Clean the table when its recovery generation is created, retaining lazy cleanup for record rooms. Add guard-loss regression and diagnostic measurements at 1,000 and 10,000 leases; expose browser profiles and await dashboard redirect before navigation.
Co-Authored-By: GPT-6 (Codex) <noreply@openai.com>
Exercise unavailable rendering and automatic lease rejoin through actual Chromium and two backend processes after disposable Redis loss. Retain traces and screenshots; frontend serving remains fixture-only and opt-in.
Co-Authored-By: GPT-6 (Codex) <noreply@openai.com>
Report unavailable presence during recovery, invalidate stale leases, and isolate controlled restores by host generation. Add store, HTTP and independent-JVM recovery acceptance. Browser, maintenance enforcement and performance qualification remain open in QQQ #1300.
Co-Authored-By: GPT-6 (Codex) <noreply@openai.com>
Require identified admin evidence and locked reference checks before deleting bytes. Preserve immutable operational receipts, exclude them from retries, and expose the standard Review/Result table action with help.
Co-Authored-By: GPT-6 (Codex) <noreply@openai.com>
Do not retire a failed-upload locator merely because DELETE succeeds. Let adapters attest authoritative settlement; otherwise retain it with bounded retry scheduling. Reproduce a PUT completing after rollback and an independent sweep.
Refs QRun-IO/qqq#1289
Co-Authored-By: GPT-6 (Codex) <noreply@openai.com>
Use batched providers without replacing host customizers. Preserve transaction visibility and parent authorization, retain tag picker behavior, and verify query, GET and export values. Requires coordinated core PR QRun-IO/qqq#1322; published dependency pin follows after its snapshot is available.
Refs QRun-IO/qqq#1196
Co-Authored-By: GPT-6 (Codex) <noreply@openai.com>
Add a reproducible opt-in shared/distinct-parent workload with SQL and latency measurements, exact visibility assertions and bounded pool checks. Correct the fixture's unpooled connection churn to match the host provider; no runtime capacity claim.
Co-Authored-By: Codex <noreply@openai.com>
Measure actual SQL calls with pg_stat_statements and preserve visible count, top-ten order and permission revocation. Focused real PostgreSQL fixture passes.
Co-Authored-By: Codex <noreply@openai.com>
Preserve exact visible counts and due-date top ten without retaining all matching rows. Recheck authorization per page and request. Full verification passes 193 tests and coverage gates.
Co-Authored-By: Codex <noreply@openai.com>
Retain cleanup locators after rollback compensation fails, keep failed batches from starving later work, and preserve pending requests across schema upgrades. Real PostgreSQL regressions and full verification pass; crash and provider acceptance remains tracked in QQQ #1289.
Co-Authored-By: Codex <noreply@openai.com>
Commit cleanup intent with attachment metadata removal before deleting storage bytes. Retry committed cleanup through the existing scheduler and verify PostgreSQL rollback, failed metadata deletion, lost callbacks and replay. Remaining upload and provider recovery acceptance stays tracked in QQQ #1289.
Co-Authored-By: Codex <noreply@openai.com>
James Maes
james.maesandCodex