GH-content sync: workflow-fragility — local branch-discipline gap producing stale-branch + chore-sync contamination (5x same-session pattern)

Context

5 PRs this session (2026-05-10) hit the same stale-branch + chore-sync contamination pattern:

5 occurrences in <90 minutes is empirical evidence for systemic workflow gap, not 5 isolated incidents. Per feedback_substrate_scope_restraint: when same friction trips 2-3 cycles, treat as substrate-quality concern (not iteration target). 5 cycles is well past that threshold.

The Problem

Two distinct sub-patterns:

1. Stale-branch from peer's pre-merge branch. Branching feature work from another peer's still-open PR branch (rather than origin/dev) bundles their content into the new PR's diff. Substrate symptom: gh pr view --json files shows files from the other peer's PR alongside the actual feature work; review surface bloated; squash-merge collision risk.

2. Local sync-pipeline auto-bundles chore: ticket sync [skip ci] commits onto feature branches before push. Substrate symptom: branch contains both feature work AND auto-sync data; PR diff shows hundreds-of-files / thousands-of-LOC stat where actual feature is ~10-100 LOC.

Both produce the same outcome: review surface signal-to-noise asymmetry + squash-merge git-history pollution (per #11116 friction-gold lesson).

The Architectural Reality

feedback_branch_from_origin_dev_explicitly MEMORY.md entry codifies the discipline:

"branching from a peer's local branch (or current HEAD) in active multi-PR cycle conflates scope; PR diff includes the upstream-branch's commits"

But the discipline is discipline-only, not enforced. 5 occurrences this session demonstrates discipline alone is insufficient operationalization.

The chore: ticket sync [skip ci] auto-bundle source is presumed to be a local sync-pipeline (e.g., npm run ai:sync-content or equivalent) that writes to git-tracked content directories. When developer runs the sync mid-feature-work, the auto-commit gets staged onto the feature branch.

The Fix — Three Path Options

Path A (lowest cost): Documented branch-from-origin/dev mandate. Add explicit pre-flight to pull-request-workflow.md:

Before opening a PR, verify branch base:
  git log origin/dev..HEAD --oneline
If output contains commits NOT belonging to your feature, rebase first:
  git checkout -b agent/<ticket>-v2 origin/dev && git cherry-pick <feature-commits>

Path B (mechanical enforcement): pre-push git hook. Detect non-feature commits on the branch (e.g., chore: ticket sync patterns OR commits authored by other peers) and warn/block before push:

.git/hooks/pre-push:
  - Check `git log origin/dev..HEAD --oneline` for chore-sync OR other-peer-authored commits
  - Block if found; suggest cherry-pick rebuild

Path C (substrate-level): isolate sync-pipeline-output from feature branches. The chore: ticket sync auto-bundle is the recurring friction source. If sync-pipeline writes to a separate worktree OR commits to a dedicated branch (not whatever HEAD is checked out), feature branches stay clean.

My lean: Path A + Path B together. Path A makes the discipline visible; Path B catches misses mechanically. Path C is heavier substrate work; defer if A+B converges the pattern.

Acceptance Criteria

• Path A patch: pull-request-workflow.md adds explicit branch-from-origin/dev pre-flight check with rebase remediation example
• Path B implementation: .git/hooks/pre-push (or equivalent project-level hook config) detects + warns/blocks chore-sync OR peer-authored commits on feature branches
• Documentation: feedback_branch_from_origin_dev_explicitly MEMORY.md entry updated to reference this ticket as enforcement layer
• Empirical validation: next feature PR opened in any session — verify (a) author-side discipline check runs; (b) hook catches misses; (c) PR diff surface matches actual feature scope (no stale-branch / no chore-sync contamination)

Out of Scope

• Modifying the local sync pipeline itself (that's Path C; defer until A+B convergence-tested)
• Cleanup of currently-in-flight PRs (#11129 RA1 + #11132 inherited stale-branch) — those are per-PR mechanical fixes; this ticket addresses the SOURCE
• AGENTS.md §0 invariant additions — this is value-tier discipline, not invariant-class

Avoided Traps / Gold Standards Rejected

Decision Matrix

1. Path A + Path B (discipline + mechanical enforcement) — Selected: Lowest implementation cost; addresses both sub-patterns (stale-branch + chore-sync). 5x empirical pattern this session demonstrates discipline alone is insufficient.

2. Path A only (documented mandate): Rejected. feedback_branch_from_origin_dev_explicitly already codifies the discipline; 5 occurrences this session post-codification demonstrates documentation alone fails.

3. Path C only (sync-pipeline isolation): Rejected as substrate-evolution scope-creep for this ticket. May become substrate-correct end-state once A+B is empirically validated; defer.

4. Per-PR cleanup discipline (continue current pattern): REJECTED. 5 occurrences in 90 minutes is the friction; continuing per-PR cleanup is treating symptom not source. Per feedback_substrate_scope_restraint: at 2-3 cycles of same iteration without convergence, choose Drop+Supersede / substrate-correct shape. We're at 5 cycles.

5. Operator-policing the discipline: Rejected. Per feedback_lead_role_decision_thresholds + just-codified Verdict-Authority discipline (#11102 / PR #11127): substrate gates should self-correct via codified peer-V-B-A; relying on operator surfacing each instance is the friction the MX loop converts to gold.

• Trap: Treating each PR's stale-branch as one-off + cleaning per-PR. Rejection: 5 same-session occurrences = pattern, not incidents.
• Trap: Codifying discipline without mechanical enforcement. Rejection: Empirically tested this session — feedback_branch_from_origin_dev_explicitly was already-codified yet 5 PRs tripped it. Documentation alone is insufficient operationalization.

Related

• feedback_branch_from_origin_dev_explicitly MEMORY.md entry (the existing discipline this ticket enforces)
• #11116 (sister friction-gold: code-vs-data-migration commit-shape; same family of "review surface signal-to-noise asymmetry" friction)
• PR #11106 cycle 1 / PR #11109 / PR #11114 cycle 1 / PR #11129 / PR #11132 — empirical-anchor cluster (5 PRs, 1 session)
• pull-request-workflow.md (Path A target)
• .git/hooks/pre-push (Path B target)
• AGENTS.md §13.2 friction → gold core value (this ticket IS the conversion artifact)

Origin Session ID: c2912891-b459-4a03-b2af-154d5e264df1 Retrieval Hint: "GH-content sync workflow-fragility", "stale-branch + chore-sync 5x pattern", "branch-from-origin/dev enforcement layer"