DUNIN7 · LOOMWORKS · RECORD
record.dunin7.com
Status Current
Path change-requests/cr-2026-157-amendment-note-v0_1.md

CR-2026-157 — amendment note — bringing a pre-charter change request current — v0.1

Version. 0.1 Date. 2026-08-03 Author. Claude.ai (drafting session). Operator: Marvin Percival. Amends. change-requests/cr-2026-157-ova-slice-1-non-member-contribution-pathway-v0_1filed 2026-07-28 and never executed. Charter. standing-notes/dunin7-standing-authorization-charter-v0_1, ratified 2026-07-30 — two days after that change request was written. Grounding. The freshness check run against the engine at 5c24d65; the change request itself, read in full. Status. This note is read alongside CR-2026-157 v0.1, and overrides it where they differ. The change request is not reissued — its design is sound and unchanged, and reissuing 300 lines to correct procedure would obscure that.


1. Why this exists

A freshness check confirmed every anchor in CR-2026-157 still resolves. Every function, table, module and route it names exists with the shape described; nothing it depends on being absent has been built; no design decision is contradicted. Its verdict — executable as written — was correct on the question it was asked.

It was asked the wrong question, and that is the drafting session's error. The instruction said: for every anchor it names — file, line, function, schema, route — confirm it still resolves. It said nothing about whether the change request's own procedures still conform to how this project works, and the change request predates the charter by two days.

An anchor check is not a discipline check. Three procedures in that change request would now cause real harm, and none of them is an anchor.


2. The three overrides

O-1 — playground_dev is never touched. Step 1's verification is replaced.

CR-2026-157 §8 Step 1 reads: "Verification: suite green, alembic upgrade head + full down/up on playground_dev."

playground_dev is the live production database. A full down/up migration cycle against it would exercise every downgrade in the chain on live data. This is the single most consequential line in the document and it is not a defect in its reasoning — it is a fence that did not exist when it was written.

Replaced with: the migration is verified on a throwaway database created for the run and dropped after, on the walk-audit dev pattern — spare port, inline environment, .env untouched, live processes untouched. playground_dev and playground_test are never connected to at any step.

This applies to the whole change request, not only Step 1. Any step, checkpoint or verification naming a database means a throwaway.

O-2 — the checkpoint structure follows the charter.

CR-2026-157 §8 reads: "Checkpoint A halts for Operator review; Checkpoint B halts for tag," and §11 adds "halt before every push."

All three are pre-charter. Charter R-2 authorizes merge, tag and push, and halts a Checkpoint A only for items needing the Operator's browser or passkey.

Replaced with:

Checkpoint A reports and proceeds on a clean report. Its live verification — synthetic engagement, issue, claim, contribute, verify held and attributed, verify the fence, admit, revoke — is run by the executing session on a throwaway stack, not by the Operator against the dev environment. The synthetic engagement is created and destroyed within the run, and the change request's own prohibition stands and is strengthened: never E0060, E0007, E0030, E0006, and never any engagement in a live database.

Halt and queue at Checkpoint A only on: the add-assertion caller enumeration revealing a disturbed authority pattern; the stub migration rippling beyond its single call site; a new failure outside the Step 0 baseline set; or any charter §6 anomaly.

Checkpoint B merges --no-ff, tags, and pushes. Authorized under R-2 including the push. Deployment remains never a session's act (F-1) — and this change request reaches no serving perimeter in any case.

O-3 — the kickoff points at the record, not at Downloads.

CR-2026-157 §11 reads: "Read /Users/dunin7/Downloads/cr-2026-157-... and execute it."

A Downloads path is a staging copy that may be stale, absent, or superseded — the hazard standing-notes/loomworks-standing-note-executable-document-versions exists to name. The record is the canonical home.

Replaced with the kickoff at §5 below.


3. Baseline corrections

Engine baseline. 1aac8155c24d65, tag stand-up-sequence-v0_1. Forty-one commits and six merges since the change request was written. Step 0's reconciliation item stands and is now essential rather than routine.

The baseline is not green. One long-standing unrelated failure, test_stele_router_mount.py::test_stele_router_mounts_and_begin_resolves_end_to_end. The change request's §9 item 1 says "suite green at every step" — that criterion cannot be met and must not be chased. Replaced with: no new failure against a baseline failure set recorded by name at Step 0.

Migration number. The change request said "next head after Step 0 reads it; do not assume 0102." That caution was correct and has now resolved: 0102 is the current head, so the new migration is 0103. Step 0 confirms rather than assumes.

Line-number drift, from the freshness check. Three files have moved by single digits: contributions.py (~5 lines), cross_engagement_memory.py (~2–6), onboard.py (~8–9). Every other anchor is exact. Step 0 re-confirms all of them; the shapes are unchanged.


4. The count that must not be trusted

The change request refers throughout to "the 63 membership-gated reads" — in its consumed decisions, §5, and acceptance gate items 3 and 9.

That figure was counted at 1aac815 and has not been re-established. The freshness check attempted a recount and correctly declined to report one it could not stand behind. Six engine merges have landed since, several of which added read endpoints.

This is the same class as three figures corrected in the build list this week — a count that grows every time a read endpoint is added, written into a document that does not turn over as fast as the thing it counts.

Override: wherever the change request says the 63 membership-gated reads, read every membership-gated read, enumerated at Step 0 and reported. Acceptance gate items 3 and 9 are satisfied against that enumeration, not against sixty-three.

Gate item 3 survives intact on its own terms — it already asks for a sample plus a fence test asserting the resolver path is untraversed, which is a structural check rather than a per-route one. The structural half is what actually holds when the count moves, and it should be treated as the primary evidence.


5. Claude Code kickoff


CR-2026-157 — OVA Slice 1, the non-member contribution pathway. Execution session.

CR: loomworks-record/change-requests/cr-2026-157-ova-slice-1-non-member-contribution-pathway-v0_1.md
Amendment, read WITH it and overriding it where they differ:
  loomworks-record/change-requests/cr-2026-157-amendment-note-v0_1.md

Read both in full before starting. The change request was written 2026-07-28,
two days before the charter; the amendment brings its procedures current. Its
design is unamended.

Charter dunin7-standing-authorization-charter-v0_1 governs.

Standing fences, and the first one overrides the CR's own text:
  playground_dev and playground_test are live databases and are NEVER touched —
  including by the migration verification, which the CR as written directs at
  playground_dev. Throwaway databases only, dropped after.
  Commit to a branch; check the current branch before the first commit.
  Stage by explicit path; never git add -A.
  Deployment is never yours.
  Append the outcome to current-status/dunin7-status-brief at close.

The engine baseline is NOT green. Record the baseline failure set by name at
Step 0; "no new failures" is the criterion, not "suite green".

Where the CR names a count of membership-gated reads, enumerate it at Step 0
and report it. Do not carry the figure.

Checkpoint A reports and proceeds; Checkpoint B merges, tags and pushes. The
amendment's section 2 names what halts.

6. What this note does not change


7. The lesson, recorded

A freshness check that verifies anchors does not verify discipline. This change request's every file, line and function still resolved; its procedures would have run a migration down/up cycle against production.

The rule that follows: when a change request predates a standing document that governs how work is done, the check is against that document, not only against the code. The engine drifted forty-one commits and the change request survived it. The project's own rules drifted further, and that is what nearly bit.


DUNIN7 — Done In Seven LLC — Miami, Florida CR-2026-157 — amendment note — v0.1 — 2026-08-03 The anchors all held. The fences were the thing that had moved.