Version. 0.1
Date. 2026-07-31
Author. Claude.ai (drafting session). Operator: Marvin Percival.
Charter. standing-notes/dunin7-standing-authorization-charter-v0_1 §2 R-3 — manifest bumps via splice-instruction documents, drafted in a drafting session, applied mechanically, checklist verified.
Applies to. current-status/current-status-manifest-v0_76.md.
Produces. current-status/current-status-manifest-v0_77.md. v0.76 is not edited — it stays as a sibling, per the record's filing rules.
Evidence. inspection-briefs/loomworks-b29-b27-b5-engine-inspection-findings-v0_1 (record e037a37); inspection-briefs/loomworks-inspection-corrections-note-v0_1 (40571cf); current-status/dunin7-status-brief-v0_2 (40571cf).
What this bump carries. One walk-audit finding struck as wrong-when-filed. One tag position corrected. A partial numeric refresh, with the excluded numbers named as still owed. Two new §3 entries. Two methodology candidates. One §5 residue resolved, one narrowed, two new.
Replace:
# Loomworks Current-Status Manifest — v0.76
With:
# Loomworks Current-Status Manifest — v0.77
Append after the existing v0.76 status paragraph:
> v0.77 (2026-07-31) absorbs the first day of governed autonomous operation under the charter: the B-25 sizing sweeps, the combined B-29 / B-27 / B-5 engine inspection, CR-2026-159's filing at three versions, and the standing-up of the status brief the charter's §7 requires. One walk-audit finding is struck as wrong when filed. The engine front is unchanged at a317051 — no engine work has been built in this window.
Append after the existing substrate-state block:
> Partial numeric refresh (2026-07-31). The §5 residue filed at v0.76 asked for five numbers. Two are refreshed here, both read from files at engine a317051:
>
> - Alembic head as declared in migrations/versions/: 0102 (0102_graph_stage_2_dependents_index.py), single head, 102 revisions. This is the repository-declared head; whether playground_dev is stamped at it is a different fact and is unread.
> - stele front: main 689de41, tag v0.4.0, working tree clean.
>
> Still owed, and the residue stays open for them: the alembic head stamped on playground_dev; engine suite pass / skip / fail counts; Operator Layer test and route counts at 99f7f64. All three need either a database connection or a run, and the inspection that produced the two above was fenced from both. This is a partial refresh and must not be read as a whole one.
>
> Tag position corrected. registry-completeness-v0_1 points at c4014d6, the --no-ff merge commit — not at a317051. a317051 is one commit later, the CR-2026-158 implementation notes, docs-only. git describe returns registry-completeness-v0_1-1-ga317051. v0.76 and several companion documents phrased the pair so the tag read as sitting on the inspected commit.
Append two candidates:
> State only what you have read back — as a session discipline, not only a surface one. The completion arc's governing rule was written for screens: a surface states only what it has read back. It applies unchanged to the sessions doing the writing. Three instances now: five positions asserted and withdrawn inside one day's drafting of the completion scoping note (v0.2 §3), each an inference where a read was available; a citation anchor asserted without a read (status brief v0.2 §8); and a source characterised from memory rather than opened. Three instances is this project's threshold for promotion. Candidate for the methodology canon. > > Do not state a specific the work does not need. The narrower rule the same day produced, with four instances: a line number that was never read; a commit identifier gestured at rather than looked up; a docstring characterised loosely; and a count of anchors that carried no information the table beside it did not already hold. None of the four changed what gets built. All four had to be caught by the executing session. The common shape is a specific supplied where none was required, and therefore none was verified. The rule that follows: state the anchor, not a count of anchors; quote what was read, not what is believed to surround it. Candidate, paired with the above.
Insert after Entry 127, before §4.
> ### Entry 128 — B-25 sized, and the eighty-three separated out (new v0.77; 2026-07-31)
>
> Two read-only sweeps against engine a317051 sized build-list item B-25 at eleven sites, not three. The walk audit had found one of three instances of the same construction in one file and missed a category of five entirely — routes that record which authentication door an action came through wrongly, destroying the one distinction memory/base.py:69-81 says the two actor kinds exist to carry. Nine of the eleven are invisible from every surface.
>
> One site moved out. compositions.py:73 relabels a Companion-triggered composition as a contributor's on the way out of the engine; the record is honest and the response schema admits only two kinds. A surface defect, not a record write — reassigned to the completion arc's engine rider. Where build list v0.7 and manifest v0.76 §5 disagreed on this, the manifest was right.
>
> A class neither sweep anticipated became its own item. provenance.wasGeneratedBy — declared the memory event id that produced this version — is fabricated at 83 sites across 37 files, and correct at exactly three. Filed as B-29, separated from B-25 because the shapes differ and the blast radius deserves its own acceptance.
>
> Four Operator rulings settled the same day: the five authentication-path sites stay in B-25; the eighty-three become B-29; the composition relabel moves to the engine rider; B-9's gate is unchanged. Governing document: inspection-briefs/loomworks-b25-sizing-sweeps-findings-v0_1.
>
> ---
>
> ### Entry 129 — The combined engine inspection: B-27 resolved, W-16 struck, the seam reshaped (new v0.77; 2026-07-31)
>
> One read-only engine session answered three questions and overturned four positions, one of them the inspection brief's own.
>
> B-27 (walk-audit W-2) is resolved. The terminal commit route for creation doors 1 and 2 was genuinely uncallable at 1aac815 — a dependency demanding a path parameter the route did not declare. CR-2026-158's rider unmounted it entirely at a425ecd; the live door-1/2 path calls the handler in process with all parameters explicit, so the dependency is never resolved and the fault cannot fire. Established from source, explicitly not from the Operator's browser pass — the class of evidence that got W-10's strike withdrawn a day earlier. The completion arc's acceptance gate is no longer blocked by it. What is not established this way is that creation completes end to end; that rests on CR-2026-158's own Checkpoint A.
>
> W-16 is overturned and should be struck rather than scheduled. Version-pinned Manifestation retrieval exists — as a query parameter on the existing route, not as a path segment — and was present identically at 1aac815, the walk-audit engine. W-16 was wrong when filed, not resolved since. Its route enumeration was correct; its method failed, looking for the assertion-side path shape and concluding absence. Its own transcript confirms the mechanism of the miss. The completion arc's engine rider loses one of its four items.
>
> B-5's staleness marker is a wire, not a build. GET /engagements/{eid}/memory-status returns both the derivation version and the source's current version in one call. One subtlety carried forward: memory_has_changed is driven by content-category counts, not raw version delta, so a Manifestation can be many versions behind and correctly flagged current. The room must choose which it shows or show both with the distinction made plain.
>
> B-29 reshaped without changing its count. 83 sites confirmed; 37 files, not 36. "It is one seam" is wrong — three raw-SQL paths bypass append_event entirely, two needing one-line repairs no seam change reaches. The three persons/personal_engagement.py sites are confirmed correct, and the inspection brief's own doubt about them is withdrawn: each pre-mints the identifier, stamps it, and passes the same value to the insert. The field has one meaning in the tree. Nothing reads the field and nothing hashes it, so the repair has an unusually small blast radius for its site count.
>
> One hazard survives B-27's resolution. The dependency chain that caused it is unrepaired — only the broken route was removed. The test suite cannot see it because conftest overrides that dependency. Filed as build-list item B-30.
>
> Governing documents: inspection-briefs/loomworks-b29-b27-b5-engine-inspection-brief-v0_1, inspection-briefs/loomworks-b29-b27-b5-engine-inspection-findings-v0_1, inspection-briefs/loomworks-inspection-corrections-note-v0_1.
>
> ---
>
> ### Entry 130 — The charter's first day, and the machinery it was missing (new v0.77; 2026-07-31)
>
> The standing authorization charter was ratified 2026-07-30. Its §5 replaces interruption with a decision queue, and its §7 puts that queue in a status brief every session appends to. The status brief did not exist. For one working day the queue had no home, so forks defaulted back to interrupting the Operator — four of them, all of which belonged either in the queue or in the drafting session's own judgment. §2 R-1's conversion of halt-before-push to halt-on-anomaly was also not applied; the session halted before every push.
>
> The brief was stood up on 2026-07-31 at current-status/dunin7-status-brief, carrying what landed, what is running, decisions taken with their reasoning and reversal cost, and the queue. The queue carries one entry — seed v0.13's commit, which is an Operator act by F-3 and cannot be delegated.
>
> Recorded as a rollout defect, not smoothed. The lesson generalises: a governance document that replaces a behaviour must have its replacement built before the old behaviour stops being available, or the old behaviour returns by default.
>
> CR-2026-159 (B-29) was drafted and filed the same day, at three versions — v0.1, then v0.2 closing an open determination, then v0.3 removing an anchor count. Each bump came from the executing session's citation audit catching a drafting-session imprecision. The two-role discipline is the mechanism that caught all five, and none reached the work.
Mark resolved:
> The surface-silence class (filed v0.76). [PARTLY RESOLVED and RENAMED at v0.77.] The class is misnamed and undercounted. Silence is a subset of unverified assertion — a surface stating what it has not read back — and the class now stands at twelve or more instances, most of them confident wrongness rather than silence. Of v0.76's four instances: E0005's stale Manifestation stands, handed to the completion arc; the unexposed provenance walk stands; the invisible moved assertion stands; and compositions.py:73 is confirmed as the response shaper relabelling a Companion-triggered composition, with the record honest underneath — reassigned to the completion arc's engine rider. The prior name and count are preserved here with the correction marked alongside, per corrections-preserved.
Insert as new residues:
> W-16 struck as wrong when filed (new v0.77). Version-pinned Manifestation retrieval exists and predates the walk. The walk-audit report and the completion scoping note both carry it as a real break and both owe an amendment. Recorded because a finding that was never true is a different residue from one that was fixed — nothing changed in the engine, and the correction is to the record of the audit.
>
> B-30 — the dependency trap that caused W-2 (new v0.77). api/deps.py's get_llm_client_with_engagement_fallback takes an engagement_id from the path. Any route mounted without that segment that depends on it 422s from birth. It bit once, silently, and the suite could not see it because conftest overrides the dependency. W-2's route was removed; the trap was not repaired. Filed as a build-list item.
>
> B-29 must land before FORAY integration opens (new v0.77). foray/reserved_emit.py's emitter is a no-op today and provenance is not in its payload; nothing hashes the payload or provenance JSONB columns. When integration opens, whatever sits in those columns becomes the anchored composition. An ordering constraint, not a preference.
>
> The eighty-three dead stampers, after CR-2026-159 (new v0.77). Seam option B leaves 83 sites minting a value that is discarded — harmless, and misleading to a reader who assumes it is stored. Sweeping them requires making the field optional across 37 files and collides with B-25 at three sites where the two items are the two arguments of one expression. Queued behind B-25, deliberately.
Insert immediately after the ## Changelog line, above the v0.76 entry:
> - v0.77 (2026-07-31): The first day under the standing authorization charter. Adds Entries 128 (B-25 sized at eleven; the eighty-three separated as B-29), 129 (the combined engine inspection — B-27 resolved, W-16 struck as wrong-when-filed, B-29 reshaped to 37 files and one-seam-plus-three-bypasses, B-5's staleness marker a wire not a build), and 130 (the charter's first day and the status brief it was missing). Corrects the registry-completeness-v0_1 tag to c4014d6. Refreshes two of §1's five owed numbers — alembic head 0102 as declared, stele at 689de41 / v0.4.0 — and names the three still owed. Renames the surface-silence class to unverified assertion and widens it to twelve or more. Adds two methodology candidates on session-level citation discipline. Four new §5 residues. No engine work landed in this window; the front is unchanged at a317051.
Replace the version line with:
*Loomworks Current-Status Manifest — v0.77 — 2026-07-31*
current-status-manifest-v0_76.md is unmodified — git log shows it untouched by this commit.DUNIN7 — Done In Seven LLC — Miami, Florida Current-Status Manifest v0.76 → v0.77 — amendment instructions — v0.1 — 2026-07-31 Eight splices. One finding struck as never-true. Two of five numbers refreshed, three named as owed.