Operator outcome: use valuable authorized work evenly across five Business workspaces, preserving the protected Pro account and existing cost, concurrency and stop controls.
266cc5b8c9f7db21f438676b2706aa37bac2e93c; fixture correction PR 11804.7106fd95deea837a81756f4448761a1ced6e0a6f, candidate 7106fd95deea837a81756f4448761a1ced6e0a6f-4618-1.d1edac1309eec02f7fa79d745ce306e312bc136f; production run 4620 succeeded./healthz returned HTTP 200, JSON ok, environment production, source revision 7106fd95deea837a81756f4448761a1ced6e0a6f. Frontend build revision matched that source. Both deployments had one ready replica.sha256:18c4ab987b838cc32698dc72c62f82b24558990798687f18cfa4e8dc806ac643.sha256:27c03f592c5afa1e70600537363459c3ea768a9da525110cabc8c2f2790b29fe.sha256:78a75538b6787c410d150e316c0a553e3c2f3874a52abf77e026e5e838d6a4d3; all ordinary account workers were ready on the designated worker host. This is preserved execution capability, not proof the full workers contain the latest selected-recovery query. Full worker publication remains required after the pending final source repair.Production migration installed codex_account_selection_rank(text) and the partial latest-claim index. Four ordinary selectors rank outstanding work and fresh provider allowance, then rotate tied accounts by latest actual claim. Paid routing gates, the protected main account, provider windows and the global 900-second new-outcome cadence remain enforced.
At 21:09 UTC all five Business accounts reported eligible capacity, with weekly use between 0% and 5%. The protected Pro account was ineligible for automation. Natural engine-health work started on account 16cc at 21:07:20; EventHorizon planning was queued on 3b0f. These observations prove resumed real work, not equal consumption over a whole provider window.
The first full candidate, run4614, failed because an older test used a minimal reset-credit schema without the new rank helper. The fixture now uses the full migration and real duration/reset telemetry; both focused REQ0147 tests passed, and successor full candidate4618 passed.
Live inspection found SKAN account b556 has 0% weekly use and no running task, but its max-1 ordinary slot is excluded by a queued pilot-only task. Ordinary worker claims cannot execute that task because its validated pilot scope is separate. The pilot task and its recorded ownership remain untouched.
Current chat 01a131bf-6050-7cf0-b57b-6930599f8b0b owns the follow-up on branch codex/workspace-balance-pilot-load-20261012: exclude queued pilot work from ordinary queue load, retain all running work in shared concurrency, and use the same count for rebalance source and target. A registered-pilot database regression covers all four selectors and preserved running concurrency. Source review passed; worker validation and delivery are pending. Update this record after verified rollout and full worker publication.
Broader paused TheWell delivery recovery is owned by chat “Find AutoCoder engine blockage”; budget increases or clearing held records are not substitutes for that recovery.
No credentials, account permissions, purchased overage, reset redemption, protected headroom, task history or operator holds were changed. Rollback, if required, must use the governed immutable candidate route and verified prior artifacts; the earlier production candidate was e5aa59f862f3c9840cb7077a601aa076ce459b13-4601-1. The additive rank helper/index do not require deleting task or quota accounting. This change record does not authorize deleting older artifacts or releasing historical capacity debt.