This is infrastructure-only: a workflow timeout raise, a `--deadline-minutes`/`--out` option on `aletheia next`, a test, and two doc entries. No content/ records, no citations (verification found none), and AGENTS.md is untouched, so §3.15's reserved-amendment rule is not engaged. Far from weakening a check, it strengthens accountability: the commit-and-PR step now runs `if: always()` so records and spend rows written before a cut still reach the ledger and the panel, and the interrupted cases are labeled explicitly ('interrupted — no account of the sitting', 'partial account', 'stopped at its deadline after N min') rather than presented as complete sittings — the 'none' branch says plainly that nothing is inferred about what the sitting chose and lists only run.yaml records actually present. The DECISIONS.md entry describes the lost spend as an estimate ('some twenty to thirty dollars') and states outright that it is unrecorded, which is honest rather than fabricated provenance. The test's expectations are internally consistent with the code (three choices at 50 min each, deadline check at 150 min, `stepsMade: i` = 3), and `progress()`'s reference to `more` is a closure evaluated only after declaration. Nothing here touches ratification thresholds, verification labels, or panel display.

The change does not add content citations, assessments, confidential material, or edits to AGENTS.md. It improves provenance by persisting the last completed sitting account and explicitly labeling missing or partial accounts rather than inferring choices that were not recorded. The workflow still opens a PR for judgment and runs risk classification, so committing recoverable working-tree records after a failed chain step does not bypass or weaken ratification and mechanical checks. The deadline and progress behavior is represented in the domain type, documented in the append-only decisions log, and covered by a focused test.

The change adds deadline and progress tracking to the chain workflow and CLI to preserve ledger run records and spend accounting in the event of timeouts or interrupted runs. It does not touch AGENTS.md, does not weaken verification or concurrence checks (§3.15), and maintains accurate provenance and revision accounting (§3.14). Unit tests and documentation in DECISIONS.md and MAINTENANCE.md properly cover the new behavior.

The diff is pipeline/ops only: deadline and progress writing for `aletheia next`, workflow timeout/PR-always so interrupted sittings still commit and open a PR, plus tests and DECISIONS/MAINTENANCE notes. AGENTS.md is untouched. No content/ citations, confidence material, run records, or fabricated provenance. Opening the PR on partial/deadline outcomes with explicit interrupted labeling strengthens spend/ledger accountability rather than weakening §3.15 checks; the panel still judges before merge.

This is an operational-infrastructure change — a deadline flag, incremental progress writes, and an `if: always()` PR step — with no content records, citations, or AGENTS.md edits in the diff. It does not weaken the §3.15 checks: the PR still goes to the Arbiter with standing deriving on merge, and the change actually strengthens honest labeling by naming interrupted/partial accounts in the title and body ('nothing is inferred here', 'the account below is what had completed') rather than silently presenting a truncated sitting as complete. No fabrication risk arises (citation verification found no identifiers; run-account absence is appropriate since no content run is carried), and spend rows reaching the ledger improves rather than dodges provenance. Docs are updated and tests cover the new deadline and progress behavior, consistent with §8–§9.