts nothing, and "some filed work order is executed" was a sentence written to give the or a Statement. A dockets field on a frame — the item ids the frame's work orders are, read by the index as the frame's work orders and by the ranking as routes under no goal, carrying no status — would represent it more honestly. Proposed in a design-log line, not applied; queued (§9).

5. Protocol and checker changes

change	rule	design-log line
`hypotheses: unread` on a record; a list mixing ids with the word refused; the record and its `and` flagged hypotheses unread in `derived.json`, the index and the view, and as a unit signal	2 (and 1 no longer reads `unread` as an id)	2026-10-23, "ruling F1 …"
`registry_id` on a record; a `priorities` pointer on every registry-built result; index schema `graph-index/2` (`items[].answered_by`, `formal_result`, `claims`); the render's Registered answers in the map line	— (index schema; the render gate reads the same nodes)	2026-10-23, "ruling F2 …"
a retired frame (`superseded_by`) is exempt from "a frame node is a plan root of some frame", holds no roots and no children, is no spell-out candidate and carries no unspelled flag	2	2026-10-23, "ruling F4 …"
the merge pattern (a claim's `and` gains a result's record, bridge and completeness; the result's `and` superseded) needs no checker change; `claims` in the index resolves the registry id	—	2026-10-23, "ruling F5 …"
docket Statements rewritten; a docket stays a shared-claim `or` root; a frame field proposed	—	2026-10-23, "ruling F6 …" and the proposal line
`graph/derived.json` and `graph/index.json` uncommitted (`.gitignore`; the ci job uploads both as artifacts; the `derived` workflow publishes `main`'s through `derived-publish`, enumerated in `AGENTS.md`); `versions.json` records both derivation hashes under `generated`; `previous_derivation` reads the branch `derived`, else the file on disk that `versions.json` at HEAD records; the `DIGEST` line names the source; the digest leaves `tasks.md`	7 unchanged (a hand edit of a generated file on disk is still refused); `WORKFLOW SCOPE` conditions 1–5	2026-10-23, "ruling F3 …"

Protocol text: §2 Formal results (unread hypotheses, registry_id, the priorities edge), §9 rule 2 (the hypotheses clause; retired frames), the file table (which generated files are committed), §6's digest paragraph, the glossary's docket and index entries. graph/MAP-AS-CANONICAL.md §6.1 is schema 2. graph/README.md step 1 and AGENTS.md Commit only what the plain run writes say which files are committed; AGENTS.md Security enumerates derived-publish.

Tests: test_record_hypotheses_unread (the flag in the three places, the refusals) and test_index_schema_two_on_the_fixture (the inversion, the claims map, the retired frames, the render line) added; two null inputs in the checker's self-test (a mixed hypotheses list; a retired frame still holding roots); the plain-run test tolerates the DIGEST line.

6. Deviations from this prompt, with reasons

**The publish step is a separate workflow, .github/workflows/derived.yml, not a step of ci.yml.** tests/workflow_scope.py refuses write scope in any workflow triggered on pull_request, and ci.yml is; AGENTS.md's condition 1 says why. ci.yml gained the read-scope half (the plain run and the artifact upload on every run); the write-scope job derived-publish lives in the new workflow behind push to main, reads the artifact of its read-scope sibling and runs .github/derived-publish.sh, which commits the two files as data and verifies the branch after the push (condition 5). The gate passes with the enumeration extended.
**The digest left tasks.md.** Its comparison is now whatever derivation is at hand (the branch derived after a fetch, or the file on disk when it is the derivation HEAD records), and a committed file that read it would change on a fetch or a rerun. The digest stays in derived.json and on the DIGEST line, which names its source as the ruling asks; tasks.md keeps the spell-out candidates, the agent work and the proposals awaiting.
**versions.json records a derivation hash, not the files' generated_hash.** The hash leaves out the digest (which varies with the comparison) and the day's audit draw (which varies with the date), so a rerun on another day or beside another derived branch leaves versions.json unchanged; each file's own generated_hash still covers all of it and a hand edit is refused as before. "A stale derivation is refused" is read as: a derivation whose hash is not the one versions.json records is never taken for the previous derivation.
**F6 was applied inside normativity and legitimacy, and a round shared by programmes of two lines places nothing.** The ruling names no line, but a docket under deference-finite-results reorganizes deference-and-trust, which F4 reserves to the second maintainer, so item-24 is listed and left; item-78 would have gone to normative-inductor on the one hit in its own line while its round also registered the legitimacy spine's results, and its title (Canonicalize the legitimacy spine) says the mechanical placement would have been wrong, so a shared round is read as no evidence.
F5's candidates are assertion-shaped and the match is tiered. A question or route leaf, an or, a cases, an and with its own completeness leaf and a leaf carrying given are listed as blocked rather than merged (the first two are not assertions a record establishes; the others the protocol cannot give a record without a further rule); the declaration's last component counts only as an identifier; and the id rule beats a round-tier match so that thin.value-of-thick goes to the node named after it rather than being lost to the eleven other thin.* results that cite the same round. Two merges result; the prompt's "the target is not zero bagged results" is taken at its word.
Completeness leaves and docket Statements carry a version bump. Rule 4 refuses a Statement change without one; 178 completeness leaves are at version: 2 (the Statement names the conjuncts and the unread hypotheses) and the five existing dockets likewise.
The ruling text drops the backticks around the two removed paths (graph/derived.json, graph/index.json in F3) and the round's PROVENANCE row names graph/index.json without backticks: tests/dead_pointers.py refuses a backticked path that does not exist and tests/untracked_pointers.py one that exists on disk and is not in the repository (the second gate does not read the historical marker the first does), which both are after the fifth commit. No ruling is reworded; an italic note after the entry carries the corrected counts.
The two new dockets' Sources cite the evidence rule, not the project marker, since the marker (normativity) is the line's and placed nothing.

7. Corrections to this prompt

"249 of 260 formal results have a frame as their only parent" — 244; five more (ro-collapse-repair, udt-trust-margin-reading, udt-trust-pointwise-refuted, udt-trust-repaired, uea-plain-agent-bound) have no parent and are attached only as the given parent of another node or row, and 11 sit under an and.
"681 script-parsed hypothesis and setting leaves", "511 hypothesis leaves and 170 setting leaves" — 646: 511 hypothesis leaves and 135 setting leaves (50 of the 178 results had no proposition binder and 43 no data binder). The 68 pre-existing hypothesis leaves include the clean-room .setting leaves; untouched.
"add the derived branch publish step to .github/workflows/ci.yml" — not possible under the repository's own gate (§6, first item); the read-scope half is there, the write-scope job is in derived.yml.
"derived.json grew to 15 MB" — 14.8 MB (14,769,185 bytes) at 86e3f35; index.json 3.7 MB; versions.json 833 KB.
"the function that reads HEAD:./graph/derived.json" is previous_derivation in checkers/graph.py; confirmed, replaced as the ruling says.
"every formal result built from a registry entry with an answers_item" — all 178 entries carry one (a string); every value resolves to an item node.
The pattern legit-composition / legit-composition.record has a fourth conjunct, legit-composition.setting, a hypothesis leaf filed by reading; a merged claim copies the result's children whatever they are, so the pattern holds either way.
Everything else named in the prompt (paths, fields, kinds, node ids, the 27 sub-frames, the 68 pre-existing leaves) exists at 86e3f35 as stated.

8. What this follow-up did not do

The reorganization of the second maintainer's lines (deference-and-trust, decision-theory, stance-and-level, abstraction-arc, enabling-work, including item-24's placement and the thirteen sub-frames there); attaching the 180 unmatched, 54 ambiguous and 8 blocked results; placing any item under the goal it serves (the dockets say so in their Statements); filing any hypothesis by reading (every registry-built record says unread); the site; a dockets frame field (proposed, not applied); confirming any drafted frame (the filed_from: wiki flags stand); any number, judgment, acceptance, priority word or human event.

9. Additions to the reserved list

Queued in DECISIONS.md, Awaiting the author, each with its turns on line and an anchor:

The ambiguous and blocked merges of F5 (54 and 8, §3). Turns on: which claim a result establishes where several summary claims cite one round, and whether a claim established by several registered declarations gets a shape of its own — a reading of the mathematics, not a mechanical rule.
How a docket is represented (§4). Turns on: whether a work order with no goal identified should count as a goal of the ranking at all — what the map's numbers are for.
**Whether tasks.md and view.md follow the derived files out of the tree** (their sizes in §1). Turns on: whether a reviewer reads the renders in the diff or on the published branch — where review happens, once the site exists.

10. Test output

WORKFLOW SCOPE: clean over 4 workflow(s); 3 enumerated write-scope job(s)
DEAD POINTERS: clean over 58 live document(s) — 926 rooted path(s) resolve, 8 of them into declared trees and each saying so; 105 unrooted and 221 glob(s) not this gate's business; 14 historical span(s) and 5 ledger citation(s) exempt
UNTRACKED POINTERS: clean — 333 round-record path(s) and 981 resolving citation(s) are in the repository, not merely on this disk
WORKSPACE STATE: valid — 6 projects, 178 modern claims, 1 foundations, 180 foundation claims, 126 rounds, 52 terms, 137 priorities, 10 interface objects
WIKI REFS: clean over 34 page(s); 304 reference(s) resolve
PRIORITIES RENDER: the committed file equals the render (137 items)
Ran 142 tests in 115.050s

OK
GATE COVERAGE: 11 gate scripts, all invoked (5 self-test only — no local form)
LEAN SORRY GATE: clean over 108 files
LEAN AXIOM DISCIPLINE: every file carries `#print axioms`
PROJECTS:
LEAN BUILD SKIPPED: set WORKSPACE_LEAN=1 to run `lake build`
ALL GREEN (80 project(s))
  … 80 PASS lines, one per project runner, elided (the run on the fifth commit's tree; the sixth commit changes documents only and its run is green too)

Second follow-up of 2026-10-23

Prompt: prompts/2026-10-22-map-as-canonical/PROMPT-followup-2.md, verbatim. Rulings G1–G4 landed as one dated DECISIONS.md entry (anchor 2026-10-23--map-second-follow-up-digest-one-shape-frames-hold-bearing-unconfirmed-want-weighs-nothing), the first commit of this follow-up, with the second maintainer's confirmation appended to the round's M1–M8 queue entry; the queued entry How a docket is represented is closed by G2. Auto-merge stays off. Every commit was preceded by python3 -m checkers.graph --root ., python3 tests/run.py and, at the commits that touch the checker, python3 -m unittest discover -s checkers/tests -t . in a detached checkout of the commit under test with no graph/derived.json on disk and no derived branch (the state CI is in). Commits above 4dc8811: c81f0ca4 (the rulings), 92b973c0 (G1), b183ad66 (G2), 1d0c2e25 (G3), 84872ad5 (G4), eaee0762 (Part 4) and the report commit.

1. Verification

Before (the branch at 4dc8811, the CI run of that head): the checker OK 2415 nodes, 2411 events, 3389 ranked units; CI red on one test, test_table_filed_on_the_fixture_passes, KeyError: 'statuses_flipped' (Ran 142 tests … FAILED (errors=1)); no other test failed there.

After (the last commit):

the checker: OK 2408 nodes, 2411 events, 1886 ranked units; research first: the decision term (printed, not applied) favours vet hull-condition-total-trust (ranked unit 16), D 0.00016, one WARN (the summary of unpointed decision anchors, 244 of 295), and DIGEST compared with graph/derived.json on disk (the derivation versions.json at HEAD records): 0 status(es) flipped, 0 root status(es) flipped, 0 level(s) moved, 0 event(s) counted since, 0 revision(s), 0 want(s) filed since, 0 proposal(s) awaiting;
tests/run.py: ALL GREEN (80 project(s)) (the gates' lines in §9);
the unit tests the way CI runs them, in the CI-shaped checkout: `Ran 146 tests in 108.321s, OK`.

Node count 2415 → 2408 (the seven docket files). Ranked units 3389 → 1886: the 177 formal results and 120 items that were goals of the ranking as shared-claim roots are held now, so the units that read them for their own entropy are gone; the units on the claims that remain roots are unchanged in kind.

Root counts per frame before (4dc8811) and after, with what each frame now holds; the 32 other frames have the roots they had and hold nothing. The top frame holds nothing.

frame	roots before	roots after	held
`authorship-and-transparency`	31	11	20
`deference-and-trust`	29	28	38
`deference-finite-results`	31	0	31
`enabling-work`	8	7	15
`integrity-and-record`	17	0	17
`legitimacy`	72	11	82
`normative-inductor`	18	1	17
`normativity`	9	8	36
`openness-and-non-capture`	4	0	4
`operative-force`	29	1	28
`public-corrigibility-line`	65	12	53
`seed-theory`	10	0	16

Held nodes in all: 357, in 12 frames. Frames whose roots became empty: deference-finite-results, integrity-and-record, openness-and-non-capture, seed-theory; each is flagged spelled; nothing yet a goal (N held) and is no spell-out candidate (spell-out candidates 13 → 13).

2. Part 1 — the digest has one shape (G1)

digest() returns the same eight keys whatever previous_derivation found: with nothing to compare with, compared_with is null, the four lists are empty, events_new is 0, and proposals_awaiting lists the proposals awaiting approval as before (see §7, deviations). digest_line prints DIGEST no previous derivation to compare with … when compared_with is null, as before. DIGEST_KEYS names the shape and a unit test (test_digest_has_one_shape_without_a_previous_derivation) pins it on a fresh tree and after a second run.

The audit of every test under checkers/tests/ that reads D["digest"]:

test	how it gets its previous derivation
`test_revise_and_draft`	three plain runs on its own temporary tree; the second and third compare with the first's `derived.json`, written by the test
`test_digest_reports_node_and_root_flips`	the same pattern, twice
`test_digest_has_one_shape_without_a_previous_derivation` (new)	asserts the keys with no previous derivation, then runs again and asserts the same keys with one
`test_table_filed_on_the_fixture_passes`	the only one that read a file a prior run left behind (the `graph/derived.json` copied from the working tree, present locally, absent in CI); it now runs the plain run once on its unmodified copy before the edit, and asserts `compared_with` names that file

Fixtures.copy_fixture now drops the two uncommitted generated files from every copy (g.UNCOMMITTED), so no fixture test can depend on a derived.json or index.json the working tree happened to hold. No test was loosened to tolerate a missing key. The CI run at 4dc8811 failed on no other test.

3. Part 2 — what a frame holds (G2)

The field, the rules and the renders are in §6. The three moves, mechanically:

Dockets dissolved. The seven docket-* nodes (every node whose id begins docket-): each removed from its frame's roots and children, its children appended to the frame's holds in the docket's order, its file deleted (not superseded). 120 items held: docket-deference (38 under deference-and-trust), docket-legitimacy (22 under legitimacy), docket-normative-inductor (1 under normative-inductor), docket-normativity (36 under normativity), docket-operative-force (1 under operative-force), docket-seed (7 under seed-theory), docket-workspace (15 under enabling-work).
Round-filed formal results. A shared-claim root of a frame at 4dc8811 that is a formal result (an and with a record child) and was a root of no frame at 4eb420f8, the commit the round began from, moved from roots to holds. That is how the round's and the follow-up's filings were told from roots that predate the round: a formal result that was already a root somewhere on 2026-10-21 (the 72 landed results under updateless-dt-programme, level-super-turing, legitimacy-weighted-deference and the other pre-existing programmes, and the two claims F5 merged results into) stays a root. 177 results held, in 8 frames.
F4's line-level roots. The 60 rows of the follow-up's moves table whose destination is legitimacy with the reason "bears on none of the three programmes" (the three sub-frames moved whole and the three wants are roots still: a frame and a want may not be held). Held by legitimacy.

What the rule left as roots: the item nodes that are shared-claim roots of a frame directly rather than through a docket (item-75, item-91, item-108, item-109, item-110; item-75 is among the 60 F4 rows and is held), the two enabling-work claims the round filed (enabling-map-as-canonical, enabling-program-site), and every root that predates the round (§8).

A gestalt on a held node: 40 AI gestalt events (@smithy-verity) target nodes that are now held and roots nowhere (the F4 line-level claims). Rule 5 ("a gestalt targets a goal") would have refused the tree; a held node ranks under --scope naming it, where its gestalt is used, so rule 5 now admits a node a frame holds as a gestalt target. No event was written, withdrawn or moved.

4. Part 3 — bearing (G3) and the want weight (G4)

G3: the render line reads Registered results bearing on this item:; the index keys are items[].bearing and formal_result.bears_on (the second renamed with the first, since a result that bears on an item does not answer it); schema graph-index/3 covers G2's membership and holds too. item_status is read from the item node and nothing in the checker or the render moves it (the fixture test asserts every item's status is the file's word).

G4: priority_value returns 0 for a want with filed_from: quote and no priority event by its owner, the word's weight once the owner writes one, and the Default medium for a want filed in a session; none still disowns. The view says priority unset (owner has not confirmed: weight 0), the pooled list's note says the want weighs nothing until the word, and the per-owner line says so too. The self-test gained three facts: a quoted want with no word weighs 0/1, with low 1/4, a want its owner filed without a word 1/2. The shipped graph: 13 wants are filed_from: quote without a word; their weight went from 1/2 to 0.

5. Part 4 — the recorded hash of `derived.json`

Cause, as far as this machine can tell. Two plain derivations of the same tree under Python 3.12.13 and Python 3.13.13 (the uv builds, this Mac) are byte-identical, and so are two under PYTHONHASHSEED=1 and =777: the interpreter's version and the hash seed are not the cause. The derivation reads the disk in one place beyond the tree, the private-source flag (source_private_only, a path's existence), and the one node it applies to names a tracked path, so a checkout of the same commit cannot differ there. The derivation holds about 30,600 floats (per unit: bits, weighted bits, rate, the decision term, the per-goal shares; per goal: weight, entropy, promise; per node: weighted informs), each computed through the platform's math library (log2, exp, pow) and serialized at full repr precision; two math libraries that differ in the last bit of log2 give two hashes of one derivation while every rendered number (three decimals) agrees, which is the symptom reported (the renders unchanged, the hash moved). That is the remaining variable, and it is the one the prompt's remedy addresses.

Fix: derivation_hash is taken over canonical(D): keys as strings and sorted, Fractions exact as p/q, sets sorted, dates as ISO strings, every float at six significant digits (HASH_FLOAT_DIGITS), a signed zero as zero. The hash is a function of the derivation at that precision. A unit test perturbs every unit's bits, weighted bits and rate by one ulp and expects the same hash, and expects a different one for a changed status or a changed number. Not verified on the other machine: this follow-up had no second platform; if a plain run there still dirties versions.json, the diff of derived.json between the two machines names the field, and §10 reserves it.

6. Protocol and checker changes, with the design-log lines

change	where	rule	design-log line
the digest has one shape (`DIGEST_KEYS`); `digest_line` reads `compared_with`	`checkers/graph.py`	— (a derived file's shape)	none: a derived-file shape, no protocol change; G1's text in `DECISIONS.md`
a frame's `holds`: attachment by membership; a held id must resolve; no node both held and a root of one frame, none held twice, no frame, valuation node or want held; a retired frame holds nothing	`checkers/graph.py` (lint), `PROTOCOL.md` §2 What a frame holds, §9 rules 1 and 2	1, 2	ruling G2
held nodes outside status, value, stake, goal set, margin, load and units; `--scope` naming a held node ranks it (kind `held`); a frame with `roots: []` and a non-empty `holds` is spelled, flagged spelled; nothing yet a goal (N held), no spell-out candidate	`checkers/graph.py` (`goal_info`, `ranking`, `derive`), `PROTOCOL.md` §2, §8	—	ruling G2
`derived.json`: `held` per frame, `held_by` and the flag held by on the node, `frame.holds`, `subframe.holds`; the index: `in_frames[].membership`, `frame.holds`, schema `graph-index/3`; the view: Held under each frame, counts on the frame lines; tasks: the count on the goal tree	`checkers/graph.py`, `graph/MAP-AS-CANONICAL.md` §6, §6.1	—	ruling G2
a gestalt may target a node a frame holds	`checkers/graph.py`, `PROTOCOL.md` §9 rule 5	5	ruling G2 (the first line)
the dockets retired: §5's item sentence, §8 Item node, docket (retired), `MAP-AS-CANONICAL.md` §7	`PROTOCOL.md`, `graph/MAP-AS-CANONICAL.md`	—	ruling G2, the three moves
Registered results bearing on this item; `items[].bearing`, `formal_result.bears_on`	`checkers/priorities_render.py`, `checkers/graph.py`, `PROTOCOL.md` §2, `MAP-AS-CANONICAL.md` §6.1	—	ruling G3
an unconfirmed quoted want weighs 0	`checkers/graph.py` (`priority_value`, the notes and lines), `PROTOCOL.md` §2, §6, §8, `graph/README.md`	— (a Default)	ruling G4
the derivation hash over `canonical(D)`	`checkers/graph.py`	—	Part 4

Self-test null inputs added (rule): a held id that resolves to nothing (1); a node both held and a root of one frame (2); a want in holds (2). Self-test facts added: the three want weights of G4. Unit tests added: test_digest_has_one_shape_without_a_previous_derivation, test_held_nodes_are_attached_and_no_goals, test_unconfirmed_want_weighs_nothing, test_derivation_hash_is_canonical; test_index_schema_two_on_the_fixture became test_index_schema_three_on_the_fixture; test_hoist_rule_on_the_fixture counts a spell-out line among the lines that rest on no AI number (an unspelled frame is read at a stand-in 1/2), because with the held results out of the goal set the spell-out of normative-record rose into the printed forty; test_personal_weights gives its quoted want the owner's word first (G4).

Documents touched beyond the protocol: graph/README.md (the frame paragraph, the want line), AGENTS.md (Goals are frames and wants), graph/MAP-AS-CANONICAL.md §1, §6, §6.1, §7, the Notes of the twelve frames whose roots moved (what each holds now and where the dockets went), graph/DESIGN-LOG.md (five new lines).

7. Deviations from this prompt, with reasons

**proposals_awaiting is not emptied** when there is nothing to compare with. The ruling says the lists are empty in that case; this list is not a comparison (it names the proposals awaiting approval in the current derivation, and tasks.md prints the same list), so emptying it would make the digest say there are none when there are. The key is always present; the comparison lists and counts are empty and zero as ruled. Reversible in one line.
**formal_result.answers renamed to bears_on** beside the ruled answered_by → bearing: the ruling's principle (a result bears on an item, it does not answer it) covers both ends of the join; leaving one as answers would have left the schema saying what the ruling retracts.
Rule 5 admits a held node as a gestalt target. Not in the prompt; forced by the tree: 40 AI gestalts sit on claims that the third move holds and no frame roots. The alternatives were to write or withdraw events (forbidden) or to leave those claims as roots (against G2).
**The schema is graph-index/3 from the G2 commit**, not the G3 commit: G2 changes in_frames and the frame records, and a schema number names a shape; the G3 commit renames the keys inside the same number. §6.1 describes schema 3 as both.
The unit tests in the CI-shaped checkout were run at the G1 commit, the G2 commit and the last commit, not before every commit: the checkout is a detached worktree of the commit under test, and the two commits between change documents and one function each; tests/run.py (which runs the same suite on the working tree, with derived.json present) ran before every commit.

8. Corrections to this prompt

The prompt's "five things to fix" says CI raises KeyError in test_table_filed_on_the_fixture_passes "and possibly others": the run at 4dc8811 failed on that one test only (Ran 142 tests, FAILED (errors=1)).
"A docket is an or with either-suffices": correct; seven dockets, not five (the follow-up added docket-operative-force and docket-normative-inductor).
The want-weight function is priority_value (the prompt says "the want-weight function"); PRIORITY_WORDS is a dict constant and keeps its values, the Default moved in the function.
"in_frames gains a membership: holds entry beside role": done so that every entry carries membership (root or holds); a held entry's role is null.
The prompt names previous_derivation's "callers": there is one, committed_files; the fix is in digest itself, which every caller gets.
Part 4's premise, "varies across interpreters": not reproduced on this machine across 3.12 and 3.13 (§5); the hash was still made canonical as the part asks.

9. Test output

WORKFLOW SCOPE: clean over 4 workflow(s); 3 enumerated write-scope job(s)
DEAD POINTERS: clean over 58 live document(s) — 940 rooted path(s) resolve, 8 of them into declared trees and each saying so; 106 unrooted and 222 glob(s) not this gate's business; 14 historical span(s) and 5 ledger citation(s) exempt
UNTRACKED POINTERS: clean — 333 round-record path(s) and 998 resolving citation(s) are in the repository, not merely on this disk
WORKSPACE STATE: valid — 6 projects, 178 modern claims, 1 foundations, 180 foundation claims, 126 rounds, 52 terms, 137 priorities, 10 interface objects
WIKI REFS: clean over 34 page(s); 304 reference(s) resolve
PRIORITIES RENDER: the committed file equals the render (137 items)
OK 2408 nodes, 2411 events, 1886 ranked units; research first: the decision term (printed, not applied) favours vet hull-condition-total-trust (ranked unit 16), D 0.00016; nothing written (--dry-run)
Ran 146 tests in 109.309s
OK
GATE COVERAGE: 11 gate scripts, all invoked (5 self-test only — no local form)
LEAN SORRY GATE: clean over 108 files
LEAN AXIOM DISCIPLINE: every file carries `#print axioms`
PROJECTS:
  … 80 PASS lines, one per project runner, elided (the run on the Part 4 commit's tree; the report commit changes documents only and its run is green too)
LEAN BUILD SKIPPED: set WORKSPACE_LEAN=1 to run `lake build`
ALL GREEN (80 project(s))

The unit tests in the CI-shaped checkout (the last commit):

$ python3 -m unittest discover -s checkers/tests -t .   # in a detached checkout of eaee0762; no graph/derived.json, no branch derived
Ran 146 tests in 108.321s
OK

10. What this follow-up did not do, and the notes for the maintainers

Not done:

The item nodes that are shared-claim roots directly (item-91 under deference-and-trust; item-108, item-109, item-110 under authorship-and-transparency) and the two enabling-work claims (enabling-map-as-canonical, enabling-program-site) stay roots: the ruling's three moves do not name them. They are goals of the ranking as the dockets were. Since the closing pass of 2026-10-24 the four item roots are held by their frames (agent-decided, reversible; the design log of that date) and the two enabling-work claims remain roots as goals of the enabling line. Turns on: whether a single work order filed under the goal it serves is a route of that goal (a child) or a thing the frame holds — the same question G2 answered for the dockets, asked of five nodes.
The 4 frames with roots: [] and held nodes have no goal; the ranking values each programme as unspelled did (no option value). Naming their goals is a maintainer's spelling-out.
Part 4 is not verified on a second platform.

Notes for the maintainers:

The branch's history holds seven versions of the derived files (the prompt says five; git log -- graph/derived.json graph/index.json lists seven commits that changed them, from 0a73a063 to a0ae3d07, before 5b1e2bba removed them): each a blob of 11–15 MB for graph/derived.json and 3–4 MB for graph/index.json. A squash merge of the pull request keeps none (the squashed tree has neither file), where the repository's convention is a merge commit, which keeps all seven in the history forever (pack loose objects 41.51 MiB, 10 packs 77.83 MiB at the Part 4 commit). Turns on: whether the maintainers trade the merge-commit convention for a small history once, or keep the convention and the copies, or rewrite the branch before merging (a force-push to the pull request, which nobody but the maintainers should do).
**stance-and-level is a plan root only because the protocol makes every frame one.** The cruxes the second maintainer filed (the computability levels as three sub-frames and nine shared claims, the updateful-versus-updateless questions among them) are roots of a frame that is a plan root of the-program; a frame node is a root only as a plan (rule 2), so the group of cruxes enters the plan decision as a programme whose steps are the level frames, and is valued as one. The second maintainer may prefer his cruxes back as shared-claim roots of the-program, grouped by a field rather than a frame. Not changed here. Turns on: whether the frame tree is the only grouping the map offers (then a group of cruxes is a frame and a plan) or a field on a root may group roots without making the group a goal — a protocol choice the second maintainer's own filing should drive.

Reserved additions to DECISIONS.md, Awaiting the author: none new; the M1–M8 entry now carries G1–G4's confirmation.

Closing pass of 2026-10-24

Prompt: prompts/2026-10-22-map-as-canonical/PROMPT-closing.md, verbatim. Two commits on the branch above 43468d7d: cee33492 (Parts 1 and 2, the prompt) and the report commit (PR-BODY.md, this section). No ruling is taken; the two agent-decided choices are in the design log and reversible. Auto-merge stays off. The body posted to pull request #131 is committed as prompts/2026-10-22-map-as-canonical/PR-BODY.md.

1. Verification

The checker after the first commit and after the second: OK 2408 nodes, 2411 events, 1886 ranked units; research first: the decision term (printed, not applied) favours vet hull-condition-total-trust (ranked unit 16), D 0.00016; one WARN (244 of 295 decision anchors unpointed); DIGEST compared with graph/derived.json on disk (the derivation versions.json at HEAD records): 0 status(es) flipped, 0 root status(es) flipped, 0 level(s) moved, 0 event(s) counted since, 0 revision(s), 0 want(s) filed since, 0 proposal(s) awaiting.
tests/run.py before each commit: ALL GREEN (80 project(s)) (Ran 146 tests in 109.499s, OK on the first commit's tree; the same suite, run before the commit; the second commit changes two documents and the run is green on the second's).
The unit tests the way CI runs them, in a detached checkout of the first commit with no graph/derived.json on disk and no derived branch (the second commit changes documents only): `Ran 146 tests in 110.614s, OK`.

Node count unchanged at 2,408; ranked units unchanged at 1,886 (the OK line before and after the move). Held nodes 357 → 361.

2. Part 1 — the residual roots

frame	roots before	roots after	held before	held after
`deference-and-trust`	28	27	38	39
`authorship-and-transparency`	11	8	20	23

item-91, item-108, item-109, item-110 moved from roots to holds of the frame each was a root of; enabling-map-as-canonical and enabling-program-site stay shared-claim roots of enabling-work. The Notes of authorship-and-transparency say where its three items went. The second follow-up's §10 "not done" bullet now says the item roots are held.

Design-log lines (both 2026-10-24, both marked agent-decided and reversible): the four item roots held, for the reason G2 gave for the 120 docketed items; the two enabling-work claims kept as roots, goals of the enabling line by their Statements.

3. Part 2 — the proposal document's principle

graph/MAP-AS-CANONICAL.md §2 gains one paragraph after the existing text: the map is a representation of the research program, only part of which is an argument; a frame's roots are its goals and enter the ranking, what a frame holds is what it is about and does not; the argument view, the research map and the contribution view are three readings of one structure that do not share inference semantics. It cites G2's anchor. Nothing else in the section changed.

4. Part 3 — the pull request body

Posted with gh pr edit 131 --body-file, committed as prompts/2026-10-22-map-as-canonical/PR-BODY.md. The title stays "The map as the canonical index of the program (2026-10-22 round)": it is the description, not a proposal. Every number in the body names its report section. The repository's pull-request template sections (the layer touched, the claim class, the model attribution and the rest) follow the six sections the prompt asked for, because tests/attribution.py reads the body from the event payload and requires a Model: line; a body without them would fail CI.

5. Corrections to this prompt

"rounds pointed at (120 of 126)": 119 of 126. The round report §6 lists six unpointed of 125; the 126th is this round's own record, which no node points at either.
"the re-owning by source attribution, follow-up report §2": the re-owning is a deviation of the round report (§2), not of the follow-up; and the queue line says five wants are @anson's and eight @abram's, while the tree holds four and nine (want-legitimacy-corrigibility-theorems is @abram's).
"78 kernel-checked declarations": the round report §2 says 78; its §4 item 1 and the queue line open--whether-clean-room-records-get-registry say 80; the index finds 75 record leaves whose Source names the clean-room directory. The body uses 78 and names the discrepancy.
"the ten frames marked filed_from: wiki": ten in the tree; the queue line open--roots-marked-owner-has-not-confirmed says "nine" and lists ten.
"verified on the second platform on 2026-10-23: a plain run on Linux / Python 3.13 leaves versions.json clean": no report section records this; the body attributes it to the first maintainer's closing prompt.
"178 registry-built, all with hypotheses unread": 178 registry entries have a formal result; the index carries 180 nodes with a registry_id because the two results F5 merged keep a deprecated alias node each, and all 180 records say hypotheses: unread.
"held nodes (357 in 12 frames)": 357 before this pass, 361 after it.
"Replace the body": replaced in substance; the template sections are kept for the reason in §4.

Nothing is reserved beyond what the body's §4 and its sub-heading list.

Provenance

Generator: Claude Fable 5.1 (Anthropic), executing the maintainer's prompt; review status ci-only. The scripts that produced the mechanical parts (the signature parser, the item transcription, the anchor writer, the survey counts) ran from the session's scratch directory and are not committed; graph/SURVEY.md's appendix gives the one-liners that reproduce every count. Model attribution trailer on every commit.

Claims it registered

None.

Items it filed or answered

None recorded in the index.

Nodes pointing at it

None.

2026-10-22-map-as-canonical

Prompt — the map as the canonical index of the program (2026-10-22)

What this round is for

Maintainer rulings

Part 0 — Survey

Part 1 — Formal results

Part 2 — Items

Part 3 — Decisions

Part 4 — Narratives

Part 5 — Rounds

Part 6 — The program and its lines

Part 7 — The structure-only index

Part 8 — The proposal document

Reserved to the maintainers

The report

Report — the map as the canonical index of the program (2026-10-22)

1. Verification

2. Deviations from the prompt, with reasons

3. Corrections to the prompt

4. What the protocol could not represent

5. Goal not identified

6. Unattached narratives and unpointed rounds

7. Protocol and checker changes

8. What this round did not do

9. Test output

10. Reserved to the maintainers

Outstanding maintainer actions

Follow-up of 2026-10-23

1. Verification

2. Part 2 — the moves (ruling F4)

3. Part 3 — the merges (ruling F5)

4. Part 4 — the dockets (ruling F6)

5. Protocol and checker changes

6. Deviations from this prompt, with reasons

7. Corrections to this prompt

8. What this follow-up did not do

9. Additions to the reserved list

10. Test output

Second follow-up of 2026-10-23

1. Verification

2. Part 1 — the digest has one shape (G1)

3. Part 2 — what a frame holds (G2)

4. Part 3 — bearing (G3) and the want weight (G4)

5. Part 4 — the recorded hash of derived.json

6. Protocol and checker changes, with the design-log lines

7. Deviations from this prompt, with reasons

8. Corrections to this prompt

9. Test output

10. What this follow-up did not do, and the notes for the maintainers

Closing pass of 2026-10-24

1. Verification

2. Part 1 — the residual roots

3. Part 2 — the proposal document's principle

4. Part 3 — the pull request body

5. Corrections to this prompt

Provenance

Claims it registered

Items it filed or answered

Nodes pointing at it

5. Part 4 — the recorded hash of `derived.json`