the map's follow-up rulings: hypotheses are filed when read, the registry join is an edge, the large derived files leave the tree, one decomposition per line, a formal result sits under the claim it establishes, and items are docketed by programme (F1–F6)

Maintainer rulings, taken by one maintainer in conversation; the second maintainer's confirmation is outstanding (landed as the first commit of the map-as-canonical round's follow-up, prompts/2026-10-22-map-as-canonical/PROMPT-followup.md; the confirmation is appended to the round's queue entry above rather than queued a second time). The review they answer found that the round of 2026-10-22 made every object of the program exist in the map and left the map flat.

F1. Hypotheses are filed when read, not parsed. A record leaf may carry hypotheses: unread in place of a list of hypothesis-leaf ids. The record's Source then carries the declaration's signature verbatim (binders included), so a reader has what the parsed leaves held. A formal result with an unread record is an and of record, bridge and completeness leaf, and the view and the index say hypotheses unread beside it. Filing the hypotheses is a human reading act or an agent act from a human's reading of the Lean; a later round may rank it as a unit, but this follow-up does not. The 511 hypothesis leaves and 170 setting leaves the round parsed from signatures are removed and the records they belonged to set to unread. Hypothesis leaves that predate the round (the 68 that existed on main at 4eb420f, filed by reading) stay as they are.

F2. The registry join is an edge. Every formal result built from a registry entry with an answers_item carries work: [{kind: priorities, ref: <n>}] for that item, and the index inverts it: each item record in the index lists answered_by (the formal-result ids) and each formal result lists answers. An item's item_status does not change — closing an item is the maintainer's (the reservation bar and epistemic debt, "closed needs a statement of record") — but the index and the render show the registered answers beside an open item.

F3. The large derived files leave the tree. graph/derived.json and graph/index.json are no longer committed. The checker still writes them on the plain run, to graph/ as now, and .gitignore excludes them; CI uploads both as workflow artifacts on every run and publishes them to the branch derived (orphan, force-pushed, one commit per main push, with the main commit hash in its message), which is where a site generator reads them from. graph/versions.json stays committed and keeps generated_hash for both files, so a hand-edited or stale derivation is still refused and a rerun still changes nothing. The digest's "previous committed derivation" (checkers/graph.py, the function that reads HEAD:./graph/derived.json) reads instead the versions.json at HEAD plus whatever derived.json is on disk or on the derived branch, and says in its digest line which it compared with. graph/tasks.md and graph/view.md stay committed: they are the human-readable renders and a pull request's diff of them is the point.

F4. One decomposition per line. The pre-existing frames normativity-line (under normativity) and legitimacy-reasons-programme (under legitimacy) are dissolved: each child is moved to the new programme whose question it bears on (seed-theory, operative-force, normative-inductor, response-learning, normative-record; integrity-and-record, openness-and-non-capture, authorship-and-transparency), or, where it bears on none, made a root of the line frame directly, never of a docket. The two frame ids survive as deprecated aliases in the report's old→new map. Every move is a row in a table in the report so a maintainer can review placements in one sitting and reverse any. The same treatment for any other pre-existing sub-frame under the two lines whose question the new programmes now ask. deference-and-trust, decision-theory, stance-and-level, abstraction-arc and enabling-work are not touched by this ruling: their reorganization is the second maintainer's.

F5. A formal result sits under the claim it establishes. For each formal result whose only parent is a frame, find the existing informal claim node it is the formal version of, and merge them: the existing claim node keeps its id and its place in every argument and becomes the and of the record, bridge and completeness leaf (the pattern legit-composition / legit-composition.record already follows); the registry-id node becomes a deprecated alias listed in the old→new map, and the registry id stays recorded on the record leaf and in the index so a claim id still resolves. The matching is mechanical and conservative, and a result with no match or more than one stays where it is, listed. The target is not zero bagged results; it is that every result that has an informal claim in the map is attached to it.

F6. Items are docketed by programme where the evidence places them. An item whose context paths lie under a round directory from which a programme's formal results were registered (origin_round), or whose project marker names a subproject with its own programme (normativity.seed → seed-theory), moves from the line's docket to a docket under that programme; the rest stay in the line's docket. Each docket's Statement is rewritten to say what it is: "Work orders filed against <programme>, not yet attached to the goal each serves; their placement is the maintainer's." A docket is a shared-claim root today; the report says whether that role or a frame field for dockets would represent it more honestly, and proposes the latter in a design-log line if so, without applying it.

Counts in the rulings as the follow-up found them (prompts/2026-10-22-map-as-canonical/REPORT.md, Follow-up of 2026-10-23, corrections): the round parsed 511 hypothesis leaves and 135 setting leaves, not 170; 244 formal results had a frame as their only parent, 5 more were attached only through given edges, and 11 sat under an and. The ruling text is the maintainer's and is not reworded.