Retraction simplification: agent handoff

2026-10-24. User-directed coordination from the conversation in /home/wsl/agent/v3-codex; review of this checkout at 2d91c8029 on fix/profile-registration-cost, including its uncommitted work.

User's intended direction

The user considers the current retraction feature overengineered and wants:

1. Delete the content from the canonical repository and commit immediately.

2. Later, sweep the content from Git history.

The user explicitly clarified: “We should be saving a record of the retraction, but not this ledger type stuff.” Keep a durable retraction record; the proposed shape is to commit it together with the content deletion. Delayed unavailability is acceptable. Do not assume a requirement for immediate suppression of every derived copy or continuous proof that no copy can reappear.

This is a direction and coordination note, not an implemented change or approval to remove current protections, rewrite history, deploy, or start paused work. Exact record fields, migration, derived-copy cleanup and history-purge mechanics still need definition. The record should contain no deleted prose and remain useful after the original content has been purged.

What changes about ongoing work

• The user requested deletion of retraction_sweep_diagnostics. Its Step 1, Step 2 and debug log were deleted in /home/wsl/agent/v3-codex; its study entry was marked canceled with historical scores retained. Do not revive that plan or implement its structured evidence-diagnostics subsystem. The existing investigation report remains historical evidence, not an active work order.

• Registration indexing, targeted profile/activity updates, bounded approval work, and ordinary status/lock recovery improvements remain useful. Continue only within their existing authorization; this note does not start the paused registration checklist.

• Revisit REG-17, REG-20 and REG-F01/F02 before treating today's retraction checks/publication protections as permanent requirements. Preserve correct authorization and transactions; distinguish those from the stronger immediate retraction freshness policy the user is relaxing.

• Treat the retraction retrospective's corrective requirements as historical proposals requiring reassessment. In particular, do not expand provenance, evidence reconstruction, publication gating or sweep diagnostics simply to satisfy R02, R09 or R19 under the old contract.

• The recent high-traffic fixes still depend on current retraction evidence: HighTrafficPublication repeatedly captures/validates it, and HighTrafficRenderInput::safetyIdentity() explicitly permits ordinary content staleness but forbids retraction/absence staleness. These are likely simplification targets. Retain useful no-rebuild rendering and performance improvements; do not blindly revert the recovery commits.

Findings behind the direction

Current browser retraction commits a signed suppression record while retaining the original post. The current RetractionSweep cleans selected generated artifacts and LLM exchanges; it explicitly excludes canonical files and history. Neither implements the user's two steps as stated.

The separate delete-record/SourceLabelRemoval path already deletes and commits content, but also has removal receipts, recovery journals and history-dependent validation. Reuse useful behavior without carrying that entire design forward. Simply deleting post files inside the existing retraction flow is insufficient: some readers expect the original post or accepted removal evidence, and operator acceptance is reconstructed from Git history.

The recommendation discussed was simplification in place, retaining the retraction record, authorization and conversation structure. A new history purge must be deliberate; deleting the diagnostics plan does not supply that feature. Keep ordinary actionable deletion/commit/purge errors without rebuilding the canceled evidence-diagnostics project.

No application changes, commits, tests, deployment or history rewrite were performed as part of this handoff. No dependencies added. Existing unrelated changes and todo.txt are untouched.

2026-10-25T19:15:34Z — Approved release-1 implementation available

The user approved Steps 1–3. Local implementation now lives on feature/retraction-simplification in /home/wsl/agent/v3-codex, through 471b8111 (planning fd5c6849, eight stage commits). Read that checkout's docs/plans/retraction_simplification/retraction_simplification_step4_implementation_summary.md for verification and limitations. This supersedes the earlier planning-only status, not its historical findings. Nothing is merged, pushed or deployed by this handoff.

Author/operator acceptance deletes the post/signature and commits one JSON record with preserved signed evidence and content-free structure. Legacy migration is explicit (retraction-sweep migrate, then ordinary rebuild). Readers/imports use accepted records without repeated Git authority reconstruction; read-model identity is now 26. PublicationGuard is removed; ordinary serialization is retained as Support/PublicationLock. Do not reintroduce retraction generations/gating in registration or publication work. Sweep uses full repeatable scans and ordinary task-queue status, without saved cursors/ledger. History purge remains a separate, unimplemented release 2.

Operational consequence: generated artifacts also expire after five minutes while accepted retractions exist, including clean copies; ordinary refresh jobs must keep running and static-only deployments must rebuild after retirement. Unknown-provenance legacy LLM payloads are conservatively scrubbed. See operator references before integration. Live qdb.us recovery remains unverified; local QDB-derived migration, rebuild and 18,931-artifact CLI sweep passed. Final focused regression suite 122/122; full-suite inherited failures and environment assumptions are documented.

Reviewed this checkout through d468c9059; no newer Host/Write/Retraction commits since 63314e6cf. Its beta.4 upgrade and unrelated work are untouched. This note is left uncommitted for the agent here; it does not authorize merging or deploying.