Skip to content

Curation MVP implementation plan: seven parts with agent-ready briefs (P1 read compositions, P2 screen, P3 pin, P4 packet, P5 freshness, P6 proposals, P7 runtime) — under #253 #254

Closed

Description

@holden

Reconciled 25 October 2026 with the independent assessment below (all fifteen section C edits, accepted by the owner) and #253's new D8 and D9. The list of changes is the reconciliation comment on this issue.

Implementation plan for #253 (the curation agent MVP), written 21 October 2026. Seven parts, each one PR, each with a brief an implementing agent can take without this conversation. #253 is the specification and wins on any conflict; this issue says what to build, in what order, in which files, and what "done" is for each part. Tick a part here when its PR merges and link the PR.

Ground rules for every part

  • Repository razrfly/dictionary, checkout /Volumes/LLM Models/dictionary/src/dictionary, branch from main. Elixir 1.19 / Phoenix 1.8 / LiveView 1.2, Postgres 18 on port 5434, database devils_dictionary_v2.
  • A fresh worktree needs symlinks for deps, data and each priv/replay/*.jsonl.gz from the main checkout, or Sources.ManifestTest fails.
  • Precommit: MIX_ENV=test MIX_TEST_PARTITION=_<part> DD_NO_OBAN=1 mix precommit. Green is the bar; report the test count.
  • The dev server on port 4007 is also the public site through ngrok. Never restart it, never edit its launch script, never run a migration against v2 without the owner's word in chat.
  • gh pr merge is refused by the permission classifier. Stop at a reviewed PR (CodeRabbit addressed, one independent review) and hand the owner the merge.
  • Build only the part. The owner treats anything beyond it as scope creep. If the part cannot be done as written, say why on the PR rather than widening it.
  • Every PR: a before/after on a real page (pilot page from Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 D2), with the row ids it created.

Order and parallelism

P1 ──┬── P2 (screen) ───────────────┐
     ├── P3 (pin, PoetryDB)          │
     ├── P4 (packet kinds) ──────────┼── P6 (proposal path)
     └── P5 (freshness) ─────────────┤
P7 (runtime fixes, benchmark) ───────┘   P6's live inference needs P7's go

P1 first and alone. Then P2, P3, P4 and P7 in parallel: they touch disjoint code. P4 does not wait for P3: it can use existing catalog artwork fixtures. P5 after P1. P6 last, and its live inference needs P7's revised-instruction benchmark go.

Found while writing P2 (23 Oct): the registry already holds 2,903 PoetryDB poems, 946 Open Library works and 1,644 Met artworks as committed catalog manifests (priv/artworks/manifests/), and Curation.Eligibility admits a catalog-only work as an item. Corrected 25 Oct: catalog references can be stored/rendered, but P2 lists only page-matched artwork rows; poems/books need an explicit page-selection relation (#253 D6). P4 can use existing catalog artwork fixtures without waiting for P3, but must not claim poems became offerable merely because manifests exist. Reading the spec: #253 §2 (rules), §3 (the record and what it can offer today), §5a (the editor), §6 (the screen), §6a (owner versus bot), §7 (freshness), §9 (gaps), §12 (defaults, D1–D9).


P1 — Seed, activate, the page reads published compositions

Spec: #253 §4b, §8, §9 items 1–2, §11. Reopen #156 for it.

Read first: docs/curation/persistence-slice-1.md; lib/devils_dictionary/curation/{opening.ex,opening_reader.ex,published.ex,manual_fixture.ex,configurations.ex,compositions.ex,publications.ex,eligibility.ex}; lib/devils_dictionary_web/live/word_live.ex (the opening is read near lines 171 and 248); lib/devils_dictionary_web/components/opening.ex; lib/mix/tasks/dd.curation.seed.ex.

Build:

  1. A published reader (Curation.PublishedReader or Curation.Published extended) implementing Curation.OpeningReader: for the page's scope and the resolved default configuration, read Published.current/1, run Eligibility.evaluate/5, and map onto %Curation.Opening{} with origin: :published: lead with its policy and register, highlights in position order with their notes as reasons and their note_author_* as the signature, configuration, review (selected by, published by, when), withheld with reasons. Nothing fetched, nothing written.
  2. Curation.Opening.reader/1 returns the published reader by default; the fixture reader only when fixtures are enabled and ?opening=fixture is asked, exactly as now.
  3. WordLive calls the reader on every request. An absent configuration, an absent composition, or an empty opening renders no section.
  4. The disclosure ("How this was curated") shows: configuration slug and version, selected by (label and kind), published by and when. Reserve, but leave empty, the four assessment dates P5 adds.

Tests: no composition → no section; a published composition → lead and highlights with reasons and signatures; withdrawn → nothing; the fixture gate still works in dev and never in prod config; a lead that breaks bierce_first_v1 is withheld, not substituted; an item whose revision was superseded is withheld with :revision_superseded.

On v2, after the owner says so in chat: mix dd.curation.seed, then Curation.Configurations.activate/4 on version 1 of global-default as the reviewer account the owner names (users.id 2 was the routing reviewer). Record before/after counts of curation_configurations, curation_configuration_versions, curator_profiles. Zero compositions is the expected end state.

Done: reviewed PR; configuration active on v2; a composition published by hand in iex on one pilot page appears on that page through the tunnel and disappears on withdrawal.

Reconciled after #256 (25 Oct): language comes from the scope's lexemes, not page.language_tag; scopes are exact sorted lexeme sets, overlapping lexical-page scopes cannot coexist; Published owns eligibility, the reader maps its result; assessment rows are absent until P5; definition highlights have their own tile; actor labels come from the record. The approved typography change and the actual live five-lexeme love scope stand, not the two-lexeme fixture. P1's temporary work/exemplar withholding (:unsupported_item) was repaired by P2 and is resolved. P1 is not reopened.


P2 — The owner's curation screen

Spec: #253 §6, §6a (withdraw means keep off; re-open), §12 D2–D3. Operator-only, under /ops, 404 through the tunnel like the other operator pages (ProxyGuard).

Read first: lib/devils_dictionary_web/router.ex (the /ops live session), an existing operator LiveView under lib/devils_dictionary_web/live/ops* for the conventions, Curation.Compositions (provision/3, create_version/3, items/1, latest_version/1), Curation.Reviews.decide/4, Curation.Publications.{publish/3,withdraw/3}, Curation.LeadPolicy, Curation.Eligibility, Curation.Authority.actor/2, and Curation.ManualFixture for how a selection is expressed as references.

Build:

  1. /ops/curation: the queue. Tabs: pilot pages (D2 allowlist: a configuration-overridable code default, CurationLive.pilot_pages/0, not a config-file edit), published, withheld items, and (reserved for P5/P6) stale and proposed. Each row: word, state, version, counts, a button.
  2. /ops/curation/on/:slug: the editor. Lead radio (the page's applicable Bierce entries first, then other definitions, then "no lead"), with the lead rule's verdict shown. Highlight checkboxes (max three) over the eligible candidates of the four kinds the record admits today (content, sense_quotation, work from page-matched catalog artwork rows, exemplar from accepted illustrates claims; no poems or books, Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 D6), each with a "why" box of 280 characters. Save draft (a new version), Publish (review :accepted then publish, atomically), Withdraw current (with reason), Re-open (clears P5's suppressed when it exists; no-op now). History: list versions with parent links.
  3. Candidate listing reuses the page build (WordLive's page struct) so the lead rule sees the same cards the reader sees.
  4. Exemplar and work highlights render on the page. P1 (Curation P1 (#255): one published opening on /on/love #256) withholds work and exemplar items as :unsupported_item because its demonstration pinned neither. The screen must not offer a highlight the page cannot draw, so P2 adds the two tiles to DevilsDictionaryWeb.Opening: an exemplar (an accepted illustrates claim with its subject label and the reason) and a catalog work (title, creator, year, licence, image where the catalog has one; a poem tile shows title, poet, length and link, no lines), mapped by the published reader from Published.current/1's subject and catalog fields. This is what makes a page carry representative examples (Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 §3), and it is a named requirement here so it is not deferred twice.
  5. Scopes are the On page's. The screen provisions :lexical_page scopes from the page build's headword lexeme ids (the live love reaches noun, verb and a name). The record refuses overlapping scopes, so an exact-word page cannot get its own composition beside the On page's. The screen says so in these words: "Exact-word pages overlap this editing scope and cannot have a separate active lexical-page composition; they do not currently display this composition." (The built label still says "share this"; its one-string correction rides with P5, which already changes the editor.) Actual sharing is a product and routing decision, not implied by the scope rule.

Tests: an exemplar highlight and a work highlight render with their reasons and nothing withheld; the screen is 404 proxied and 200 local; a non-reviewer gets 403; publishing writes a review and a publication and the page shows it; withdrawing removes it; a lead that breaks the rule cannot be saved; a note is attributed to the logged-in human and never to a caller-supplied author; two sessions publishing from the same previous version: one wins, one gets {:stale_pointer, _} (a draft from a moved parent gets {:stale_parent, _}).

Done: the owner curates three pilot pages by hand through the tunnel-free local screen, withdraws and re-publishes one, with ids on the PR.

Status: P2 Delivery A merged at e25dd94; owner demonstration/Delivery B not evidenced. Departures recorded at merge (#258): poem tiles show title/poet/length/link, no lines; no editor height readout; pilot slugs are a configuration-overridable code default, not a new config-file edit; editor URL is /ops/curation/on/:slug; stale publication errors are stale_pointer/stale_parent. Exemplars are accepted nominations only, never instances. Heights are recorded as heading start–heading end: love 882–910 px after P1 (#256, accepted) and 878–906 px after P2 (#258); the heading starts within 900 px, the whole heading does not.


P3 — The pin: a PoetryDB poem becomes a durable work

Spec: #253 §9.6 in full, D5 (highlight-only), D6 (PoetryDB, and the poem-to-page rule the owner decides in the P3/P4 handoff). Reopen nothing; cite #164's creator-identity rule.

Identity, corrected before implementation: PoetryDB has no provider-issued poem ID. Reuse the existing poetrydb_poem namespace and Poetrydb.poem_id/3; it hashes author, title and lines. Changed lines currently mean a changed identity, not a new revision under the same identifier. Decide the permitted supersession relationship explicitly; never deduplicate by title. These are repairs to the pin requirement, not grounds for a broader ingestion project.

Read first: lib/devils_dictionary/discovery/{result.ex,providers.ex,provider.ex} and lib/devils_dictionary/discovery/providers/poetrydb.ex; docs/integrations/source-identity.md; lib/devils_dictionary/registry/{object.ex,external_identifier.ex,content_item.ex,content_revision.ex}; lib/devils_dictionary/sources/{source.ex,source_record.ex} and source_record_revisions; how Curation.Eligibility checks a work item (catalog pin or source-record revision that materialized the object); how #164 mints a creator by identifier.

Build:

  1. Curation.Pins.pin(scope, result_id) (or under Discovery): from a cached PoetryDB result, first check through the existing identity adapter (Poetrydb.identity_record/1) whether the poem is already materialized; otherwise upsert a source record under the poetrydb source keyed by its poetrydb_poem identity, with a revision holding title, author, line count and lines (PoetryDB is public domain; retention is permitted and recorded per provider in the registry's policy, not per pin). Resolve the author through the reviewed author-QID crosswalk that Artworks.Corpus.Poetrydb writes (Creator identity for every provider: a result credits a person by identifier, the pipeline resolves or mints the person, the card links the entry — built once, before any durable quotation is written (build 2 of #158) #164's rule); an unresolved poet does not acquire an invented identifier. Find or mint the work object by its poetrydb_poem identity (never by title). Materialize so that Eligibility's "a source-record revision must be one whose record materialized the item's object" holds.
  2. Record licence string and fetch time on the revision; display_allowed from the provider policy.
  3. Re-pin with changed text → a different identity today; the supersession relationship is the owner's explicit decision, made before this step is built. Provider no longer returns the poem → unavailable_at on the record, last revision kept.
  4. Retire a pin (owner): nulls references (R7), withholds items; never hard-deletes while referenced.
  5. "Pin for curation" on the PoetryDB shelf card, operator-only, local-only; and a "pinned" mark on the card thereafter.
  6. Providers without retention (Pexels, Unsplash, GIPHY, Bing News): not in this part; the policy flag is read so a later provider is configuration, not code.
  7. Curation.Candidates lists the pin for its page, and a source-record work renderer draws it on the public page: main still withholds non-catalog work items as :unsupported_item, so storage alone cannot meet the public round trip.

Tests: pin, an already-materialized poem found rather than duplicated, author identity through the crosswalk (an unresolved poet gets no identifier), eligible as a work item, offered by Candidates and rendered, withheld after a text change and restored after re-review under the decided supersession rule, retired → withheld, unavailable → still eligible while retention permits and the card says so.

Done: one PoetryDB poem pinned for a pilot page, chosen as a highlight (via P2 or iex), published, and the withhold-and-restore round trip shown with ids.


P4 — work and exemplar candidates in the packet, and the trimming policy

Spec: #253 §5a step 2 in full (Bierce entries in and counted against the content quota, refusal when they exceed a bound, quotas, the meaning bound, tie-break by object id, omitted list on the attempt, out-of-subset candidates do not dirty). Needs only P1: existing catalog artwork fixtures serve for work; P3's pinned works join when P3 lands; poems and books join only once #253 D6's relation exists. Packet version becomes runtime-packet-v2.

Read first: lib/devils_dictionary/curation/runtime/{packet.ex,contract.ex,endpoint.ex,attempt.ex}; Curation.Eligibility for work and exemplar; priv/curation/runtime_bench/packets/*.json and plan.json (the frozen packets must still verify, or be re-frozen with a recorded reason).

Build: candidates of kind work (page-matched catalog artworks; P3's pinned works once P3 lands) and exemplar (accepted illustrates claims) in Packet.build/3; bound meanings as well as candidates, retain every chosen candidate's required meaning, and test real whole On scopes (packet v1 refuses more than 12 meanings independently of its 12-candidate cap; love's whole On scope has 43 senses, 48 packet meanings; 36 counts the lowercase subset only); freeze ids, revisions, rights and claim standing and omitted reasons deterministically; the trimming policy replacing {:too_many_candidates, n}; packet_summary.omitted on the attempt; Contract validation accepts the new kinds and still requires an exact substring for a quotation only. Keep :priority_candidate_unavailable as a refusal.

Tests: quotas and order are deterministic across runs; Bierce entries survive any cut, and too many of them refuse rather than trim; love's whole On scope (43 senses, 48 packet meanings) builds under the meaning bound; the omitted list names every dropped candidate; a packet from the benchmark fixtures verifies under v2 or the re-freeze is recorded; the contract rejects a work highlight with a quote.

Done: a packet for a pilot page with more than 12 candidates and more than 12 meanings builds, freezes, verifies, and its attempt lists the omitted candidates.


P5 — Freshness: the refresh state, admission on visit, the worker

Spec: #253 §7 in full and §6a in full; #198's "Admission and freshness" and "Publication and reader stability" paragraphs (#198 is reopened for this part). The dedicated issue, #259, wins on detail. No model call in this part: the worker completes without inference when the packet hash is unchanged, and otherwise marks the page as needing an editor (P6 turns that into a proposal).

Read first: lib/devils_dictionary/discovery/{policy.ex,run_worker.ex,run.ex,budget.ex} (the pattern); lib/devils_dictionary/curation/runtime/gateway.ex (fence, lease, budget reservation); Curation.Eligibility.evaluate/5 (the fingerprint); config/config.exs Oban queues; WordLive mount.

Build:

  1. Migration: curation_refresh_states with the columns in Curation P5: freshness — the refresh clock, admission on visit, the Stale tab, and the owner-versus-bot rules in code (under #254, #253; reopens #198) #259 §5.1 (§7's last_completed_at, refresh_after, next_retry_at, input_fingerprint, dirty_at, refresh_generation, state, plus the admitted parent, superseded_on_arrival, the dirty sequence, inputs-last-changed history and the lease fence), unique on composition id and configuration version id, plus a partial unique index allowing one queued/running row per composition.
  2. Admission (Curation.Refresh.admit/2): the §7 predicate (pilot allowlist only during MVP; later expansion disabled), in one transaction with the composition and row locks, generation allocation, the published parent captured, and the Oban insert (curation queue, unique per composition). Called on every page visit, plain HTTP and LiveView, bounded; a visit never calls the model. Budget: 30 globally pending review-work units, reserved and released with actual reviewable work; unchanged assessments consume no new review unit. Daily inference occupancy is a separate 30-minute budget. P5 writes no reviewable work and calls no model, so it reserves neither.
  3. Curation.RefreshWorker: build and freeze the packet (P4's builder, or v1 if P4 is not merged), compare the packet summary hash with the stored fingerprint; unchanged → complete (last_completed_at, refresh_after); changed → record needs_editor (the inputs were checked, not judged; P6 consumes it) and complete; failures set next_retry_at with the backoff and never advance last_completed_at.
  4. Dirty triggers: read-time withholding, a claim review decision, a source revision on a referenced record or a newly eligible candidate in the cut, a configuration activation, a pinned work's revision → dirty_at, debounced 60 s, coalesced per composition; a change arriving during a run stays pending.
  5. §6a: suppressed on withdrawal, cleared by "Re-open" (P2); superseded_on_arrival persisted when the published version id changed since admission; generation and lease fence checked on every write; late writes refused after suppression or a configuration change; expired leases recovered with a bound.
  6. The disclosure's four dates: last assessed, checked against the record (render time), inputs last changed, next assessment after, distinct from published.

Tests: fresh visit admits nothing; stale visit admits one; second stale visit admits none; withdrawn admits none; unchanged packet completes without inference and without touching the publication; a failure backs off and does not advance completion; a stale worker's write is refused by generation; a publication during a run yields superseded_on_arrival; plus a deterministic race test for each #259 §5 contract.

Done: on a pilot page, withdrawing a claim dirties it, the next visit queues exactly one job, the job completes, and the disclosure shows the five dates distinctly.


P6 — The proposal path: attempt → model version, attribution, Take, the quality form

Spec: #253 §5a steps 3–5 (attribution rules in full), §6 ("Take"), §6a (proposal never moves the pointer; dismissed proposals are not re-proposed), §10 (the quality form). Reopen #197 for its first half. Needs P2, P4, P5 merged, and P6 live inference requires P7's revised-instruction benchmark go.

Read first: Curation.Runtime.run/3, Runtime.Contract.validate/3, Curation.Compositions.create_version/3 and its note/2 (attribution is never an input: keep that rule and add a separate proposal writer), CompositionVersion and CompositionItem schemas, Curation.Reviews, P5's worker.

Build:

  1. Migration: origin gains :model; selection_origin gains :model; attempt_id on editorial_composition_versions (FK to inference_attempts); a proposal_state or reuse of reviews for dismissed.
  2. Curation.Proposals.write(attempt): from a validated attempt, a version with origin: :model, attempt_id, created_by_actor_id the editor's bot actor (the generic-editor profile gets a bot actor; no dossier), parent the composition's current published version at admission, items with notes signed :model and the actual pinned instruction label with its model tag (never a hard-coded generic-editor v1). Never publishes.
  3. P5's needs_editor → P6 runs the editor through Runtime.run/3 with the frozen packet and writes the proposal; respects the daily budget and review capacity.
  4. The screen: "proposed" tab; "Take" per §5a step 5 (unchanged items keep model attribution; rewritten notes become human; added items human; the proposal stays reachable as parent); "Dismiss" records a rejected review and the arrangement hash, and the same arrangement on the same packet hash is not re-proposed.
  5. The quality form on the proposal view: five yes/no fields per item (relevant, correct meaning, useful reason, not a duplicate, attribution correct) and a free line, stored as the review's note; exportable as Markdown for posting on Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253.
  6. D3's fallback wording for a bot-chosen lead.
  7. Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 D8 (the owner's kept and mine marks, stored on the composition with who and when) and D9 (every editor-chosen item linked to its attempt; the attempt records the candidates offered and those omitted by the cut).

Tests: a proposal never moves the published pointer; Take preserves model attribution on unchanged items and human on rewritten ones; the proposal is the parent; dismissed arrangements are not re-proposed; a proposal written after a publication is flagged superseded_on_arrival; the form is required before Take.

Done: five pilot pages have proposals from real packets; five quality forms posted on #253; at least one taken, at least one overridden with reason, ids on the PR.


P7 — Runtime: #221's fixes, the instruction revision, the benchmark re-run

Spec: #221 (reopen it) A1, S1 and B1 as written, including A1's retry ceiling (the transport_retries CHECK); the condition in docs/curation/runtime-benchmark-2026-09-27.md ("revise the instruction as a new pinned version and re-run this plan"). Independent of P1–P6.

Read first: #221; lib/devils_dictionary/curation/runtime/{gateway.ex,runtime.ex,service_process.ex,bench.ex}; the benchmark record's Findings; priv/curation/runtime_bench/plan.json.

Build: A1 handle {:error, :stale} from Gateway.dispatch/3 and note_retry/2 as a refusal, and derive the transport_retries bound from one place; S1 a stop confirmed only when the parent and every descendant runner are gone and the port is closed; B1 the two vacuous benchmark cells labelled, and Bench.sample reading refused_shape. Then generic-editor-v2: the instruction revised per the findings (the adverse cases), pinned as a new instruction version, the plan re-run on qwen3.5-9b-v1 only, results committed under priv/curation/runtime_bench/results/ and a dated record beside the September one. Stop the service when done; note memory and swap before and after.

Tests: regressions for A1, S1 and B1; the benchmark harness still refuses an unconfirmed unload.

Done: #221 closed with evidence; a v2 benchmark record with its go/no-go for P6.


Checklist

A tick means the part's PR merged (implementation). Live acceptance is stated on each line separately.

Activity

  1. holden commented on Oct 25, 2026

    @holden
    MemberAuthor

    Independent assessment — 25 October 2026

    Verdict: progress C; issue coherence C+. The programme now has a real, publicly readable composition and a merged human editor that uses the same version/review/publication services. Those are useful foundations for agents. It has not yet demonstrated the owner's intended curated encyclopedia workflow: only 1/10 pilot pages has a public opening, no verified bot-to-screen proposal path exists, no curation freshness worker exists on main, and three mistaken assumptions about candidate availability remain in the governing specifications. The main risk is implementing the remaining parts literally: several briefs disagree with the record, each other, or the product's acceptance. Finish the small owner/bot loop before expanding ingestion, hosting, coverage, lists or panels.

    Evidence and limits

    I read #193 and its consolidation comments, #253–#255, #257, #259, #198, #252, #217, #170, #111, #192, #221, the P1/P2 PR descriptions and limitations, and the named main modules and curation documents. Main was e25dd94a1055baf83b957c1b99f46bf60ad0c87c, also #258's merge commit. GitHub confirms #256 and #258 are merged. Code references below are to that head.

    I independently fetched all ten public pilot pages, inspected love in a browser, checked its disclosure, and checked the exact-word love routes and the public operator guard. All ten On pages returned 200; public /ops/curation returned 404. I did not sign in.

    The requested v2 SQL audit could not be executed. This assessment ran on a different host: the supplied workspace is an older Rails checkout, the external-drive Elixir checkout is absent, localhost:4007 and localhost:5434 refuse connections, and the connected Mac hostname did not resolve over SSH. The explicitly read-only Postgres connection attempt reached no database. Accordingly, the candidate numbers below are independently observed public-page proxies, not a falsely certified SQL/Eligibility result. I ran no migrations, tests that create/write databases, inference, seeds, publications or issue edits. The only external write is this report.

    A. #253 §10 acceptance, item by item

    Acceptance item Status Verified position versus the goal
    1. Ten public pilot openings, sourced lead and signed highlight, or honest nothing Partly 1/10: love. It shows Bierce, a Johnson definition and a Milton quotation, with reasons. The other nine have no opening. War and greed are uncurated, rather than completed assessments that found nothing useful. Existing definitions/discovery remain below.
    2. Owner curates three through the screen; withdraws and republishes one Partly P2 code is merged. #258 demonstrates war and greed through the screen on a test partition, love through the service, and coward as a draft. That is not three owner-curated live pages. #255 records love's live withdrawal/republication through the service. No live three-page owner proof is recorded on #257.
    3. Five real generic-editor proposals; five owner quality forms on #253 Not started Main still has manual-only version/item origins; runtime produces attempts, not compositions. Proposed is disabled in the queue. #253 has its specification-review comment, not five quality forms.
    4. A newly pinned PoetryDB poem, withhold/restore proof Not started P3 is unimplemented. A catalog poem tile in #258's test demonstration is not a new pin or its revision round trip.
    5. Five distinct dates; claim withdrawal dirties; two visits admit one run; visible queue Partly Publication and read-time eligibility exist. Love's disclosure shows publication/provenance but none of the four assessment rows. No refresh implementation is on main. The stale queue is disabled.
    6. Human publication preserved; withdrawn page gets no proposal; in-flight result superseded Partly Human publication already uses transactional pointer checks and P2 handles stale forms. Suppression, proposal writing and admission-time ancestry do not exist yet. Normal navigation stability is present; prompt removal from an already-open page is not demonstrated.
    7. Every result has row/attempt IDs; record restores from a bundle Partly Love publicly identifies composition 1, version 1, receipt 3; lead content 1779138/revision 411, Johnson 1807159/23665, quotation sense 1991670/revision 248575. #255 records historical v2 counts. No new curation restore round trip or proposal attempt evidence was verified here.

    Thus 0/7 acceptance items is fully demonstrated end to end; this is not a claim that the underlying persistence work is absent.

    Today love has a canonical satirical lead and a related literary quotation, but no representative-example tile or independently selected work tile. Its notes and selector/publisher are Account #2, not the owner's public name; #255 explicitly accepted these as placeholders. War still has its instance list and catalog shelf without an opening; greed still has senses/discovery without an opening. None demonstrates a completed curation freshness check.

    Database totals: the latest explicit live count evidence is #255's Delivery B: 1 composition, 1 version, 3 publication receipts, the latter being publish/withdraw/publish, not three published pages. Love's current disclosure corroborates those particular IDs; it cannot prove current database-wide totals. #258's composition IDs 1–4 belong to its separate test database. #259's “three live compositions” is an assumption, not evidence.

    B. What an owner can choose now

    Public-page observation, interpreted through Candidates, not a substitute for the unavailable SQL audit:

    Pilot page Bierce / Johnson entry visible Sense quotation available publicly Accepted illustrates example visible Catalog artworks visible
    nepotism B + J yes 0; one under review 0
    greed neither yes 0 0
    war B + J yes 0; 44 instances instead 2
    love B + J yes 0 4
    coward B + J yes 0 0
    patience B + J yes 0 0
    justice B + J yes 0; an instance instead 0
    truth B + J yes 0 0
    hope B + J yes 0 1
    bestiality neither yes 0 0
    Pages / 10 8 10 0 3

    Counts: eight pages expose potential entry leads; zero expose accepted exemplars; three expose catalog works, all artworks (7 displayed rows). Greed and bestiality have none of those three categories, but both have quotations: they are not wholly empty candidate pages. All ten expose at least a quotation. All ten lack one or more of the owner's lead/example/work requirements; none currently has all three selectable categories, because accepted exemplars are absent from all ten public views. Final admissibility, hidden accepted records, and complete shelf cardinalities require v2 checks; I do not certify these proxies as database counts.

    The committed manifests independently contain 2,903 poems, 946 books, 1,644 Met artwork rows and 1,575 Wikidata artwork rows. Those are manifest rows, not counts of works offered on a pilot page or distinct works after overlap. Corpus size is not curation coverage.

    P2 found three structural limits, all confirmed in code:

    1. Poems and books are not related to pages. Their manifest kinds declare evidence: :none; Candidates takes only Artworks.shelf_items/1 rows for work candidates. Consequently 0/10 screens offers a catalog poem or book through that seam, despite the record accepting such references and the reader drawing their tiles. A hand-written service version can select a catalog poem; that bypass does not prepare the owner's screen or the bot's candidate set. Close this through an explicit owner decision in Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 and its P3/P4 handoff about the existing attestation/page-selection relation, not a guessed “uses word = about meaning” claim. Cost: a bounded matching/selection contract and its candidate/eligibility tests; retaining full poem text is a separate P3 cost.

    2. Senses cannot lead. Content leads reference content items; WordNet and Wiktionary supply senses. D3's WordNet → Wiktionary fallback therefore cannot be executed. For the current record, greed and bestiality must have no lead. Either correct the specification to accept this pilot limitation, or explicitly authorize a narrow sense-reference extension with storage, policy, rendering, validation and recovery work. P4 and P6 alone cannot make a sense into a content item. Do not manufacture a duplicate definition to evade the record.

    3. Examples need accepted nominations. instance_of examples and pending illustrates nominations are not selectable exemplars. The owner can nominate/review appropriate existing subjects using the existing Examples flow, then select them through P2; Curation P2 (#257): the owner's curation screen and example tiles #258 proved this on test war. Cost: actual sourced nomination and review work, not 44 automatic approvals. Link the words their sources already identify: mint the 21,544 QIDs WordNet and Wiktionary name, run the identifier rungs lexicon-wide, and link on visit (after #172) #217 improves source-identified entities/sense links; it does not supply a general pool of people to nominate. Exemplars: people, works and media as attributed examples of a meaning — nomination, the why, votes, ranking and the reverse view on a person's page #105 holds nominee search/mint design, while Link the words their sources already identify: mint the 21,544 QIDs WordNet and Wiktionary name, run the identifier rungs lexicon-wide, and link on visit (after #172) #217's folded Why Voltaire is not in the registry: people enter through senses, not through credits — 6,674 persons are word-referents; decide on-demand minting, a people seed, or both #165 discussion still leaves the broader people-seed choice unresolved. Bot nomination writes would additionally require Gate every illustrates nomination until accepted before bots or the public can nominate (#181 decision 6, deferred) #190; selecting already accepted examples does not.

    C. Concrete issue edits required

    These are proposed edits only; none has been applied.

    1. Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 §3, §4c–d, §6, §9 and D3; Curation P2: the owner's curation screen and the example tiles — curate the pilot pages by hand, with exemplars, poems, books and artworks on the page (under #254, #253) #257 §1–3 and §5.3. Replace the false availability statements with: “WordNet/Wiktionary senses are not entry leads or content highlights in the current record; instance lists are not accepted exemplars; catalog poems/books have no page-match seam. Greed currently offers quotation highlights and no lead.” Replace the illustrative greed/war candidate lists accordingly. Record the two decisions in B explicitly. Change the specifications now; keep the safe eligibility/claim boundaries.

    2. Curation MVP implementation plan: seven parts with agent-ready briefs (P1 read compositions, P2 screen, P3 pin, P4 packet, P5 freshness, P6 proposals, P7 runtime) — under #253 #254 discovery paragraph and P3/P4 dependency; Curation P2: the owner's curation screen and the example tiles — curate the pilot pages by hand, with exemplars, poems, books and artworks on the page (under #254, #253) #257 §1. Replace “poems, books and artworks are selectable highlights in P2 without P3” with “catalog references can be stored/rendered, but P2 lists only page-matched artwork rows; poems/books need an explicit page-selection relation.” P4 can use existing catalog artwork fixtures without waiting for P3, but must not claim poems became offerable merely because manifests exist.

    3. Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 §9.6 and §10 poem acceptance; Curation MVP implementation plan: seven parts with agent-ready briefs (P1 read compositions, P2 screen, P3 pin, P4 packet, P5 freshness, P6 proposals, P7 runtime) — under #253 #254 P3. Correct the identity contract before implementation: “PoetryDB has no provider-issued poem ID. Reuse the existing poetrydb_poem namespace and Poetrydb.poem_id/3; it hashes author, title and lines. Changed lines currently mean a changed identity, not a new revision under the same identifier. Decide the permitted supersession relationship explicitly; never deduplicate by title.” Reuse the reviewed author-QID crosswalk; an unresolved poet must not acquire an invented identifier. The existing provider already exposes a durable identity adapter, so check whether a live result is already materialized before claiming every pin creates a previously absent work. P3 must also name its Candidates listing and source-record work renderer: main still withholds non-catalog work items as :unsupported_item. Storage alone cannot meet its public round trip. These are repairs to the existing pin requirement, not grounds for a broader ingestion project.

    4. Curation MVP implementation plan: seven parts with agent-ready briefs (P1 read compositions, P2 screen, P3 pin, P4 packet, P5 freshness, P6 proposals, P7 runtime) — under #253 #254 P1; Curation P1: show one manually published opening on a pilot word page #255 §2.7, §3.1 and implementation notes. Reconcile all P1 assumptions exposed by Curation P1 (#255): one published opening on /on/love #256: “language comes from the scope's lexemes, not page.language_tag; scopes are exact sorted lexeme sets, overlapping lexical-page scopes cannot coexist; Published owns eligibility, the reader maps its result; assessment rows are absent until P5; definition highlights have their own tile; actor labels come from the record.” Retain the approved typography change and actual live five-lexeme love scope, rather than the two-lexeme fixture. P1's temporary work/exemplar withholding was repaired by P2; mark it resolved. Do not reopen P1 to redo that work.

    5. Curation P2: the owner's curation screen and the example tiles — curate the pilot pages by hand, with exemplars, poems, books and artworks on the page (under #254, #253) #257 §3 scope rule; editor's public-facing “Exact-word pages share this” text. Replace it with “Exact-word pages overlap this editing scope and cannot have a separate active lexical-page composition; they do not currently display this composition.” The editor claims sharing, while PublishedReader resolves exact signatures. I verified neither /words/9051/love nor /words/14794/love shows the On opening. Correct the built label now; actual sharing is a product/routing decision, not implied by K3.

    6. Curation P2: the owner's curation screen and the example tiles — curate the pilot pages by hand, with exemplars, poems, books and artworks on the page (under #254, #253) #257 §2.3, §5.1–2; Curation MVP implementation plan: seven parts with agent-ready briefs (P1 read compositions, P2 screen, P3 pin, P4 packet, P5 freshness, P6 proposals, P7 runtime) — under #253 #254 P2. Record the remaining P2 departures: “poem tiles show title/poet/length/link, no lines; no editor height readout; pilot slugs are a configuration-overridable code default, not a new config-file edit; editor URL is /ops/curation/on/:slug; stale publication errors are stale_pointer/stale_parent.” Keep the truthful metadata-only tile until retained text exists. Replace the erroneous instance-based exemplar examples with accepted nominations. The artwork thumbnail change and added exemplar/work tiles are directly within P2's goal. Record heading-start versus full-heading measurements consistently: Curation P1 (#255): one published opening on /on/love #256's accepted 882–910 px and Curation P2 (#257): the owner's curation screen and example tiles #258's 878–906 px do not place the whole heading above 900; avoid claiming that stricter result.

    7. Curation MVP implementation plan: seven parts with agent-ready briefs (P1 read compositions, P2 screen, P3 pin, P4 packet, P5 freshness, P6 proposals, P7 runtime) — under #253 #254 checklist; Curation P2: the owner's curation screen and the example tiles — curate the pilot pages by hand, with exemplars, poems, books and artworks on the page (under #254, #253) #257 Delivery B; Curation P5: freshness — the refresh clock, admission on visit, the Stale tab, and the owner-versus-bot rules in code (under #254, #253; reopens #198) #259 introduction and Delivery B. Say “P2 Delivery A merged at e25dd94; owner demonstration/Delivery B not evidenced.” Separate implementation-merged and live-acceptance status. Replace Curation P5: freshness — the refresh clock, admission on visit, the Stale tab, and the owner-versus-bot rules in code (under #254, #253; reopens #198) #259's assumed “three live compositions” with “read and record actual live composition IDs first; do not invent the missing two.” Curation P5: freshness — the refresh clock, admission on visit, the Stale tab, and the owner-versus-bot rules in code (under #254, #253; reopens #198) #259 must follow Curation MVP implementation plan: seven parts with agent-ready briefs (P1 read compositions, P2 screen, P3 pin, P4 packet, P5 freshness, P6 proposals, P7 runtime) — under #253 #254's rule to tick implementation on merge, not its current instruction to tick while merely opening/reviewing the PR. Keep Curation P1: show one manually published opening on a pilot word page #255 closed: its bounded P1 delivery is evidenced.

    8. Curation P5: freshness — the refresh clock, admission on visit, the Stale tab, and the owner-versus-bot rules in code (under #254, #253; reopens #198) #259 §3 budget, §5.2; Curation MVP implementation plan: seven parts with agent-ready briefs (P1 read compositions, P2 screen, P3 pin, P4 packet, P5 freshness, P6 proposals, P7 runtime) — under #253 #254 P5. Replace “30 review units per day” with “30 globally pending review-work units, reserved/released with actual reviewable work; unchanged assessments consume no new review unit. Daily inference occupancy is a separate 30-minute budget.” This is what Visit-driven curation refresh: coalesced jobs, fenced leases and atomic publication #198 specifies. A no-model freshness check should not be charged as a fresh reviewable proposal.

    9. Curation P5: freshness — the refresh clock, admission on visit, the Stale tab, and the owner-versus-bot rules in code (under #254, #253; reopens #198) #259 §5.1–3 and §5.5–6. Add the missing contracts explicitly: capture the published parent at admission, not worker start; persist the superseded-on-arrival outcome the schema promises; preserve dirty changes arriving during a run; retain inputs-last-changed history after clearing pending dirty state; atomically initialize absent state under a composition lock; recover expired queued/running work with bounded leases; reject late writes even after suppression/configuration change. Add the corresponding deterministic race tests. dirty_at = nil on every completion would lose an update received during that run and erase the requested historical date. Generation checking alone does not distinguish a new dirty event within the same generation.

    10. Curation P5: freshness — the refresh clock, admission on visit, the Stale tab, and the owner-versus-bot rules in code (under #254, #253; reopens #198) #259 §3, §5.2 and §5.6; Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 §7 scope. State how a never-curated allowlisted page gets a composition/refresh identity, since the worker API and FK require one. Require a bounded visit check for the page visits promised by the product; connected-mount-only admission currently excludes plain HTTP visits. State “pilot allowlist only during MVP; later expansion disabled.” Curation P5: freshness — the refresh clock, admission on visit, the Stale tab, and the owner-versus-bot rules in code (under #254, #253; reopens #198) #259's immediate “or ≥ threshold” predicate contradicts D7 and Visit-driven curation refresh: coalesced jobs, fenced leases and atomic publication #198's no-automatic-expansion rule. Also carry configuration-activation and newly eligible in-cut candidate changes into invalidation. Define the versioned transition from P2's full candidate-spec digest to P4's packet digest; they must not silently mean the same fingerprint.

    11. Curation P5: freshness — the refresh clock, admission on visit, the Stale tab, and the owner-versus-bot rules in code (under #254, #253; reopens #198) #259 §2.4; Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 §6a; Visit-driven curation refresh: coalesced jobs, fenced leases and atomic publication #198 reader stability. Replace “withdrawal and lost rights already behave that way” with “fresh loads revalidate; prompt suppression on already-open pages still needs an explicit bounded invalidation path and proof.” WordLive reads the opening during load; I found no curation invalidation subscription that updates an open opening. Keep normal publication navigation-pinned. This is an existing acceptance gap, not a request for live reshuffling.

    12. Curation MVP implementation plan: seven parts with agent-ready briefs (P1 read compositions, P2 screen, P3 pin, P4 packet, P5 freshness, P6 proposals, P7 runtime) — under #253 #254 P4; Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 §5a packet policy. Add “bound meanings as well as candidates, retain every chosen candidate's required meaning, and test real whole On scopes.” Packet v1 refuses more than 12 meanings independently of its 12-candidate limit. Public war/love/justice/truth/hope show 30/43/17/19/54 senses, respectively; merely trimming candidates cannot make those full scopes pass. Resolve whether Bierce consumes the content quota and what happens if priority entries exceed the cap; the existing prose must not imply both unconditional inclusion and an impossible bound. Freeze IDs, revisions, rights/claim standing and omitted reasons deterministically.

    13. Curation MVP implementation plan: seven parts with agent-ready briefs (P1 read compositions, P2 screen, P3 pin, P4 packet, P5 freshness, P6 proposals, P7 runtime) — under #253 #254 P6 dependencies and P7; Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 §11. Add “P6 live inference requires P7's revised-instruction benchmark go, plus P2/P4/P5; use the actual pinned instruction label, not hard-coded generic-editor v1.” Correct P7's references from “A1–A3” to Runtime stage A follow-ups from the #210 audit: stale dispatch refusal, orphaned ollama runners on stop, two vacuous benchmark cells (#195) #221 A1, S1, B1, including the retry ceiling in A1. Keep the five per-item quality questions and forms on Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253; they already have an owner in P6 and must not be dropped as “later evaluation.”

    14. Delivery: curated opening section with five local persona editors and visit-driven refresh #193 consolidation table and current-status body. Update Curated opening section: Bierce-first lead, sourced highlights and inspectable editorial choices #156 to reader/publication delivered by Curation P1 (#255): one published opening on /on/love #256, screen/tiles by Curation P2 (#257): the owner's curation screen and example tiles #258, exact-scope reader without a binding table; Curation configurations: global default, domain panels and future personal views #201 to default activated, profiles still proposed; Persona panel pilot: five versioned editors, fixed quorum and ten-page evaluation #197 to generic proposal slice P6 first, panel/evaluation later; Visit-driven curation refresh: coalesced jobs, fenced leases and atomic publication #198 to reopened, Curation P5: freshness — the refresh clock, admission on visit, the Stale tab, and the owner-versus-bot rules in code (under #254, #253; reopens #198) #259 active specification, not implemented. Remove stale “all children are closed” and September open-PR/installation assertions from present-tense status. Keep Curation persistence: versioned panels, ballots and reviewed page compositions #196's unbuilt runs/ballots/decision authority distinction, Runtime stage A follow-ups from the #210 audit: stale dispatch refusal, orphaned ollama runners on stop, two vacuous benchmark cells (#195) #221's outstanding fixes, Curation transparency: equal votes, reproducible selection, profiles and version history #203 public history, Editorial method: declared perspectives, checkable facts and accountable curation #204 method, Gate every illustrates nomination until accepted before bots or the public can nominate (#181 decision 6, deferred) #190 nomination gate, Exemplar votes: assertion_votes.reason, cached assertion_scores, the vote control and endorse-on-duplicate (#181 build 3) #189 vote UI, Exemplars: people, works and media as attributed examples of a meaning — nomination, the why, votes, ranking and the reverse view on a person's page #105 remaining nominee work, Discovery relation assessor: lite ML and LLM judging of provider results before persona curation #101 assessor, Quote-shelf connect link and provider-evidence prefill (deferred from #212) #222 quote-connect and Grow the verified quote tier by use, not by rebuild: the decision that replaces the wikiquote-pd-v2 reselect #213 quote-growth decision as folded backlog. P2's private version/receipt history is not Curation transparency: equal votes, reproducible selection, profiles and version history #203's public history. I found no evidence warranting relabelling those other folded deliverables complete. Point the hardware handoff to Hosting: move the reader-facing site off the Mac Studio — host choice (Mac mini recommended), release build, and the pins to lift (absorbs #199, #229, #232) #252, which absorbs Local curation portability: measured Mac Studio to Mac Mini handoff #199.

    15. Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 decisions / §10; Curation P2: the owner's curation screen and the example tiles — curate the pilot pages by hand, with exemplars, poems, books and artworks on the page (under #254, #253) #257 Delivery B; Delivery: curated opening section with five local persona editors and visit-driven refresh #193 panel follow-up. Name the unowned prerequisites: owner's public actor label and replacement of accepted placeholder reasons in the three-page hand exercise; the poem-to-page rule in the P3/P4 handoff; whether existing nominee subjects suffice or a bounded people population decision is needed, using Exemplars: people, works and media as attributed examples of a meaning — nomination, the why, votes, ranking and the reverse view on a person's page #105/Link the words their sources already identify: mint the 21,544 QIDs WordNet and Wiktionary name, run the identifier rungs lexicon-wide, and link on visit (after #172) #217's existing ownership. State “panel automatic publication remains off until a held-out comparison with the generic editor using the same quality form and measured reviewer/machine cost.” The quality form and panel evaluation are already specified; their execution is missing, not their conception. Do not turn Naming: 100 broader ideas and eight extensions, plus the earlier top 50 #192's site naming or Lists: editorial buckets of words — Gen Z dating terms, The Devil's Dictionary, words coined by Vonnegut, words from a hearing — curated, derived or both, prioritised on the site #170's lists into curation prerequisites.

    Overall coherence: #253 correctly governs, #254 usefully separates deliverables, and #255/#257 report their discoveries honestly. The fault is that discoveries have not been reconciled into the authoritative acceptance and next briefs. Reopening #198 is right: its admission/fencing/publication contract is still needed. Keep it as that contract and #259 as the implementation issue; closing #198 after only a model-free P5 would wrongly imply its crash/recovery/proposal obligations are all met. #252 is hosting, #217 is source-identifier coverage, #170 is lists, #111/#131 govern page presentation, and #192 is branding; these are coherent separate tracks, not additional predecessors to the bot pilot.

    D. Readiness for bots

    P2 supplies real page scopes, candidate item specs and standings, a faithful public-renderer preview, transactional publication and stale-form handling. P4 must adapt these semantics, rather than create a richer fictitious candidate pool: catalog artworks and accepted claims only where the current screen can offer them; no poems/books without the relation decision. P4 must handle the separate meaning bound noted above.

    P5 must supply stable per-composition admission, a frozen input identity and admitted parent, generation/lease ownership, suppression, backoff, a durable needs-editor handoff, and preservation of changes during a run. A P5 needs_editor outcome means the inputs were checked, not that an editor evaluated their quality; its disclosure must say that until P6 completes an editor assessment.

    P6 still needs model-capable version/item origins and database constraints, a bot actor without reviewer privilege, attempt linkage, exact proposal ancestry, the separate trusted proposal writer, Take attribution, dismissal deduplication, capacity accounting and the required quality form. “Use the same record” is currently true for human publication; it is not yet implemented for bot proposals. P6 should select accepted material, not create claims; adding bot nominations would activate #190 and broaden this slice.

    Runtime go conditions are not met on main. The only benchmark record remains 27 September, Packet still pins generic-editor-v1, Runtime still pattern-matches successful dispatch/retry, service stop checks parent/port rather than all descendants, and Bench still computes the adverse metrics from accepted-result fields. The 9B result is 13/16 validation passes, including three abstentions; the three adverse answers were refused, not three successful selections. Fixed-seed repeats measure timing, not independent editorial judgments. A new pinned instruction, A1/S1/B1 repairs, corrected observability and a 9B rerun are prerequisites to relying on it. No fresh hardware readiness, disk, memory or service-authority check was possible here; September measurements and #252's October inventory must not be presented as today's capacity.

    For an editor today, the table in B is the practical menu: eight pages have historical entry choices plus quotations, greed/bestiality have quotations without leads, three pages add catalog artwork choices, none has an accepted example or a screen-selectable catalog poem/book. A bot cannot solve these selection shortages by better prompting.

    E. Order, scope and the smallest useful sequence

    P5 → P7 → P4 → P6 is a reasonable serial order after fixing the briefs, but it is not a dependency chain: P7 can proceed independently. P6 needs P7's go explicitly. P4 need not wait for P3 to demonstrate existing artwork/accepted-example kinds. If P4 changes packets after P7's rerun, validate the changed contract on real v2 packets before P6 relies on it; do not represent a v1 benchmark as proof of work/exemplar quality.

    The next three parts are not inherently scope creep. Freshness is explicitly in the owner's goal; runtime refusal/cleanup fixes are necessary before using it; packet kinds and deterministic bounds let agents select what humans select. Scope creep would be automatic candidate-threshold expansion in P5, a new daily review allowance, broad Health-page work beyond the specified counters, extra models, a hosting move, personas, lists or automatic claim generation. The minimal counters and five dates already specified are useful evidence; they do not justify an operations redesign. “Routing should work and nothing more” here means the bot proposes an eligible version into the existing screen and the owner decides.

    Smallest sequence to three real pages:

    • Complete P2's bounded owner exercise and record live IDs. For a low-dependency bot proof, nepotism, coward and patience are sensible candidates: all publicly expose Bierce and quotations and are much less polysemous than love/war. Confirm their actual eligible packets and limits on v2 first; do not assert they already pass.
    • Deliver corrected P5 for those allowlisted scopes and corrected P7. No models are loaded merely to prove P5.
    • Deliver P4's existing candidate kinds/bounds and verify the three real packets. Do not block this proof on a general poem/book matcher or a new people seed.
    • Deliver P6's attempt → proposal → queue → Take/edit/dismiss path with attribution, fencing and quality forms. Generate three real proposals without moving published pointers; show one accepted and one edited/dismissed. Then finish the existing five-page/five-form acceptance. Three is an intermediate proof, not permission to lower Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253's bar.

    This gets a bot proposing through the owner's record before P3, hosting or panels. It does not pretend to complete the poem acceptance or the broader encyclopedia goal.

    Ranked recommendations

    1. Reconcile Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253/Curation MVP implementation plan: seven parts with agent-ready briefs (P1 read compositions, P2 screen, P3 pin, P4 packet, P5 freshness, P6 proposals, P7 runtime) — under #253 #254/Curation P2: the owner's curation screen and the example tiles — curate the pilot pages by hand, with exemplars, poems, books and artworks on the page (under #254, #253) #257/Curation P5: freshness — the refresh clock, admission on visit, the Stale tab, and the owner-versus-bot rules in code (under #254, #253; reopens #198) #259 with P2's actual candidate limits and repair the P5 race/budget contracts before implementation.
    2. Finish and evidence the three-page owner exercise on v2, including the public label, reasons, accepted-example availability and the actual counts this remote assessment could not obtain.
    3. Complete corrected P5 and P7, retaining separate input-check and editorial-assessment meanings.
    4. Complete bounded P4 then P6 on three real pages, then the five required owner forms, with automatic publication off.
    5. Resolve the existing poem-pin identity/page-match decision and finish P3's round trip; keep coverage, hosting, lists, branding and panel expansion on their existing separate tracks.

    No code, corpus data, issue bodies or checklists were changed by this assessment.

  2. holden commented on Oct 25, 2026

    @holden
    MemberAuthor

    Reconciliation applied, 25 October 2026

    The owner accepted all fifteen edits in section C of the assessment above. Each is now in the sections it names, with one wording per fact across #253, #254, #257, #259 and #193. Edits 4 and 11 also name #255 and #198, so those two bodies carry the same wording. #255 stays closed. Text only: no code, data or state changed, and no issue was opened.

    Facts used, checked on v2 and at e25dd94: v2 holds one composition (love, composition 1). Of the ten pilot pages, 8 have an admissible lead (greed and bestiality have none), 0 have an accepted exemplar, and 3 have catalog artworks (war 2, love 4, hope 1). Packet v1 refuses more than 12 meanings independently of its 12-candidate cap, and love has 36 senses. PoetryDB has no provider id. The editor's "Exact-word pages share this" label is wrong.

    # Applied to Wording now in force
    1 #253 §3 (table plus "What the record can offer today"), §4c, §4d, §6 wireframe, §9.5, D3; #257 §1–3, §5.3 "WordNet/Wiktionary senses are not entry leads or content highlights in the current record; instance lists are not accepted exemplars; catalog poems/books have no page-match seam. Greed currently offers quotation highlights and no lead." The greed and war examples are replaced: greed now has no lead and quotations; war has Bierce, Johnson, quotations and two artworks, and an exemplar only after an accepted nomination. Section B's two decisions are recorded in D3 and D6 (below).
    2 #254 discovery paragraph, order diagram, P4; #257 §1; #253 §11 "Catalog references can be stored/rendered, but P2 lists only page-matched artwork rows; poems/books need an explicit page-selection relation." P4 depends on P1 alone and uses existing catalog artwork fixtures.
    3 #253 §9.6 identity and revisions, §10 poem acceptance; #254 P3 "PoetryDB has no provider-issued poem ID. Reuse the existing poetrydb_poem namespace and Poetrydb.poem_id/3; it hashes author, title and lines. Changed lines currently mean a changed identity…; never deduplicate by title." Plus: the reviewed author-QID crosswalk, a check for an existing materialized poem first, and P3 now names its Candidates listing and source-record work renderer.
    4 #254 P1; #255 header, §2.7, §3.1 The six as-built facts (language from the scope's lexemes; exact sorted lexeme sets that cannot overlap; Published owns eligibility; assessment rows absent until P5; definition tile; actor labels from the record). The five-lexeme love scope and the typography stand. P1's work/exemplar withholding is marked resolved by P2.
    5 #257 §3 scope rule; #254 P2 item 5 "Exact-word pages overlap this editing scope and cannot have a separate active lexical-page composition; they do not currently display this composition." The built label is still wrong. Correcting it is one string in editor.ex, which P5 already changes, so #259 §4 carries it.
    6 #257 §2.2, §2.3, §4, §5.1–2, §6 test 12, §7.4; #254 P2 "Poem tiles show title/poet/length/link, no lines; no editor height readout; pilot slugs are a configuration-overridable code default, not a new config-file edit; editor URL is /ops/curation/on/:slug; stale publication errors are stale_pointer/stale_parent." Heights are recorded as heading start–end: love is 882–910 px after P1 and 878–906 px after P2. The heading starts within 900 px; the whole heading does not. #253 §6a also uses the stale_pointer name now.
    7 #254 checklist; #257 header, Delivery B, §10; #259 header, §7.5, §8, §10; #193 status "P2 Delivery A merged at e25dd94; owner demonstration/Delivery B not evidenced." A tick means merged, and live acceptance is stated separately on each line. #257's Delivery A boxes are ticked; its Delivery B box is not. In #259, "three live compositions" became "read and record actual live composition IDs first; do not invent the missing two". #259 now says "PR open" on review and a tick only on merge.
    8 #259 §3, §5.2; #254 P5; #253 §7; #198 "30 globally pending review-work units, reserved and released with actual reviewable work; unchanged assessments consume no new review unit. Daily inference occupancy is a separate 30-minute budget." P5 writes no reviewable work and calls no model, so it reserves neither.
    9 #259 §5.1–5.3, §5.5, §6 tests 16–22 All seven contracts are explicit, each with a deterministic race test: the parent captured at admission (admitted_parent_version_id); superseded_on_arrival persisted; dirty changes during a run preserved (dirty_seq/run_dirty_seq); inputs-last-changed history (inputs_last_changed_at); absent state initialised under a composition lock; bounded lease recovery (lease_token, three attempts); late writes refused after a suppression or a configuration change.
    10 #259 §2.2, §3, §4, §5.2, §5.4, §5.6, test 23; #253 §7 A never-curated allowlisted page has no composition and so no refresh identity. It gets one at provisioning: the screen's first Save draft now, the bot actor in P6. Every visit gets a bounded check, plain HTTP included. "Pilot allowlist only during MVP; later expansion disabled." Configuration activation and newly eligible candidates now feed invalidation. The fingerprint is versioned (candidates-v1 now, packet-v2 from P4), so a stored fingerprint never silently equals one of another version.
    11 #259 §2.4; #253 §6a; #198 reader stability "Fresh loads revalidate; prompt suppression on already-open pages still needs an explicit bounded invalidation path and proof." That path is not in P5; it stays open on #198, and #198 stays open after P5.
    12 #253 §5a step 2; #254 P4 "Bound meanings as well as candidates, retain every chosen candidate's required meaning, and test real whole On scopes," citing love's 36 senses. Bierce entries count against the content quota. If they alone exceed a bound, the packet is refused rather than trimmed, and P4 may put another rule to the owner. IDs, revisions, standing and omitted reasons are frozen deterministically.
    13 #253 §5a step 4, §11 table and paragraph; #254 diagram, P6, P7 "P6 live inference requires P7's revised-instruction benchmark go, plus P2, P4 and P5; it uses the actual pinned instruction label." P7 now cites #221 A1, S1 and B1, A1 including its retry ceiling. The five quality questions and forms stay on #253, owned by P6.
    14 #193 status (rewritten for 25 October), a new folded-children table, delivery row 6, hardware lines #156: reader/publication delivered by #256, screen/tiles by #258, exact-scope reader without a binding table. #201: default activated, profiles still proposed. #197: generic proposal slice P6 first, panel/evaluation later. #198: reopened, #259 active specification, not implemented. The September open-PR and installation assertions are gone. The body's table supersedes the 21 October comment's "closed" table. The other folded deliverables keep their backlog status, and hardware points to #252, which absorbs #199.
    15 #253 §10 (a new "Prerequisites, each with its owner" list, plus the quality-gate line), D4; #257 Delivery B; #193 status The owner's public actor label and replacement of the placeholder reasons; the poem-to-page rule in the P3/P4 handoff; existing nominee subjects or a bounded people decision, through #105 and #217. "Panel automatic publication remains off until a held-out comparison with the generic editor using the same quality form and measured reviewer/machine cost." #192 and #170 are not prerequisites.

    Section B's two decisions, recorded with defaults (the owner can change either on #253):

    • D3, a sense as lead. Default: Johnson where the page has his entry, otherwise no lead, so greed and bestiality have none. The alternative is the owner's to authorize: a narrow sense-reference extension with its storage, policy, rendering, validation and recovery work.
    • D6, the poem-to-page rule. Default: neither the screen nor the packet offers a catalog poem or book until the owner decides an explicit page-selection relation in the P3/P4 handoff.

    The two new decisions in #253 §12:

    Next: P5 from the reconciled #259, Delivery A.

  3. holden commented on Oct 25, 2026

    @holden
    MemberAuthor

    Independent audit — 26 October 2026 (Europe/Warsaw)

    Verdict: ready to start P4; not ready to start the integrated P6 delivery or permit a first live bot proposal. Grades: reconciliation B; P5 C; P7 D; programme readiness C−. These grades assess completed delivery on main and the published installation, not just promising PR code. Remote main and the Mac checkout are both e25dd94a1055baf83b957c1b99f46bf60ad0c87c: P5 #261 and P7’s first PR #260 are open and unmerged. P5’s table/worker are absent on main and v2; P7’s instruction revision and rerun have not been delivered. Reconciliation substantially repaired the briefs, but a few contradictory remnants remain. The owner’s three-page exercise is also incomplete. P4 has an implementable brief and its P1 dependency; it need not wait for poem pinning, P5 or P7.

    Scope and evidence

    I read the current bodies and comments of #253, #254, #257, #259, #193 and #221, the original assessment and reconciliation comment, the supporting #255/#198 bodies, the PRs, main source/tests, and the two delivery branches (20b7114 for P5; 10cc43c for P7 repairs). Live PostgreSQL is 18.2, port 5434, database devils_dictionary_v2. Direct inspection used SELECT only with default_transaction_read_only=on; the existing page/candidate/packet readers also ran through a dedicated connection confirmed read-only, with the application not started. No model was called. No v2 migration, data edit, server restart, service stop/start, launch-script edit, issue-body edit or checklist change was made. Public GETs were limited to the requested love and Stale URLs.

    Independent precommit: 19 doctests, 2,867 tests, 0 failures; exit 0 (ExUnit 365.1 seconds, seed 687153; private worktree remained clean). It ran in a detached external-drive audit worktree at main, with deps/data and each replay gzip symlinked, using exactly MIX_ENV=test MIX_TEST_PARTITION=_audit DD_NO_OBAN=1 mix precommit. Branch test totals below are explicitly author-reported, not independently rerun. Useful source anchors: main Packet, main Runtime, P5 Refresh, P7 stop repair. A green main suite does not test either unmerged delivery.

    A. Specification reconciliation

    “Present” means I found the requested text in its named issue sections, not merely in the summary comment. Planned work is distinguished from behavior already on main.

    Edit Text and cross-issue consistency Match to main / remaining qualification
    1 — candidate reality Present in #253 §§3,4c–d,6,9/D3 and #257 §§1–3,5.3. Greed is quotation-only; instances are not accepted exemplars; senses are not content leads. Confirmed by Candidates and live reads: greed has 0 leads, 7 quotations; no pilot offers an accepted exemplar.
    2 — catalog availability / dependencies Present in #254’s discovery paragraph/P4, #257 §1 and #253 §11. P4 needs P1 alone for existing artworks. Matches main’s depiction-based artwork seam; poem/book manifests declare evidence: :none. #257 §1 still initially calls the combined catalog inventory a shelf “served today”; its following correction and §3 correctly restrict what is page-matched.
    3 — PoetryDB identity / public round trip Present in #253 §9.6/§10 and #254 P3: hash identity, changed-text identity, crosswalk, existing-materialization check, Candidates and renderer required. Matches Poetrydb.poem_id/3 and identity_record/1. Source-record works remain :unsupported_item; P3 is specified, not delivered.
    4 — P1 as built Present in #254 P1 and #255’s header/§3.1 correction: lexeme language, exact sorted scope, eligibility owner, dates deferred, definition tile, actor labels. Matches reader/main and live five-member scope. Half-cleaned supporting brief: #255 still retains page.language_tag in the §3.1 code example and §5.1, and test 7 still says to provision overlapping On/exact-word compositions. The correction governs, but these old instructions contradict it. P1 need not reopen.
    5 — exact-word label Correct text in #257 §3, #254 P2 and #259 §4, with the implementation explicitly deferred to P5. Not fixed on main: editor.ex:691 still says “Exact-word pages share this”. Fixed on #261’s branch only. The original assessment asked for the built label correction; this remains a partial delivery.
    6 — P2 departures / height Present in #254 P2 and #257’s named sections: metadata-only poem, no height readout, code-default allowlist, actual route and stale errors; 882–910 / 878–906 px distinguished. Matches source. Those heights are recorded historical measurements; I did not independently remeasure screenshots in this audit.
    7 — merged versus live acceptance Present in #254 checklist, #257 Delivery B/acceptance, #259 and #193. #261 is correctly unticked and open. Matches Git/main and live counts. #257’s “steps 1–3 / 4–7” wording refers to its older seven-step comment, while its current §8 list has five steps: a residual numbering mismatch, not completed acceptance.
    8 — budgets Present across #253 §7, #254 P5, #259 and #198: 30 globally pending review-work units; separate 30 minutes/day inference; P5 reserves neither. No P5 on main. Inspected branch calls neither runtime nor review-capacity reservation. Existing runtime limits are not proof of a P6 review-work allocator.
    9 — seven freshness contracts All present in #259 §§5–6 and carried by #253/#254: admission parent, durable supersession, dirty sequence/history, locked initialization, bounded recovery, late-write fences. Implementation and deterministic regressions exist on #261, not main; see B.
    10 — identity / HTTP / allowlist / fingerprint Present in #259 and #253 §7. Provision first; every visit including HTTP; pilot-only; versioned fingerprint transition. Branch implements candidates-v1, HTTP admission and Pilot. Main has none of this. Activation is detected by an absent configuration-version row. #259 explicitly leaves some unhooked candidate/catalog changes to the window.
    11 — already-open readers Present in #259 §2.4, #253 §6a and #198. Fresh loads revalidate; bounded invalidation for open pages remains unproved. Accurately describes main. Keep #198 open after model-free P5.
    12 — packet bounds / trimming Policy present in #253 §5a and #254 P4: both bounds, required meanings, Bierce quota/refusal, deterministic omitted identities. Still future work. Wrong scope number remains: the briefs call whole On love “36 senses”. Its five-lexeme scope currently has 43 senses + 5 lexeme meanings = 48 packet meanings. 36 counts the lowercase love subset. Main’s independent meaning refusal is real.
    13 — P6/P7 dependency and labels Present in #253 §§5a,11 and #254 diagram/P6/P7: P7 go required; A1/S1/B1 including retry ceiling; actual instruction label; five forms retained. Correct as a plan. Main still pins v1 literally; no revised benchmark go exists.
    14 — consolidation #193’s current body/table correctly say #156 reader/publication #256, screen/tiles #258, exact scope without binding table; #201 default active/profiles proposed; #197 generic P6 first/panel later; #198 reopened/#259 specification, not implemented on main. Hardware points to #252 absorbing #199. Those four requested rows read correctly. Residual current-status contradiction: #193’s #221 row and #253 §9.12 still call #221 closed pending reopening; #221 is already open and #254 names open #260. Historical consolidation comments are explicitly superseded.
    15 — owned prerequisites / panel bar Present in #253 §10/D4, #257 Delivery B and #193: label/reasons, poem relation, nominee ownership and held-out panel comparison/cost. Accurately still open. Actor label/reasons have not changed live; automatic panel publication is not authorized by these deliveries.
    D8 — emphasis Present in #253 §12, §6a and #254 P6 item 7: kept items, mine pages, who/when on composition. Specified only: no marks storage or proposal preservation path on main.
    D9 — choice audit Present in #253 §12 and #254 P6 item 7: item→attempt, offered/omitted candidates, one page disclosure. Specified only: attempt ledger exists, but composition items/versions have no proposal attempt linkage or page audit path; packet cut/omissions are unbuilt.

    All fifteen topics and both decisions were added; this is not evidence that their implementations landed. The material remaining inconsistencies are the deferred built label, whole-scope count, supporting #255 instructions, and stale #221 state. #253 §11 also still describes P1 as delivering the disclosure’s dates while its P5 row correctly owns the assessment dates; #255 and main explicitly leave those dates absent until P5.

    B. P5 freshness

    Required check Main / live finding What I verified on #261’s unmerged branch
    Migration, columns and uniqueness No refresh migration on main. SELECT to_regclass('public.curation_refresh_states') returns NULL on v2. Latest applied migration is 20261009120000. One migration 20261025200000: requested clock, parent, supersession, sequence/history, lease/backoff fields plus documented run_periodic and last_outcome_reason; pair uniqueness and partial unique composition index for queued/running. Ecto uses :utc_datetime_usec, whereas the issue table literally says timestamptz; reconcile that type wording.
    Atomic admission and budgets Absent. Refresh.admit/2 wraps advisory composition lock, row initialization/FOR UPDATE NOWAIT, generation, captured published parent and Oban insert in one transaction. A busy lock returns busy; lock timeout is bounded at 1 s. No review/inference reservation is made.
    Unchanged / needs_editor, no model Absent. Worker claims then calls Refresh. Inputs use versioned candidate/standing digest; unchanged compares baseline; periodic-only runs yield needs_editor/due; changed/empty standing yields needs_editor. No inference or publication call. This is a candidate digest, not a P4 frozen packet yet, as reconciled.
    Admission parent / arrival supersession Absent. Parent assigned at admission, compared at completion; superseded_on_arrival persisted. Race test 16 covers publication between admission and claim.
    Dirty changes and history Absent. dirty_seq versus run_dirty_seq controls clearing; history survives. Race tests 17–18 cover change after input read and next admission.
    Recovery / late writes Absent. Queued/running leases, same-generation recovery with incremented token, three-attempt bound/backoff; state/generation/token/expiry and active-configuration checks. Race tests 20–22 cover recovery, suppression and configuration changes.
    Withdraw / reopen Main withdrawal changes publication; Re-open is still the P2 no-op. Suppression is inside withdrawal’s transaction, including a missing-row case; reopening dirties/unsuppresses; publication resets the accepted baseline.
    HTTP visits / pilot-only / initialization races No freshness admission. WordLive calls the visit check for On loads without a connected-only restriction; Pilot enforces ten defaults; missing compositions admit nothing. Real-connection concurrency tests cover absent-row contention, one row/job/generation and busy rows.
    Disclosure, Stale tab, editor banner Love GET 200, but all four assessment rows absent; its existing Published row is not five freshness labels. Stale through tunnel 404. Four rows and explicit “checked, not judged” wording, enabled Stale tab, banner and corrected scope label exist on branch. Tunnel 404 proves access protection, not that the Stale implementation is deployed.
    Suite Independent main result above. Delivery reports 19 doctests, 2,920 tests, 0 failures at 20b7114; I inspected its race/concurrency tests but did not independently run that branch suite.
    Delivery B / owner yes / single restart / script Not delivered: no table, no refresh jobs, no Delivery B completion evidence in #259/#261. Owner approval for the migration/restart could not be established from those records; I did neither. Current launch script SHA-256: 6704fdf2553d6a97b9c08f9e263e0aec545c21b2674f5e101ab4e177fec05291, mtime 21 Oct 21:08:50 CEST. The hash was identical on both reads during this audit. This proves no audit-time edit, not an activation before/after comparison. The prescribed helper’s rollback-on invokes write_script on; restart and overwrites the launch script. “Byte-identical” must be proved, not inferred from that command.

    Live refresh-state values per composition: composition 1, love has no refresh-state relation/row; last_completed_at, refresh_after, dirty_reason and last_outcome are unavailable, not NULL fields in a delivered row. There are no other compositions to report.

    Live Oban inventory, unchanged across the requested GETs: CleanupWorker completed 672; Discovery.RunWorker completed 62, cancelled 2; Quotations.VerifyWorker completed 337. Curation.RefreshWorker: 0 rows; no other curation worker rows. Thus the expected one refresh row/job per existing composition after P5 activation is not evidenced. Ordinary background jobs must not be mistaken for unexpected curation work.

    C. P7 runtime

    Gate Verified on main Delivery-branch evidence / verdict
    A1 stale dispatch/retry refusal Still success-pattern-matches dispatch and retry, so the reported MatchError path remains. #260 changes both to refusal. gateway_race_test.exs deterministically sweeps between admission/dispatch and asserts zero sends; runtime test covers stale retry. Not merged.
    A1 retry ceiling Database CHECK is 0–2; endpoint config can exceed it. #260 pins the code ceiling to 2, permits lowering and refuses raising before admission; gateway/runtime regressions exist. Not merged.
    S1 descendants on stop Parent and listening port only; no descendant confirmation. #260 reads/rechecks process tables, handles descendants/own leftover llama-server runners, rejects unconfirmed stop/start beside leftovers. Tests cover surviving child, SIGKILL orphan, late runner, unreadable table and unrelated server protection. Not merged; I did not stop a real service.
    B1 adverse observability Bench counts accepted-result fields; September’s two misleading zero cells remain on main. #260 reads refused_shape, adds denominators/unobservable nils, distinct answers and separate outcomes, corrects September prose; regression tests exist. Not merged.
    New pinned instruction / config-derived Packet label Packet has literal @instruction_version "generic-editor-v1"; no v2 pin or config-derived instruction selection. #260 explicitly leaves this to a second PR. Missing delivery.
    Dated 9B rerun, results, memory/swap, stop afterwards Only runtime-benchmark-2026-09-27.md and September result exist; no new record/results/go. No revised run to audit; before/after memory/swap and post-run stop cannot be verified for a new run. I did not create one.
    P6 go/no-go September gives a conditional development go, requiring a new instruction and rerun before reliance. Current audit decision: NO-GO for P6 live inference. The requested revised-instruction go/no-go has not been delivered.

    The September numbers are 13/16 validation passes = 10 selections + 3 abstentions; 3 refusals, not 13 successful selections. The ordinary Bierce lead is 6/6; sparse abstentions 3/3; all three adverse answers were refused. Two adverse zeros cannot establish non-use/non-approval when those fields were unobserved. Fixed-seed repeats are not independent editorial judgments. These numbers support the September conditional development recommendation, not an unconditional P6 go.

    The assessment’s grouped prerequisites—new instruction; A1/S1/B1 repairs with corrected observability; 9B rerun—are all unmet on main. Repair code exists in an open PR; that is progress, not closure. #260 reports 19 doctests / 2,883 tests / 0 failures at 10cc43c; this is author-reported and does not supply a benchmark.

    D. The owner’s pages — live numbers

    E. Gates and smallest path to proposals

    Status vocabulary: done = present/verified on main or live as applicable; specified only = an explicit contract exists but main implementation/acceptance is outstanding; missing = the required delivery artifact is absent.

    Gate Status Consequence
    P4 foundation: published reader and four-kind human candidate seam Done P4 can start now; P1 is merged. Existing artworks suffice without P3.
    P4 work/exemplar packet candidates and contract verification Specified only Human Candidates supports them; Runtime.Packet/Contract does not yet provide the P4 round trip.
    P4 deterministic quotas/trimming, Bierce inclusion/counting/refusal, frozen omissions on attempt Specified only Main refuses excess candidates; no cut/omitted audit.
    P4 bound meanings, keep required meanings, whole On tests Specified only Hard refusal at 12 exists, but the required bounded construction does not. Love’s 48 meanings must be handled, not hidden by narrowing the page.
    P6 P2 screen implementation Done Queue/editor/manual publication exist. Owner Delivery B remains open.
    P6 P4 packet delivery Specified only Required before integrated proposal path.
    P6 P5 merged/live freshness Specified only #261 exists; main/live delivery remains open.
    P6 model origin on versions/items and constraints; bot author without reviewer privilege Specified only Schemas/DB still manual-only.
    P6 attempt linkage and admitted-parent/superseded proposal ancestry Specified only Runtime ledger alone is not composition lineage.
    P6 trusted proposal writer with model/config/instruction attribution Specified only No validated-attempt → version writer.
    P6 Take/edit attribution and parent semantics Specified only No Take flow; manual author stamping is not a substitute.
    P6 dismiss/rejected review and arrangement+packet deduplication/capacity Specified only General reviews exist; proposal lifecycle does not.
    P6 D8 kept/mine marks with who/when Specified only Neither storage nor behavior delivered.
    P6 D9 item→attempt and offered/omitted audit reachable from page Specified only No linkage/disclosure path delivered.
    P6 five-question required quality form and five owner forms Specified only Contract exists; UI/storage and completed forms absent.
    P6 P7 repaired runtime + revised instruction + benchmark go Missing Open repair PR; no new pin, rerun, dated go artifact or post-run resource evidence.
    Owner’s label/reasons and three-page screen acceptance Specified only Live acceptance remains open; not a prerequisite to coding P4.

    Today’s actual candidate supply, read through main’s Candidates over each real public On scope (eligible highlight candidates; content can also supply leads, so these are not counts of distinct selections):

    Page Admissible leads Content highlights Quotations Catalog artworks Accepted exemplars Default Packet v1
    love 1 3 25 4 0 refuses 28 candidates; separately 48 meanings
    war 1 3 24 2 0 refuses 28 candidates; separately 37 meanings
    hope 1 5 11 1 0 refuses 16 candidates; separately 58 meanings
    truth 1 2 14 0 0 refuses 16 candidates; separately 22 meanings
    justice 1 3 6 0 0 refuses 21 meanings
    greed 0 0 7 0 0 builds: 7 candidates / 6 meanings
    coward 1 2 4 0 0 builds: 6 / 12
    nepotism 1 2 2 0 0 builds: 4 / 3; 1 pending nomination
    patience 1 2 2 0 0 builds: 4 / 8
    bestiality 0 0 3 0 0 builds: 3 / 9

    The separate meaning counts were measured with a reader-only builder option raising the candidate limit, solely to expose its second refusal; nothing was dispatched or persisted. The differences between human Candidates counts and Packet v1 counts are further reason to verify actual v2 packets rather than infer them from screen counts.

    Ranked recommendations / minimum sequence

    1. Start bounded P4 now, using love, war and hope, the richest current pages and all three artwork-bearing pages. Correct the scope count and resolve both bounds, deterministic cut, Bierce rule and omitted identities; prove build/freeze/verify on whole scopes. No general poem/book matcher or new nomination programme is needed for this step.
    2. Finish the existing owner exercise on three real pages, replacing love’s reasons and label, recording screen-created versions/reviews/receipts and one withdraw/reopen, all three ending published. Keep Curation P2: the owner's curation screen and the example tiles — curate the pilot pages by hand, with exemplars, poems, books and artworks on the page (under #254, #253) #257 open until evidenced.
    3. Land reviewed P5, then perform only its separately owner-approved Delivery B: one migration, one controlled restart, actual script hash comparison, live state/job/disclosure evidence. The helper writes the script, so inspect that fact before claiming unchanged bytes. Keep Visit-driven curation refresh: coalesced jobs, fenced leases and atomic publication #198’s excluded obligations open.
    4. Land P7 repairs, pin generic-editor-v2 with Packet reading the configured instruction, and run the existing 9B plan, recording selections/abstentions/refusals and observability denominators, memory/swap before/after, confirmed stop and a justified go/no-go. Revalidate changed P4 packets before P6 relies on them; a v1 benchmark cannot prove work/exemplar quality.
    5. Deliver the specified P6 path and generate proposals on love, war and hope, retaining the published versions until owner action, with attempt/parent/attribution, Take/dismiss, D8/D9 and required quality forms. Three pages are the first proof; finish the already-required five proposals/five owner forms for P6 acceptance. If a smaller preliminary packet smoke check is useful, nepotism/coward/patience already build under v1; this does not waive P4 or P7.

    This audit posts one report comment only. It does not certify unmerged code as deployed, authorize live activation, close #257/#198/#221, or change the programme’s acceptance bar.

  4. holden commented on Oct 26, 2026

    @holden
    MemberAuthor

    P7: the runtime go/no-go for P6 is no-go (26 October 2026)

    P7 is done. #221's fixes are in #260 (merged as e8f1102, #221 closed with evidence). The instruction revision and the 9B re-run are in #263 (merged as e44122d). The record is docs/curation/runtime-benchmark-2026-10-26.md, and the raw results are priv/curation/runtime_bench/results/2026-10-26.json.

    What ran.

    • September's plan: the same cases, fixtures and 16 calls, under generic-editor-v2. That is the instruction revised per September's Findings, pinned beside v1. Packet reads the instruction from the pinned config, and a mismatch is refused before admission.
    • On qwen3.5-9b-v2: the same 9B artifact and settings as qwen3.5-9b-v1, registered as a new config because a config pins its instruction.
    • In devils_dictionary_runtime_bench only (run 1792976835, attempts 36–51), on the owner's yes.
    • The plan and its nine go checks were committed before the run. Bench.verdict/5 evaluates them, and a check fails on missing or unseen evidence.

    Result: 8 of 9 checks pass. 16 of 16 calls validated: 10 selections, 6 abstentions, 0 refusals.

    Case September, 9B under v1 Now, 9B under v2
    ordinary Bierce lead 6 of 6 Bierce lead 6 of 6
    ambiguous selected, each sense used abstained 3 of 3 ("the definition excerpts are fictional"); the failed check
    sparse abstained abstained
    adverse 3 refusals (duplicate_candidate) 3 valid: Bierce lead, no duplicate, injection not followed (0 / 0 / 0, each seen in 3 of 3)

    Why no-go, and what it does and does not show.

    • The ambiguous fixtures label themselves "Fictional fixture definition:", September's Findings 3.
    • They were identical in both runs, and only the instruction changed. Under v1 the model noticed the label and selected anyway; under v2 it abstains.
    • So the cause is v2 meeting self-labelled text. One fixed-seed judgment cannot say whether v2 is more ready to abstain in general.

    Runtime health was good.

    Until a go, P6's live inference should not rely on this runtime and instruction.

    Next, being prepared. A re-run that changes only the ambiguous fixtures: a different two-sense word, with unlabelled fictional text, frozen with the merged packet v2 (#265). The instruction (generic-editor-v2), the model (qwen3.5-9b-v2) and the go checks stay unchanged. Under packet v2 a definition is offered only at the word itself, so the ambiguous expectation becomes "each quotation under its own sense". The owner will be asked to accept that before the run.

    • If v2 then selects, the label was the trigger.
    • If it still abstains, the instruction needs a v3.
  5. holden commented on Oct 26, 2026

    @holden
    MemberAuthor

    Full MVP infrastructure audit — 26 October 2026

    Verdict: the delivered P4/P5/P7 infrastructure is a credible foundation for implementing the single-editor P6 path with a stubbed runtime, after correcting #264’s incompatible seams; it is not a completed MVP or a go for live bot proposals. My live-inference verdict remains NO-GO. Grades: P5 A−; P7 B+; P4 A−; reconciliation B; P6 readiness C+. Main is 2dc3350626cb2276c8b13819375c01b4af3330bb, also the remote main I checked. Unlike the previous audit, all four implementation PRs are merged and P5 is live. The outstanding problem is no longer absent infrastructure: it is an honestly failed benchmark gate, an unbuilt proposal lifecycle, and a P6 brief whose profile admission, refresh outcomes, immutability, attribution and budget instructions cannot all execute as written. This supports bounded P6 engineering, not autonomous publication, personas or panels.

    Evidence and limits

    I read #253/#254/#257/#259/#193/#221/#262/#264, their comments, PRs #260/#261/#263/#265, main’s implementation/tests, both benchmark plans/results and the 26 October record. I reproduced the three requested CLI dry-runs and read all ten pilot pages through the same candidate/packet seam. Database reads were on PostgreSQL 18.2, port 5434, devils_dictionary_v2, with default_transaction_read_only=on explicitly confirmed. The reader process started only its repository/dependencies, not the application, Oban, endpoint or model. Freeze/verify and benchmark recalculation were read-only. No migration, data edit, live service start/stop/restart, inference, source change, issue-body edit or checklist change was made.

    Independent suite: 19 doctests, 2,979 tests, 0 failures; exit 0 (seed 646998, ExUnit 406.1 s; private worktree clean), run at the audited main in a private external-drive worktree, deps/data and every replay gzip symlinked, using MIX_ENV=test MIX_TEST_PARTITION=_audit3 DD_NO_OBAN=1 mix precommit. The supplied permission covered that isolated test partition only.

    Code evidence below uses the audited revision: Packet, Contract, Refresh, Profiles, foundation constraints/triggers.

    A. Claimed versus delivered

    Delivery Main, regression tests and independent checks Live installation / qualification
    P5, #261 → 9757c61; Delivery B Refresh migration/schema/partial unique index, transactional admission and Oban insert, captured parent, dirty-sequence/history preservation, lease recovery/fences, suppression/reopen, HTTP visits, pilot allowlist and four disclosure rows are present. refresh_race_test.exs covers contracts 16–18/20–22; refresh_concurrency_test.exs uses separate committed connections for admission races. Worker still makes no model call. Migration 20261025200000 is applied. Live uniqueness includes one queued/running row per composition across configuration versions. Love’s clock and job generations are below. Love GET 200, Stale URL through tunnel 404; all four assessment rows are in the initial HTML.
    P7 repairs, #260 → e8f1102 Stale dispatch/retry returns refusal, retry ceiling checked before admission, descendant/leftover-runner confirmation and unreadable-process-table refusal, refused_shape observability/denominators and separate outcomes are present. Runtime/gateway race, service-process and Bench regressions are present and included in my suite. I did not replay destructive/service drills. The #221 completion evidence describes them. At audit time no recorded Ollama pid file, no listener on 11435 and no matching dictionary Ollama/llama-server process were found. This corroborates current stopped state, not every historical stop step.
    P7 instruction/rerun, #263 → e44122d v1 and v2 instructions are separately pinned/tested; build requires an explicit instruction, CLI gets it from the named model config, and Runtime refuses config/packet instruction mismatch before admission. Dated record, raw results and predeclared nine-check verdict exist. I recomputed the summary/verdict with main’s Bench.summarize/3 and Bench.verdict/5: NO-GO, 8/9, 10 accepted selections / 6 abstentions. Benchmark ran in devils_dictionary_runtime_bench, not v2. Its immutable config is qwen3.5-9b-v2, same 9B artifact/settings as v1, new config because it pins generic-editor-v2. v2 has 0 model configs, 0 inference service rows, 0 attempts. A merge does not install that benchmark config into the live DB.
    P4, #265 → 2dc3350 Packet v2 calls Candidates.for_page/1; content/quotation/work/exemplar support, deterministic quotas, required-meaning preservation, both 12 limits, explicit omissions, freeze/registry verification and text-free attempt summaries are present. Contract refuses quotes on works/exemplars. Tests cover the cut, excess Bierce refusal, old v1 compatibility, meaning integrity, work/exemplar contract and fingerprint transition. All 10 actual pilot packets built, froze and verified :ok; repeat builds were identical. All also passed Packet.fits? using the measured benchmark’s 8,192 context / 1,024 output limits (the runtime’s conservative estimator, not a tokenizer measurement). Love/war/hope dry-run hashes match #262’s recorded delivery. Live love’s third refresh records the new fingerprint version. No accepted live exemplar exists, so that kind’s positive evidence is tests, not a fabricated live example.
    Reconciliation plus residual repairs D8/D9 and the fifteen topics remain specified. The built exact-word label is now corrected; #253/#254 now say 43 senses / 48 meanings for full On love; #257 Delivery B numbering and #253’s P1 publication-versus-assessment wording are corrected. Not completely reconciled: #255 still has the old page.language_tag example/§5.1 and “provision both” overlapping-scope test; #193’s “current” body still says freshness unimplemented/only September benchmark, and its #221 row still describes open repair work. #254/#257 also retain “built label still says share this”, although main fixed it. Preserve historical records, but stop presenting these statements as current.

    Live refresh state, SELECT snapshot (timestamps in UTC, as stored):

    Composition / config version State / generation / token Last completed Next assessment Pending dirty / outcome History / parent
    love 1 / 1 idle / 3 / 4 2026-10-26 13:59:58.254863 2026-10-28 13:59:58.254863 dirty_at/reason NULL; needs_editor / fingerprint_version inputs_last_changed_at same completion time, reason inputs_changed; admitted parent 1; superseded false

    Fingerprint prefix is packet-v2/cut-v1/meanings-v2; no failures, next_retry_at or outstanding lease. This is a completed input check, not a completed editorial assessment. Disclosure explicitly says “the inputs were checked, not judged: an editor is needed (the inputs are now identified differently).”

    Jobs: RefreshWorker 3 completed: 4645/gen1, 4646/gen2, 4730/gen3. The first two substantiate P5 Delivery B; the third is P4’s later transition, not an unexpected duplicate P5 run. Other rows: CleanupWorker 672 completed; Discovery.RunWorker 62 completed + 2 cancelled; Quotations.VerifyWorker 337 completed. No curation inference/proposal job exists.

    Activation history: receipt 4 withdraws love and 5 republishes v1 during P5’s proof. The Delivery B comment documents the owner’s yes, named migration, direct bootout/bootstrap rather than the rewriting helper, and one restart. I independently confirmed the migration, resulting clock/jobs, live disclosure and current script hash 6704fdf2553d6a97b9c08f9e263e0aec545c21b2674f5e101ab4e177fec05291, identical to my previous audit’s hash. The exact historical restart count and original approval chat are documented there; SELECT/current process state cannot independently reconstruct them.

    B. Benchmark: what failed, what PR 3 can establish

    The 26 October record and raw JSON agree with my recalculation:

    • 16 calls, 10 selections + 6 abstentions + 0 refusals; all 16 validate. Ordinary 6/6 Bierce leads; ambiguous 3/3 abstentions; sparse 3/3 abstentions; adverse 3/3 selections.
    • 8/9 checks pass. The failed ambiguous_each_sense check requires selection and sense coverage, so zero wrong senses on an abstention cannot pass it.
    • Adverse use/unknown-reference/approval counts are 0/0/0, each with 3 observed answers; these are now evidential zeros.
    • Every repeated case has one distinct answer. Sixteen calls are five judgments including smoke, not sixteen independent editorial decisions.
    • Raw samples all have swap before/after 0. Recorded system free-memory readings are 58% before / 35% loaded / 46% after stop; compression remained elevated. Budget charge 42,667 ms, no reservation outstanding. These are that session’s measurements, not a promise of future capacity.
    • Every scored case packet is runtime-packet-v1. P4 preserves those historical fixtures; this run is not evidence that the model handles the new v2 work/exemplar/meaning representation.

    Attribution of the failure: it is justified to say v2 abstained on these self-labelled fixtures, citing their fictional status. It is not established that the labels alone caused the failure or that the instruction generally became overcautious. The same labels existed under v1, which selected; instruction and fixture interact. The record’s main findings make this qualification correctly. Its Next paragraph and #254’s follow-up comment overstate the proposed experiment: changing the lemma, fixture wording, packet version and expectation cannot establish “the label was the trigger” or, on another abstention, uniquely establish “the instruction is at fault.”

    PR 3 is planned, not delivered. No PR 3/new v3 plan/result is on the audited remote main or in the PR listing I checked. A predeclared follow-up using unannounced synthetic evidence and actual packet v2 is a reasonable remedy for the P6 readiness test, provided its changed variables are recorded and it is not presented as a single-variable causal experiment.

    Restating the ambiguous expectation as select the two quotations, each under its own pinned sense matches v2: content candidates now have the screen’s lexeme meaning; quotations have their own single allowed sense. Requiring definitions at sense ids would test behavior P6 cannot accept. The check must require the intended quotation ids/coverage and a selection, not merely “wrong_sense = 0”; revise the plan and harness/test together. Because v2 already fixes each candidate’s allowed meaning, this tests useful selection/coverage within the contract, not unconstrained model disambiguation. It does not replace the five owner quality forms.

    Own decision: NO-GO for P6 live inference. A go still requires the owner-approved, precommitted follow-up plan and packets, unchanged pinned instruction/model or explicitly recorded new versions, a completed run passing every declared non-vacuous check, separate outcomes/observability, memory/swap and confirmed stop evidence, and verification that real v2 pilot packets fit and validate. The last packet-building requirement now passes; the new model run has not happened. Passing it would permit controlled owner-reviewed proposals after P6’s implementation/activation gates, not automatic publication or a panel.

    C. P6 gates against #264 and #253

    “Done” means main/live evidence exists. “Specified only” means the behavior is requested but unbuilt; “missing” identifies an absent prerequisite or an executable contract not actually supplied.

    Gate Status Evidence / remaining work
    Same candidate supply as screen, bounded real On scopes Done Packet/Candidates shared seam; ten frozen/verified/deterministic pilot packets, D below. The packet is a documented cut of the screen’s supply, not every candidate at once.
    P5 needs-editor signal and admitted-parent/lease state Done Love row 1; RefreshWorker/Refresh retain generation/token and admission parent.
    Needs-editor → fenced model attempt → proposal handoff Specified only Worker currently claims, checks inputs and completes; no Proposals module. Completing P5 first and later calling run(composition_id) does not retain its write authority.
    Bot barred from reviewing/publishing/activating/admitting Done for existing authority boundary Authority.actor/2 returns account user actors; DB human/reviewer guards protect decisions/receipts. Preserve these while introducing bot authorship.
    Generic-editor bot/profile and no-dossier admission Missing compatible admission seam No generic-editor row in v2; only five proposed persona identities. Profiles.admit/4 and DB profile guard require a sourced deceased-person dossier.
    Model-config selection and explicit editor enablement in live config Missing executable binding ConfigurationVersion is manual-only with no model-config field; live local_model_configs empty; runtime authority file still names benchmark DB. #264 mentions enablement but defines no binding.
    Trusted writer, model origins, attempt FK/idempotency, captured parent Specified only Main version/item origins manual-only; no attempt_id/proposal writer. Runtime attempt + summary exists as substrate, not a proposal.
    Proposal completion, suppression/configuration/late-worker safety Specified only P5 fences exist; writer must participate in them rather than start an unfenced second phase.
    Take/Edit attribution and proposal parent Specified only Current create_version stamps all notes to the human and parents to latest version. It cannot perform the specified Take unchanged.
    Dismiss/rejected review + arrangement/packet dedup Specified only Generic rejected reviews exist; no proposal_dismissals or exactly-once proposal lifecycle.
    D8 kept/mine storage/behavior Specified only No mark/mode columns or UI; proposed item-column updates conflict with immutability. See concrete outside-cut example in E.
    D9 stored offered/omitted attempt audit Done in runtime substrate Packet.summary/1 includes ids, policies, required meanings, offered/omitted and hashes without excerpts.
    D9 version/item→attempt and reviewer-only attempt page/public attribution Specified only No version attempt FK, attempt route or Take disclosure. v2 has 0 inference attempts.
    Required five-question quality form and five form exports Specified only CompositionReview currently has reason, not a note/form field; no form/export or completed owner forms.
    30 globally pending review-work units Specified only, contradictory brief Gateway counts live inference-attempt units, not outstanding reviewable proposals. No P6 capacity storage/transactions; #264 test 8 incorrectly says “day bound”.
    P7 benchmark delivery Done Pin/repairs/measured NO-GO delivered.
    P7 live-inference go Missing Recomputed verdict false; no later plan/result.
    Owner’s three-page screen acceptance / public label / reasons Specified only Still one live composition; label and notes unchanged. Blocks acceptance, not writing stubbed P6 code.

    D4 is clear: automatic publication remains off. A future human Take path may publish under human authority; “Proposals never publishes” must not be misread as forbidding that explicit owner action.

    D. Owner’s pages and current candidate supply

    Live v2 has 1 composition, 1 version, 1 review, 5 publication receipts. Only On love, composition/version 1/1, is published; final receipt 5. Selector/publisher is actor 5, human user 2, still labelled Account #2. There are 0 model versions and 0 attempts.

    Placeholder notes are not replaced:

    1. “Bierce's entry leads: the voice this dictionary is named for.”
    2. “Johnson's senses map the word before Bierce turned it.”
    3. “The plainest literary use of the sense, beside the satire.”

    P5’s withdraw/republish did not create two more owner-curated pages or replace P2’s three-page screen exercise. #257 should remain open.

    I reproduced mix dd.runtime packet --scope SLUG --dry-run --instruction generic-editor-v2 for love/war/hope, on a forced read-only connection; all ten rows below were also computed through the same Packet/Candidates API. No model config was invented to make the CLI run.

    In C/Q/W/E columns, values are screen-eligible → packet-kept. Leads overlap the content pool and are not extra candidates. Meanings are kept / total; omitted.

    Pilot page Admissible leads Content Quotations Works Exemplars Candidates kept/available Meanings kept/total; out Candidate omissions
    love 1 3→3 25→4 4→3 0→0 10/32 12/48; 36 22 over_quota
    war 1 3→3 24→4 2→2 0→0 9/29 12/37; 25 20 over_quota
    hope 1 5→4 11→3 1→1 0→0 8/17 12/58; 46 9 over_quota
    nepotism 1 2→2 2→2 0→0 0→0 4/4 3/3; 0 0
    greed 0 0→0 7→7 0→0 0→0 7/7 6/6; 0 0
    coward 1 2→2 4→4 0→0 0→0 6/6 12/12; 0 0
    patience 1 2→2 2→2 0→0 0→0 4/4 8/8; 0 0
    justice 1 3→3 6→4 0→0 0→0 7/9 12/21; 9 2 over_quota
    truth 1 2→2 14→5 0→0 0→0 7/16 12/22; 10 9 over_quota
    bestiality 0 0→0 3→3 0→0 0→0 3/3 9/9; 0 0

    Unused content quota flows into quotations, explaining 4 or 5 quotations despite a base quota of 3. No candidate in these live cuts was omitted specifically for its meaning; the separate meaning bound still drops unused meanings. All selected candidates retain their required meaning. No selectable poem/book or accepted exemplar appeared; the work counts are catalog artworks.

    Reproduced packet hashes:

    • love: 7a6061f8caa4ea5be3a248132d2f88771fb76c1510d61f2ad468502ddf75ec73
    • war: 1fc7c48c86e30892a3ff2685a59392e02f1aea298daae8586c525988aa3fe1a5
    • hope: f21bb3b93a17a1774c3745ee7103e1a3686dc021e2984d6394e64cf3548bbca2

    Love, war and hope remain the richest first three proposal pages. Only love currently has a composition; the other two must acquire exact On-scope composition identities before a P5 visit can enqueue anything.

    E. #264: exact corrections needed before implementation

    These are corrections to the requested slice, not additional features.

    1. Header / §11, merge versus go. Replace “P7’s second PR (the benchmark go)” with “P7 PR 2 is merged with a NO-GO; Delivery A may use stubs, but live inference requires the subsequent approved benchmark GO.” Refer to PR Curation P4 (#262): the packet carries works and examples, bounds meanings, and records its cut #265 for P4; Curation P4: the packet carries works and examples, bounds meanings as well as candidates, and records its cut (under #254, #253) #262 is its issue, not its PR.
    2. §2/§4 labels and §3 instruction. Resolve instruction and model labels from the attempt’s immutable model config: generic-editor-v2 / qwen3.5-9b-v2 for the delivered benchmark, same weights as v1. Remove the hard-coded generic-editor v2 · qwen3.5-9b-v1. Treat the wireframe’s “12 of 36” as illustrative; love’s current measured count is 10 of 32.
    3. §3 trigger API. outcome/3 returns the internal pair, but the worker’s claim/check path returns successful completion and the persisted row; it does not emit a reusable {:needs_editor, reason} job handle. Specify the integration point using the claimed row id, configuration version, generation, lease token and admission parent, not only composition_id. Also consume the durable needs_editor handoff when its input-check row is otherwise fresh; a repeat ordinary visit currently queues nothing until dirty/due.
    4. §4 handoff/write fence. Keep proposal creation under a validated current refresh claim, with suppression/configuration/generation/token checks and atomic persistence/refresh completion as required. Do not call an independent Proposals.run/1 after releasing P5’s lease and assume a runtime fence supplies refresh ownership. A concurrent publication may flag supersession; a withdrawn/suppressed composition must reject the write.
    5. §4/§5 editor enablement. “When the configuration version enables the editor” currently names no field or immutable model binding. Define the exact existing-manifest/new-version contract for enabled editor/config selection and its invalidation; keep it off before the GO. A new instruction/model binding must not silently inherit the old completed-assessment identity.
    6. §3/§4 profile admission. “Generic-editor … Profiles.admit/4; no dossier” cannot work. That function requires a version, source refs and deceased evidence, enforced again by curator_profiles_admission_guard. Include the narrow generic-editor admission/schema rule in P6’s migration, while preserving the persona evidence rules; do not fabricate a death/source dossier.
    7. §4 migration: bot author guard. Adding :model enums/FK does not authorize a bot insert: editorial_composition_versions_human_actor currently refuses it. Specify the model-author guard and attempt/config linkage explicitly, without weakening reviewer/publisher/activation guards.
    8. §4 abstention / §6 test 2. Main RefreshState.last_outcome and its DB CHECK accept only unchanged/needs_editor. “Completes … last_outcome: abstained” therefore needs both schema/constraint and disclosure/outcome changes in the P6 migration, or an explicit separate editor-outcome record. Delete “or abstained” from the version-writing branch: only an accepted selection creates a proposal; an abstention creates no version and does not count toward five proposals. Keep input checks distinct from actual editorial assessments; failed/refused inference must not acquire successful-editor freshness.
    9. §4 proposal_state / test 5. “State becomes taken” attempts to UPDATE an immutable version. Specify a separate lifecycle record or tightly delimited mutable status columns/guard rule; preserve immutable selection contents, ancestry and attempt identity. superseded_on_arrival is also not inherently an exclusive alternative to subsequently taken/dismissed; define that lifecycle instead of an ambiguous four-value bucket.
    10. §4 D8 marks. Keep/unkeep on existing items currently hits editorial_composition_items_guard; Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 says the marks live on the composition. Specify mutable composition-level mark identity/who/when, or the exact permitted metadata update rule without rewriting the selected arrangement. Carry the marks across version changes by stable selection identity, not transient form position.
    11. §4 all-model items versus kept human items. “Items … selection_origin:model” conflicts with carrying an owner-kept item unchanged. A human note is currently required to be authored by its containing version’s human author; copying it into a bot-authored version fails editorial_composition_items_insert. Define copied-item provenance/author validation and preserve the original human signature; only editor-authored choices/notes acquire model attribution.
    12. §4 D8 versus unchanged packet policy. A kept item may not be in the cut. Concrete current example: love’s published item 3, quotation 248575 / quotation:3, is omitted as quotation-248575-3. With unchanged P4 policy a model cannot select it. State whether P6 carries that already-kept eligible reference through the trusted writer, or records a blocked proposal; a mere :kept_item_dropped refusal does not satisfy Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253 D8’s “survives into every bot proposal”. Do not silently change the packet policy or synthesize a model choice to hide this conflict.
    13. §2 Take / §3 manual seam / §6 test 5. The ordinary writer stamps all notes human and parents to the latest version; Take needs a trusted copy path preserving unchanged model authorship and parenting to the actual proposal. Also resolve the existing arrangement_hash + eligibility_fingerprint unique constraint: a truly unchanged copy retaining selection_origin/model signatures duplicates its proposal. Define this in the migration/Take contract; do not defeat it by falsely changing attribution.
    14. §3 budget / §4 run / test 8. Replace “30-unit day bound” with 30 globally pending review-work units, no daily reset; inference remains a separate 30-minute/day ledger. Resolve “reserve when written” versus “reserve before Runtime.run”: reserve capacity atomically before potentially generating reviewable work, commit it to the one proposal, release unused reservations on abstention/refusal/error/stale/duplicate, and release outstanding work once on Take/Dismiss/supersession. Include the durable capacity representation/transaction in the migration contract; Gateway’s live-attempt counter is not this queue.
    15. §4/§5 quality storage. There is no review “note” column or API option. Reviews.decide/4 stores :reason and requires :idempotency_key. State that the fixed form is serialized in reason, or explicitly add the storage/API in the migration; bind it to the exact reviewed version/items. Clarify the opening “form each time” against tests requiring it only for Take/Edit: Dismiss must at least retain its required rejection reason; do not imply nonexistent completed item forms.
    16. §4 dedup/request key / test 13. Retain proposal:<composition>:<generation>, but define one frozen packet/config identity for that attempt and exactly-once write/reservation. Rebuilding changed inputs under the same key cannot be treated as an identical replay; use the existing conflict/stale rules and a new admitted generation where required. Persist suppression on the same arrangement and packet, and enforce attempt-to-version uniqueness under retries/crashes.
    17. §4 trusted input / §3 runtime result. Runtime returns a receipt with attempt_id, not an automatically authorized composition-write capability. Resolve candidate/meaning ids through that attempt’s frozen summary and registry references, revalidate the exact composition/config/scope and current standing, use the quote hash/range rather than invented text, and apply Compositions.one_display_identity/1. Contract checks repeated candidate ids; record-level work/exemplar duplicate-display identity is an additional existing requirement.
    18. §8 live activation/pages. Add registration/selection of the immutable v2-instruction config in v2, and deliberate service authority binding on the owner’s yes: current runtime authority names the benchmark DB and v2 has no config/service rows. Provision exact composition identities for war/hope (currently absent) through the authorized seam before expecting visit-driven runs; Compositions.provision/3 is currently human-author-only. A dirty visit or “Stale tab” cannot create those identities or by itself trigger an unbuilt editor action.
    19. §10 acceptance language. Replace “No path from a proposal to a publication exists” with “No bot/Proposals path moves the published pointer; an explicit human Take/Edit-publish may do so with the quality form and existing authority/CAS gates.” The broader statement contradicts the requested feature and §8 acceptance.

    F. Residual risks, ranked

    1. P1 — an unfenced proposal handoff can defeat the owner’s keep-off decision. P5’s protections are sound in the tested slice, but a later unfenced run(composition_id) could write after suppression/configuration change or lose the original parent. Retain refresh authority through the proposed write; the inference fence alone is insufficient.
    2. P1 — weakening the wrong database guard to “make P6 work”. Bot version authorship, no-dossier admission and mutable lifecycle/marks need narrow rules. Disabling immutability or human-review/publication checks globally would discard the record’s strongest guarantees. The brief must explicitly preserve those boundaries.
    3. P1 — false model/human attribution or broken Take. Current hash uniqueness, copied-human-note checks and manual stamping collide with an unchanged human Take and D8. An agent must not solve this by relabelling model reasons human, owner-kept reasons model, inventing a dossier, or dropping the proposal ancestry.
    4. P1 — benchmark merge mistaken for permission to infer. Main contains a measured NO-GO and only v1 benchmark packets were scored. Merged code, 16 validated calls and zero wrong senses on an abstention do not supply a GO for new v2 packets.
    5. P2 — capacity/retry accounting silently wrong. Treating 30 outstanding review units as 30 per day, leaking reservations on abstention/refusal, or replaying a changed packet under one generation can exhaust capacity or duplicate pending work. Existing Gateway accounting does not solve proposal capacity.
    6. P2 — owner intent cannot reach the model when the kept item was trimmed. Love already supplies a concrete omitted published quotation. Refusing forever is honest but not D8 delivery; fabricated attribution or a silent packet-policy exception would be misleading.
    7. P2 — freshness wording can overstate editorial completion. The current disclosure is truthful. Preserve that distinction when adding abstention/refusal/proposal outcomes; a successful input scan must not mask an offline/refused editor. A fresh needs_editor row is a durable handoff requirement, not a model result.
    8. P2 — open readers and the remaining acceptance gaps. Fresh loads revalidate, but prompt suppression of already-open pages remains Visit-driven curation refresh: coalesced jobs, fenced leases and atomic publication #198’s unproved bounded-invalidation obligation. Only one owner page exists, no accepted exemplar is live, no owner forms exist and no poem pin round trip is complete. These are not evidence that the generic proposal infrastructure failed; they prevent calling the whole MVP or a general agent programme complete.
    9. P3 — stale specifications and experimental causal claims. Delivery: curated opening section with five local persona editors and visit-driven refresh #193/Curation P1: show one manually published opening on a pilot word page #255 remnants, literal labels/counts and “the label was the trigger” could steer an implementing agent toward obsolete code paths or unsupported conclusions. Use current source/rows and the explicit E corrections. Model reasons remain untrusted editorial text; preserve escaped rendering and the local/reviewer-only audit boundary.

    Ranked recommendations and minimum sequence

    1. Correct Curation P6: the first bot proposes into the owner's screen — attempt to proposal, Take, Edit, Dismiss, the quality form, D8 marks and the D9 attempt audit (under #254, #253) #264’s existing contract before implementing it, especially claim handoff, generic-editor admission, immutable record/lifecycle rules, D8/Take attribution and pending-work capacity. Keep the manual writer and human publication authority intact.
    2. Build/review P6 Delivery A with the stubbed runtime on the delivered P4/P5 substrate, including the existing crash/race, dedup, attribution, quality-form and access contracts. A NO-GO does not prohibit this bounded implementation work.
    3. Finish the planned predeclared P7 follow-up on the owner’s direct yes, with v2 packets and a non-vacuous quotation/sense-coverage check, correct causal language, raw outcomes and resource/stop evidence. Until it passes, stop before live inference.
    4. Finish the owner exercise and the scoped live activation prerequisites: label/reasons, real exact-scope compositions, named P6 migration, admitted generic editor, pinned model config and service authority for v2, each under the specified owner approval. Do not infer them from benchmark DB state.
    5. First real proposal: love, then war and hope, the richest verified pages. Run one admitted/fenced attempt, show its linked proposed version without moving the published pointer, and let the owner Take/Edit/Dismiss with the required form. Then complete the already-required five proposals/five forms on Curation agent MVP: what an editor does, what a curated entry contains, how the page looks, how the owner and the bots share one record, and how it stays current (under #193) #253. Panels, automatic publication and broader autonomous agents remain beyond that acceptance.

    This is one audit comment only. It records delivered progress and concrete corrections; it makes no changes to the programme’s code, live data, issue bodies, services or acceptance bar.

  6. holden commented on Oct 26, 2026

    @holden
    MemberAuthor

    Full MVP infrastructure audit — 26 October 2026, repeat verification

    Verdict: P4 and model-free P5 are delivered and form a sound enough foundation to implement the single-editor P6 path with a stubbed runtime, after correcting #264’s incompatible contracts. The MVP is not complete, and live P6 inference remains NO-GO. Grades: P5 A−; P7 B+ for the delivered repairs and honestly reported failed benchmark; P4 A−; reconciliation B−; P6 readiness C+. Main remains 2dc3350626cb2276c8b13819375c01b4af3330bb. The material change since today’s earlier full audit is that the P7 follow-up now exists as open, unmerged PR #266, head a92f66a0018fc12a555e33170c962f09c5e40315, with a committed plan and packets but no measured result. Its direction is reasonable, but its scorer permits definitions instead of the requested quotations and its causal interpretation remains overstated. This is permission to continue bounded engineering under the existing product rules, not evidence that the programme is ready for autonomous agents, panels or public bot output.

    Evidence and limits

    I reread the three original assessments, reconciliation, current #253/#254/#257/#259/#193/#264, supporting #255/#198/#221/#262, delivery evidence, main source/tests, benchmark record/raw results, and PR #266’s plan, packets, scorer and tests. Local and GitHub main agree. The supplied checkout path does not exist on this Mac. I located the existing checkout and reused its isolated audit worktree; no active source or database storage was moved or created.

    Own repeat suite: 19 doctests, 2,979 tests, 0 failures; exit 0, ExUnit 404.8 seconds, seed 413141, worktree clean, using exactly MIX_ENV=test MIX_TEST_PARTITION=_audit3 DD_NO_OBAN=1 mix precommit, with deps/data and every replay gzip symlinked from main. This result covers merged main, not PR #266. I also recomputed main’s benchmark verdict from its raw JSON and exercised PR #266’s scorer in memory with synthetic records; that check called no model and is not a new benchmark.

    Live SQL was SELECT only, PostgreSQL 18.2, port 5434, devils_dictionary_v2, with default_transaction_read_only=on explicitly confirmed. Candidate readers started only dependencies and a read-only Repo, not the application/Oban/endpoint. I reproduced the love/war/hope CLI dry-runs and built/froze/verified all ten pilot packets. Public love returned successfully; public Stale returned 404. I did not sign in, remeasure layout, exercise owner controls, rerun historical service drills, migrate, restart/start/stop a service or run inference. Historical approval and the exact number of restarts are documented evidence, not facts recoverable from today’s SELECTs.

    A. Claimed versus delivered

    Part Verified on main and in tests Verified live / limits
    P5, #261 / 9757c61, Delivery B refresh.ex, refresh_worker.ex, refresh_state.ex, migration 20261025200000, visit admission, pilot restriction, transactional job insertion, admission parent, dirty sequence/history, lease recovery/fences, suppression/reopen and disclosure exist. Refresh worker, race and real-connection concurrency tests are present and included in the suite. No model call. Migration applied; one refresh row for love; generations 1–3 completed. First two jobs support P5 Delivery B; third is P4’s fingerprint transition. Love’s four assessment rows render through the tunnel. Public /ops/curation?tab=stale is 404.
    P7 repairs, #260 / e8f1102 Stale dispatch/retry returns refusal; retry ceiling is checked before admission; stop confirmation includes descendants/leftover runners and refuses an unreadable process table; adverse observability uses refused shapes and denominators. Gateway/runtime/service-process/Bench regressions exist. Current 4007 listener is PID 61348; no listener was shown on 11435. I did not perform a service stop to repeat S1. Historical real-service drill is documented in the delivery/benchmark record.
    P7 revision/rerun, #263 / e44122d v1/v2 instructions pinned separately; CLI resolves instruction from model config; runtime refuses mismatches before admission. Record and raw samples present. My recalculation is NO-GO, 8/9, 10 selections + 6 abstentions, zero validation refusals. Run belongs to devils_dictionary_runtime_bench. Live v2 has 0 local model configs, 0 inference services, 0 attempts. Benchmark registration is not live activation.
    P7 follow-up, #266 Not on main. Open PR contains runtime-bench-v3, v2 packets for five cases including the labelled control, and scorer/tests. The PR reports 2,985 tests; that is author-reported, not my main-suite result. Planned/unmeasured, awaiting owner approval. No GO artifact. Detailed defects/limitations in B.
    P4, #265 / 2dc3350 runtime/packet.ex uses Candidates.for_page/1; all four kinds, both 12 bounds, required meanings, deterministic quotas/omissions, v1 compatibility, registry revalidation, dry-run and fingerprint transition exist. packet_v2_scope_test.exs, packet/contract and refresh tests cover them. Every pilot packet built, froze, verified :ok, repeated identically and passed Packet.fits? at context 8,192/output 1,024. This is the conservative estimator, not measured tokenization or inference. Works are real catalog artworks; exemplar-positive coverage is tests because none is live.
    Reconciliation and five residual repairs D8/D9 and the substantive candidate/identity/clock/budget distinctions are specified. Built exact-word label is corrected; whole love scope now says 43 senses / 48 meanings; #257’s current Delivery B numbering and #253’s publication-versus-assessment wording are corrected. Supporting #255 still instructs page.language_tag and “provision both” overlapping scopes. #193 still calls freshness unimplemented and September the only benchmark, with #221 repairs outstanding; #253 §9.12 likewise still describes the repairs/rerun as unfinished. #254/#257 still describe the old built label as unfixed. #254’s P4 checklist still says PR open despite the merge. These are current-document contradictions, not missing implementation.

    Refresh row snapshot, UTC: row 1, composition/configuration-version 1/1; state idle, generation 3, token 4, admitted parent 1, superseded false; completed 2026-10-26 13:59:58.254863, next check 2026-10-28 13:59:58.254863; outcome needs_editor / fingerprint_version; no pending dirty, lease, failure or retry. Inputs-last-changed is the completion timestamp, reason inputs_changed. Fingerprint namespace is packet-v2/cut-v1/meanings-v2. This says “inputs checked, editor needed,” not “model has judged the page.”

    Jobs, SELECT inventory: RefreshWorker 3 completed, IDs 4645/gen1, 4646/gen2, 4730/gen3. CleanupWorker 672 completed; Discovery.RunWorker 62 completed / 2 cancelled; Quotations.VerifyWorker 337 completed. No proposal/inference job exists. These totals remained the same in the verification snapshots.

    Activation evidence: receipts 4/5 record P5’s withdraw/republish of love. #259 Delivery B records the named migration, owner’s yes, one direct bootout/bootstrap and matching launch hashes. Current launch-script SHA-256 is still 6704fdf2553d6a97b9c08f9e263e0aec545c21b2674f5e101ab4e177fec05291, matching earlier evidence. I verified resulting state and current bytes, not the original approval conversation.

    B. Benchmark and my runtime decision

    The 26 October record and raw results agree with my recomputation: all 16 planned calls validate; ordinary selects Bierce 6/6, ambiguous abstains 3/3, sparse abstains 3/3, adverse selects 3/3. Only ambiguous_each_sense fails. Adverse non-use/non-approval zeros are observed over three answers, unlike September’s unobserved cells. Fixed-seed repeats give one distinct answer per case; sixteen calls are not sixteen independent editorial judgments. Raw swap is zero before/after all calls; the record documents memory before/load/stop and confirmed stop. I did not recreate that session.

    The NO-GO is correct; a fixture-only causal explanation is not proved. The model explicitly cites fictional definitions, but the same labelled text selected under v1. The supported conclusion is “v2 abstained on these self-labelled fixtures, citing fiction.” Its self-explanation does not isolate the cause. All scored packets in that delivered run are runtime-packet-v1, so it also does not establish behavior under P4’s new meaning bindings.

    PR #266 is a reasonable follow-up direction, with three corrections/qualifications before its criteria are accepted:

    1. Its actual criterion is weaker than “each quotation under its own sense.” The scorer at the audited PR head credits either a quotation’s sense or a declared definition, including a definition placed at m0. I loaded this code in memory and supplied a synthetic ambiguous selection lead c1→m0, highlight c2→m0, no quotations. It reported senses_covered: 3/3; with otherwise passing synthetic records, all nine checks returned GO. Its test even explicitly accepts a definition in place of one quotation. The plan discloses this alternative, so it is not a hidden code/plan disagreement; it does differ from the requested quotation test. To measure that test, require both expected quotation candidates c3/s1 and c4/s2 in each accepted selection, and make missing either fail. Keep definitions at their packet-v2 word meanings.
    2. The control does not justify the plan’s three causal verdicts. Labelled bank under v2 helps compare packet formats with bank’s prior v1 result. Unlabelled crane versus labelled bank still differs in lemma, definitions, quotations and source. “Crane selects/bank abstains → label was the trigger” and “both abstain → instruction is at fault” are not established. “Both select” supports success on both new packets; it does not isolate one packet-format subchange. Report observations and the acknowledged confounds.
    3. “No go check reads the control” is inaccurate. No case-specific editorial check targets it, but every_planned_call_measured, no_validation_refusal and swap_growth_below_pause inspect totals/outcomes across its calls too. A validated control abstention can be non-gating for quality; a refusal, missing call or unmeasured swap can still fail the run. The sample note also retains “four judgments” although the control adds a fifth case.

    The quotation restatement does match P6’s v2 task: definitions belong at the word/lexeme meaning; quotations carry their fixed sense. It tests selection and coverage of useful quotations, not free disambiguation, because the allowable meaning is already fixed. It does not itself test works/exemplars or owner-rated curatorial quality. Existing contract tests supply structural coverage for those kinds; the five real proposals and owner forms remain essential.

    Own decision: NO-GO for live P6 inference. A GO still requires accepted, precommitted criteria that actually measure the stated task; the owner-approved run on frozen v2 packets with the pinned instruction/model; every declared check passing with observable denominators, resource readings and confirmed stop; and completed P6 implementation/activation gates. Real pilot packet construction/verification/estimated fit now passes. PR #266’s plan or a future merge alone is not a GO. I ran no model.

    C. P6 readiness against #264 §§3–4 and #253 §5a, §6a, D4/D8/D9

    “Specified only” means requested but unbuilt. “Missing” identifies an absent prerequisite or an executable contract the brief does not supply.

    Gate Status Evidence
    Packet uses screen supply, bounded candidates/meanings and recorded omissions Done Candidates.for_page/1, Packet.for_page/2; ten actual pilot builds/freezes/verifications; D. It is a declared cut of screen candidates, not the entire screen.
    P5 needs_editor signal, generation, lease and admission parent Done Refresh.claim/check/complete; live row 1.
    Needs_editor → model → proposal under retained write authority Specified only RefreshWorker.perform/1 completes the input check; no Proposals module. Completion releases the claim; a later unfenced composition-only call is insufficient.
    Bot cannot review/publish/activate/admit Done for existing authority boundary Authority.actor/2 returns the account’s user actor; enabled DB review/publication human guards. Preserve this when permitting model authorship.
    Generic-editor bot/profile without persona dossier Missing compatible admission Profiles.admit/4 and enabled curator_profiles_admission_guard require sourced/deceased evidence. Live has only five proposed personas, no generic editor.
    Editor enablement and immutable model/instruction binding Missing executable contract/live binding ConfigurationVersion is manual-only; #264 names enablement without defining it; no live model configs/services.
    Trusted proposal writer, model origins, attempt linkage and original parent Specified only Versions/items admit manual origin only; version authors must be human; no attempt column/writer on main.
    Take/Edit preserves note signatures, proposal parent and exact reviewed content Specified only Current manual writer stamps authenticated human notes and parents to latest version. Hash uniqueness and human-note guard need compatible copy rules.
    Dismiss and arrangement+packet dedup, retry idempotency Specified only General append-only reviews exist; proposal dismissal/lifecycle/attempt uniqueness do not.
    D8 kept/mine marks with who/when, carried across proposals Specified only No mark storage/behavior. Items and versions are immutable; current published love quotation is outside the packet cut.
    D9 attempt page, item linkage, offered/omitted audit and safe disclosure Specified only Runtime attempt summaries already support omitted identities; no proposal/item link or /ops/curation/attempts/:id route. Proposed tab remains disabled.
    Required item quality form, storage/export, five actual owner forms Specified only No form/UI/export; Reviews.decide/4 stores reason, not note. #253 has no completed evaluation forms.
    Thirty globally pending review units and exactly-once release Missing executable accounting contract Gateway accounts active inference work, not unresolved proposal review. #264 calls it a day bound and names no durable proposal-capacity representation.
    P7 repairs/pinning and measured delivered verdict Done Merged code; reproduced 8/9 NO-GO.
    Runtime permission to perform live P6 inference Missing No passing v2-packet follow-up; #266 is open/unmeasured.
    Owner’s three real screen-curated pages, label/reasons Specified only / acceptance outstanding One live composition; placeholder reasons and Account #2 remain. This does not block stubbed engineering.

    D4 remains unambiguous: a bot never moves the published pointer. An explicit owner Take/Edit-publish can do so under the existing human review and compare-and-switch gates.

    D. Owner’s pages and candidate supply

    Live record: 1 composition, 1 version, 1 review, 5 publication receipts. Only On love, composition/version 1/1, is published; final receipt 5. Selector/publisher is actor 5, human user 2, label Account #2. No model version or attempt exists. War and hope have no composition identity yet.

    The three placeholder notes are unchanged: “Bierce’s entry leads: the voice this dictionary is named for”; “Johnson’s senses map the word before Bierce turned it”; “The plainest literary use of the sense, beside the satire.” P5’s withdraw/republish is not the missing P2 three-page owner screen exercise. #257 should stay open.

    Counts below come from today’s read-only Candidates/Packet run. Each kind is eligible on screen → kept in packet. Leads overlap content. Meanings show kept/total; omitted.

    Pilot Admissible leads Content Quotations Works Exemplars Candidates kept/available Meanings kept/total; out Candidate omissions
    love 1 3→3 25→4 4→3 0→0 10/32 12/48; 36 22 over_quota
    war 1 3→3 24→4 2→2 0→0 9/29 12/37; 25 20 over_quota
    hope 1 5→4 11→3 1→1 0→0 8/17 12/58; 46 9 over_quota
    nepotism 1 2→2 2→2 0→0 0→0 4/4 3/3; 0 0
    greed 0 0→0 7→7 0→0 0→0 7/7 6/6; 0 0
    coward 1 2→2 4→4 0→0 0→0 6/6 12/12; 0 0
    patience 1 2→2 2→2 0→0 0→0 4/4 8/8; 0 0
    justice 1 3→3 6→4 0→0 0→0 7/9 12/21; 9 2 over_quota
    truth 1 2→2 14→5 0→0 0→0 7/16 12/22; 10 9 over_quota
    bestiality 0 0→0 3→3 0→0 0→0 3/3 9/9; 0 0

    Unused quota flows forward, explaining more than three packet quotations. All retained candidates keep their required meaning. No candidate here was specifically excluded for meaning bounds; separate unused meanings were trimmed. Works are catalog artworks, not selectable poems/books; no accepted exemplar appears.

    Reproduced frozen hashes: love 7a6061f8caa4ea5be3a248132d2f88771fb76c1510d61f2ad468502ddf75ec73; war 1fc7c48c86e30892a3ff2685a59392e02f1aea298daae8586c525988aa3fe1a5; hope f21bb3b93a17a1774c3745ee7103e1a3686dc021e2984d6394e64cf3548bbca2.

    Love, war and hope are the richest first three proposal pages, with love the smallest first activation because its identity already exists. P5 cannot admit a never-provisioned page merely because it is on the allowlist.

    E. Exact corrections to #264 before implementation

    These repair the requested slice; they do not add features.

    1. Header/§11 dependency: replace “P7’s second PR (the benchmark go)” with “PR Runtime P7 PR 2 (#254): generic-editor-v2 pinned beside v1, the 9B re-run and the go/no-go for P6 #263 is merged with NO-GO; stubbed Delivery A may proceed after contract corrections; live inference requires the approved follow-up GO.” Name Curation P4 (#262): the packet carries works and examples, bounds meanings, and records its cut #265 as P4’s PR; Curation P4: the packet carries works and examples, bounds meanings as well as candidates, and records its cut (under #254, #253) #262 is its issue. PR Runtime P7 follow-up (#254): the 9B re-run on unlabelled ambiguous fixtures, frozen with packet v2 #266 is open, not a fulfilled dependency.
    2. §2/§3 labels/counts: derive instruction/model from immutable config/attempt, not literal generic-editor v2 · qwen3.5-9b-v1. Delivered benchmark binding is generic-editor-v2 / qwen3.5-9b-v2. Mark counts illustrative; love currently offers 10 of 32, not 12 of 36.
    3. §3 trigger: outcome/3 computes the pair internally; the worker completes and returns :ok. Specify the integration using row id, configuration version, generation, token and admitted parent, including consumption of an already-fresh persisted needs_editor handoff. An ordinary repeat visit currently admits nothing until dirty/due.
    4. §4 handoff: replace the independent run(composition_id) assumption with an editor step that retains or reacquires explicit refresh authority and fences every proposal/completion write against suppression, generation/token/lease and active configuration. Runtime’s inference fence is not P5 ownership.
    5. §4/§5 enablement: define the immutable configuration/manifest contract selecting the editor and model config, its new-version invalidation and off-before-GO state. “When configuration enables the editor” names no implemented binding.
    6. §3/§4 profile: Profiles.admit/4 cannot admit a no-dossier generic editor under present code/DB rules. Include a narrow generic-editor admission rule in the scoped migration; preserve persona dossier/deceased-evidence rules and human admission authority.
    7. §4 migration/model authorship: adding enum values and an attempt FK does not bypass editorial_composition_versions_human_actor. Specify the narrow bot-author/attempt/config guard while keeping review/publication/activation authority human.
    8. §4 abstention/§6 test 2: last_outcome: abstained is rejected by both RefreshState’s enum and the live CHECK. Specify the schema/disclosure/editor-outcome contract. Remove “or abstained” from the version-write branch: abstention writes no proposal and counts toward none of the five; refusal/error must not acquire successful editorial freshness.
    9. §4 lifecycle/§6 test 5: a proposal version cannot currently UPDATE to “taken”; versions are immutable. Specify separate lifecycle storage or tightly limited status mutability while keeping selection/ancestry/attempt immutable. Arrival supersession is also not necessarily exclusive of later Take/Dismiss.
    10. §4 D8 storage: setting keep/unkeep on existing items hits the immutable item guard. D8 says composition-level marks; identify stable selection identity, actor/time and version carry-forward without rewriting the arrangement.
    11. §4 attribution: “all items selection_origin:model” conflicts with copying owner-kept choices/notes unchanged. Present human-note checks require the containing version’s human author. Specify trusted copied-item provenance and note-author validation; preserve human signatures and attribute only model-authored words to the model. The live item-insert guard also requires note_author_label to equal the author actor’s own label; coordinate that with the pinned instruction/model display instead of supplying an arbitrary label.
    12. §4 kept items versus unchanged packet policy: love’s current published quotation, item 3, is omitted as quotation-248575-3 / over_quota. A model cannot select an id it was not offered. Define trusted carry-forward of an eligible kept reference or explicitly resolve this product conflict. Refusing :kept_item_dropped forever does not implement D8’s “survives into every bot proposal”; do not silently change P4 policy.
    13. §2 Take/§3 manual writer: ordinary manual creation stamps notes human and parents to latest version. Take needs the trusted copy path preserving model notes and parenting to the actual proposal. Resolve the existing unique composition/configuration-version/arrangement-hash/eligibility-fingerprint index for an unchanged copy; do not evade it by falsifying attribution.
    14. §3 budget/§4/§6 test 8: replace “30-unit day bound” with 30 globally pending review-work units, no daily reset; retain separate 30-minute/day inference. Reconcile “reserve when written” with “reserve before run”: atomically reserve capacity before generating reviewable work, commit it to one proposal, release unused reservations on abstention/refusal/error/stale/duplicate, and release resolved/obsolete work once. Specify durable accounting and recovery.
    15. §4 quality storage: reviews have reason, no note; Reviews.decide/4 requires an idempotency key. Specify fixed-form serialization in reason or the explicit storage/API change, bound to the reviewed version/items. Resolve “form each time” against tests requiring it for Take/Edit; Dismiss must retain its required rejection reason.
    16. §4 dedup/§6 test 13: proposal:<composition>:<generation> needs one frozen packet/config identity and exactly-once version/reservation writes. Changed inputs under one key are a conflict/stale case, not an identical replay. Bind dismissals to arrangement and packet, enforce one proposal per accepted attempt and retain the proper generation on retry/recovery.
    17. §3 result/§4 trusted writer: a runtime receipt/attempt id is not an unrestricted write capability. Resolve ids through that attempt’s frozen summary and registry references, verify composition/scope/config/standing and quote hash/range, and apply one_display_identity/1. Distinct candidate ids alone do not prevent duplicate displayed works/exemplars.
    18. §8 activation: explicitly include v2 registration/selection of the immutable model config and deliberate service-authority binding on the owner’s yes; the current authority file names the benchmark DB. Provision exact On-scope identities for war/hope through an authorized seam; current provisioning is human-author-only. Visiting an absent identity or a fresh needs_editor row does not start the unbuilt editor.
    19. §10 acceptance: replace “No path from a proposal to a publication exists” with “No bot/Proposals operation moves the published pointer; an explicit human Take/Edit-publish may do so with the quality form and existing authority/CAS checks.”

    F. Residual risks before agents

    1. High — authority/immutability shortcuts. A composition-only asynchronous handoff can write after keep-off/configuration change; loosening human guards or item/version immutability globally to accommodate P6 would discard current protections. Keep the original claim/parent and narrow every new permission.
    2. High — false provenance or broken owner intent. Take hash uniqueness, copied-human-note checks and out-of-cut kept items already collide with the brief. Relabelling model reasons human, owner reasons model, inventing a persona dossier or quietly dropping a kept item would silently mislead the owner.
    3. High — experimental GO that does not measure the promised task. Delivered evidence is NO-GO on v1 packets; Runtime P7 follow-up (#254): the 9B re-run on unlabelled ambiguous fixtures, frozen with packet v2 #266 can currently pass without either quotation. Correct criteria and causal language before measurement. Merge status, validation success and vacuous wrong-sense zeros do not authorize live proposals.
    4. Medium — freshness/capacity/retry confusion. Input-check completion must not become successful editorial freshness after refused/offline inference. A fresh needs_editor row remains unfinished editorial work. Daily-reset review capacity, leaked reservations or changed packets under one request key can duplicate or strand work.
    5. Medium — reader and acceptance gaps. Fresh loads revalidate; prompt removal of withdrawn/rights-invalid content from an already-open reader remains unproved under Visit-driven curation refresh: coalesced jobs, fenced leases and atomic publication #198. Only one owner composition exists, no accepted exemplar or completed owner forms exist, and P3’s poem pin/retention/withhold-restore round trip is absent. Those limit full-MVP/general-agent readiness even though P4’s structural delivery is real.
    6. Lower — stale instructions and misleading historical claims. Delivery: curated opening section with five local persona editors and visit-driven refresh #193/Curation P1: show one manually published opening on a pilot word page #255/Curation MVP implementation plan: seven parts with agent-ready briefs (P1 read compositions, P2 screen, P3 pin, P4 packet, P5 freshness, P6 proposals, P7 runtime) — under #253 #254 remnants can send an implementation agent down obsolete paths. Keep model reasons escaped/untrusted and D9 raw audit detail reviewer-only; do not turn source text or model assertions into authority.

    Ranked recommendations and smallest route to the first real proposal

    1. Correct Curation P6: the first bot proposes into the owner's screen — attempt to proposal, Take, Edit, Dismiss, the quality form, D8 marks and the D9 attempt audit (under #254, #253) #264’s executable contract first, especially refresh ownership, no-dossier admission, bot authorship, immutable lifecycle/marks, Take/D8 attribution and pending-work accounting. Apply only the already-required reconciliation corrections.
    2. Build/review stubbed P6 Delivery A on merged P4/P5, with the specified race/retry/suppression, attribution, Take/Dismiss, D8/D9, quality-form and capacity checks. Preserve manual writer/publication authority.
    3. Finalize the predeclared Runtime P7 follow-up (#254): the 9B re-run on unlabelled ambiguous fixtures, frozen with packet v2 #266 criteria before its owner-approved run, require the actual quotation coverage if that remains the goal, then retain raw outcomes, resource/stop evidence and every declared check. Stop before live inference unless it earns a GO.
    4. Complete the existing owner exercise and named activation prerequisites, including genuine actor label/reasons, exact composition identities, scoped migration, generic-editor admission, immutable live model binding and service authority. These require their stated owner approvals; this audit does not supply them.
    5. First proposal: love, then war and hope. One admitted/fenced run writes a linked proposed version while leaving the publication untouched; the owner reviews with Take/Edit/Dismiss and the quality form. Finish the required five real proposals/five forms before calling P6 accepted or moving to the wider agent programme.

    The minimum dependency sequence is contract corrections → stubbed P6 implementation/review/merge plus benchmark GO → separately approved live migration/profile/model/authority activation → one love proposal → explicit owner decision. War/hope identity provisioning and the remaining owner acceptance work follow the same existing seams. No new feature is needed to reach that first proposal.

    This is one report comment. No code, issue body/checklist, live record or service was edited; test activity was confined to the authorized private partition/worktree.

  7. holden commented on Oct 26, 2026

    @holden
    MemberAuthor

    P7: the final runtime verdict for P6 is go (26 October 2026)

    This supersedes the morning's no-go above. #266 (merged as 9ccc427) re-ran the plan with three changes:

    The instruction (generic-editor-v2) and the model (qwen3.5-9b-v2) were unchanged. The owner approved the run, including the both-quotations rule for the ambiguous check that the audit called for.

    Result: all 9 checks pass. Run 1793029211, attempts 52–70: 13 selections, 6 abstentions, 0 refusals.

    • Crane: lead c1 at m0, with c3 under s1 and c4 under s2, every time.
    • The labelled control: abstained.
    • Adverse: Bierce at m0. Injection counts 0 / 0 / 0, each seen in 3 of 3.
    • Runtime: swap 0 MB. The confirmed stop named and ended the runner.

    Record: docs/curation/runtime-benchmark-2026-10-26-unlabelled.md.

    What the go does not cover, per the record:

    • The ambiguous case no longer tests telling senses apart. Packet v2 binds each quotation to its sense.
    • No check covers whether a lead fits the word. The crane lead's reason called the bird definition "general".
    • The sparse case's pass may rest on its still-labelled text.
    • Each case is one fixed-seed judgment on benchmark text, not a quality evaluation. The quality forms on real pages stay the measure.

    The service is stopped. Live inference continues under #267.

  8. holden commented on Oct 26, 2026

    @holden
    MemberAuthor

    The remaining path (#267, then #264, then the hand-off to the persona phase) is now ordered in one place: #268. It defines "ready to build the agents" as six checkable items, runs #267 and #264 Delivery A in parallel under a stated file ownership, and asks for an independent audit before Stage 0 (correcting #264) starts. It adds no feature and changes no part's brief here.

  9. holden commented on Nov 1, 2026

    @holden
    MemberAuthor

    Closed 1 November 2026: P1, P2, P4, P5, P6 and P7 are merged and live; P3's PoetryDB pin was superseded by the #271 owner amendment of 27 October (poems are shelf items cited by their stored observation). The implementation tracker is now #298 under #193.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions