[j3w1]

Pod 0.21.12: the records milestone of #109. Bookkeeping limits no longer strand a live objective: receipt histories hold 64, any row can be renewed identically, and a full map can still record its next required finding.

Closure without the owner (#109 R4, PoD#4) is the next milestone (0.21.13) and is not in this PR. This PR does not close #109.

What changes

• R150, receipts (PoD#1). Each row retains up to 64 receipts; one write still selects at most 16. The 65th refuses atomically with the exact renewal request. A review admission reserves one receipt and 32 finding records for every pending serving attempt, so a consumable review result always fits. Ordinary checkpoint receipts cannot take reserved slots.
• R151, identical renewal (PoD#2). A checkpoint renew entry replaces any row of any kind or provenance with an identical copy: a fresh id, a renews lineage and no owner decision. The old row becomes renewed, a terminal state, and keeps every receipt, finding, acceptance and decision. Waits, effective parents, correction order and the PoD mapping follow the new row. Renewal never copies proof, never ticks or unticks a PoD, and refuses any changed definition field. Renewal waits until every serving result is consumed or dispositioned.
• R152, rows (PoD#3). A map holds 128 live rows and 512 in total. Withdrawn and renewed rows leave the live count, may be omitted from restatements, and move to immutable history positions, as do triaged findings. A report's update is normalized before triage, so a large map records a finding group from a small request. The physical 512 KiB context bound is unchanged.

Accepted limitation

A review whose report Pod keeps refusing for route mismatch still holds its reserved slots and the renewal hold. The existing route-decision rule already holds that objective, so this adds no new stranding. That route-decision hold is pre-existing and will be tracked separately.

Evidence (separate facts)

• Implementation: done by supervised workers plus coordinator corrections. Worker success is not acceptance.
• Local focused checks: 27 affected modules (470 tests) pass on Python 3.13 and 3.14 at the head commit. Skill validation, inventory, source audit, suite-lane check, compile and whitespace also pass.
• Full local gate: tools/gates.py on the exact head passed on Python 3.13 (411 s) and 3.14 (423 s). Each run had 3107 tests with 0 failures, and all 11 gates passed.
• Independent review: six rounds by fresh reviewers under the installed 0.21.11 mechanism, with required corrections fixed in between. The final audit of the head returned ACCEPT with two advisories.
• #164 shape: a sanitized fixture with 96 rows, 53 corrections and a context near the physical bound. It consumes the same settled three-finding review into one correction group. After that, the next review asks for one mechanical renewal, and further groups record on the renewed row.
• #208 shape (PoD#1, 16 → 17 receipts): proved by a disposable probe on the frozen head under both interpreters. No tracked test covers this exact shape yet; one follows in 0.21.13.
• Hosted CI, merge, installation and live recovery (R6): not run yet.

🤖 Generated with Claude Code