research: multi-AI feedback packets verbatim preservation (2026-04-30)

docs/backlog/P2/B-0112-stale-2026-04-27-project-file-internals-bleed-out-cleanup-2026-04-30.md

@@ -0,0 +1,128 @@
+---
+id: B-0112
+priority: P2
+status: open
+title: Stale 2026-04-27 project file internals-bleed-out cleanup
+tier: discipline-cleanup
+effort: S
+ask: deepseek-flag-via-aaron-forward
+created: 2026-04-30
+last_updated: 2026-04-30
+trigger: any tick that touches memory/project_laptop_only_source_integration_*.md OR any tick that scopes work into ../scratch / ../SQLSharp / ../no-copy-only-learning-agents-insight integration
+composes_with:
+  - feedback_no_copy_only_learning_from_sibling_repos_aaron_2026_04_30.md
+  - project_laptop_only_source_integration_scratch_sqlsharp_features_or_designs_high_priority_2026_04_27.md
+tags: [discipline-cleanup, no-copy, sibling-repos]
+---
+
+# B-0112 — Stale 2026-04-27 project file internals-bleed-out cleanup
+
+## What
+
+The 2026-04-27 project file
+`memory/project_laptop_only_source_integration_scratch_sqlsharp_features_or_designs_high_priority_2026_04_27.md`
+predates the 2026-04-30 no-internals-bleed-out discipline
+(`memory/feedback_no_copy_only_learning_from_sibling_repos_aaron_2026_04_30.md`).
+Its body contains characterizations of `../SQLSharp` and `../scratch`
+internals that the new discipline forbids in Zeta documents.
+
+A `status_note:` was added to the file's frontmatter in PR #916
+flagging the situation, but Deepseek's review correctly noted
+the prose-flag-without-mechanical-trigger anti-pattern: "Queued
+where? There's no backlog row, no deferred note in an active
+trajectory, no scheduled follow-up... a known gap flagged in
+prose, never actioned."
+
+## Why this is filed (not deferred indefinitely)
+
+The discipline is fresh; doing the cleanup later when it's stale
+is harder than doing it now. But landing the cleanup-PR mid-round
+mid-incident is substrate-rate-incorrect — too many in-flight PRs
+already, plus the cleanup needs careful work to extract the
+generalizable pattern findings without the specific internals.
+
+So: file as B-0112 with an explicit trigger condition. Future-Otto
+hits the trigger naturally when next touching this file or scoping
+work into the named sibling directories.
+
+## Trigger condition
+
+Any tick where one of the following is true:
+
+1. The agent edits `memory/project_laptop_only_source_integration_*.md`.
+2. The agent scopes new work into `../scratch`, `../SQLSharp`, or
+   `../no-copy-only-learning-agents-insight`.
+3. The TS+Bun expert baseline (Gate B per task #351 + the
+   no-copy-only-learning composition Deepseek named) is being
+   drafted and would benefit from references to those siblings.
+4. A future round explicitly drains P2 backlog or files
+   substrate-rate audit with bandwidth available.
+
+When any trigger fires, the agent does the cleanup pass:
+
+- Re-read the body of the 2026-04-27 file.
+- Identify each section that contains characterizations of
+  `../SQLSharp` or `../scratch` internals.
+- For each section, decide: (a) generalize-and-keep (rewrite to
+  the privacy-class generalized-about framing per the no-copy
+  discipline), (b) move-to-no-copy-discipline-file (if the
+  generalization is itself the discipline content), or (c) drop
+  (if the section was internals-only and the generalization adds
+  no Zeta-relevant value).
+- Land as a single PR with explicit "discipline-aligning rewrite"
+  framing.
+
+## What's allowed under the no-copy discipline (refresher)
+
+Generalized "about" framings ARE allowed:
+
+- "a database-related sibling project"
+- "a pre-DBSP event-stream-processing prototype"
+- "a package-manager seed in the ace-package-manager direction"
+
+What's NOT allowed (privacy-class internals):
+
+- Specific company names, customer names, architecture details
+- Specific subdirectory structures or named experiments
+- Verbatim or near-verbatim copies of code/design from siblings
+
+The cleanup pass keeps the generalized abouts; rewrites the
+specific-internals sections to either generalize or drop.
+
+## Why P2 not P0/P1
+
+The bleed-out is in a memory file that already exists with the
+problematic content. The damage (if any) is from existing-state,
+not ongoing-state. The new no-copy discipline (PR #916) governs
+all NEW substrate going forward; the existing file is a known
+exception flagged in its own frontmatter. P2 because:
+
+- Not blocking anything.
+- Not actively producing new bleed-out (frozen content).
+- Cleanup is mechanical once the trigger fires.
+- Aaron's framing was "add specific identifiers MUST stay inside
+  the sibling repo" — the existing file's identifiers exist
+  inside Zeta but came from before the rule.
+
+If the file becomes load-bearing for new work (e.g., the TS+Bun
+expert baseline cites it), promote to P1 at that point.
+
+## Composes with
+
+- `memory/feedback_no_copy_only_learning_from_sibling_repos_aaron_2026_04_30.md`
+  (the discipline this cleanup aligns to)
+- `memory/project_laptop_only_source_integration_*.md`
+  (the file being cleaned up)
+- task #351 (TS+Bun expert baseline — composes per Deepseek's
+  enhancement-opportunity note)
+- Deepseek 4th review (`docs/research/2026-04-30-multi-ai-feedback-packets-this-session.md` § Deepseek — fourth review)
+  — the explicit ask that triggered this row
+
+## Acceptance
+
+When the trigger fires AND the cleanup pass lands, mark this row
+`status: closed` with reference to the cleanup PR.
+
+If 6 months elapse without the trigger firing, re-evaluate:
+either fire the cleanup voluntarily during a quiet round, or
+explicitly defer further with a stale-decay note.