[JacobStephens2]

Summary

Grader panel retained-call History
├── capture every expected process pair × seat × order
├── estimates → all pending calls, including unfinished Trials
├── execution → terminal pairs' missing calls → atomic publication
└── reports → both orders for every seat → existing tally

Corrupt retained calls now stop cost confirmation and resume before paid work, even behind an earlier missing order or with --yes. Saved filenames, JSON, voting, fixed executables and per-run spending policy stay compatible. Resume documentation and the domain-to-code index describe the shared owner.

Closes #1968

Evidence

Execution evidence

• Before: cargo test --test experiments corrupt_later_panel_calls_stop_resume_before_any_more_sessions -- --nocapture
  failed with left: 2, right: 1: a second Grader launched before malformed order 2 was reported.
  After: the same command passed: 1 passed; 0 failed.
• Before: cargo test --test experiments panel:: -- --test-threads=3 failed: 14 passed; 1 failed; the new sparse-history fixture incorrectly started from a completed Experiment.
  After: the same command passed: 15 passed; 0 failed (100.87s), including the retained sparse multi-seat regression.
• cargo check --bin thirdshift passed. cargo test --bin thirdshift evals::experiment::panel:: -- --nocapture passed: 7 passed; 0 failed.
• npm run docs:check --prefix site and cargo fmt --check passed.
• Earlier implementation validation: scripts/affected-tests refs/remotes/origin/main passed: site checks passed in 100.28s; the full Rust suite (cargo test fallback) passed in 1933.62s, including compilation (1m 16s).
• GitHub CI passed for the current head 81eed964995fb65640f5a874963522851a72f99d, including Rust, site and Console checks. The successful full suite is reused under the repository's exact-head CI policy; no product changes were needed in this continuation.
• This continuation: cargo test --bin thirdshift evals::experiment::panel:: -- --nocapture passed (7 passed; 0 failed); cargo test --test experiments panel:: -- --test-threads=3 passed (15 passed; 0 failed, 93.82s execution); npm run docs:check --prefix site passed after npm ci --prefix site.

Test seams: the retained-call History interface with Native/Observed reads and temporary storage; the Experiment CLI's cost confirmation, pause/resume and reports. Existing voting, family, counted-people, low-limit and unpriceable-session regressions are retained.

Visual evidence

Manual resume documentation, same expanded Experiments section and scroll position. Before was captured from main at 6ce3c940; After shows this change.

1440×900 desktop:

390×844 mobile:

Review coverage

Reviewed the whole branch from refs/remotes/origin/main (84c5856e) on independent Standards and Spec axes: Standards 0 findings; Spec 0 findings. Both axes read all 11 changed files, including generated HTML. Changed files left unread: none. Generated-output exclusions: none. The existing sparse-history fixture correction and its regression remain in the branch.

Unaddressed findings

Standards

None.

Spec

None.

Security

None.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

No saved-data migration; the refactor can be reverted.

Blast Radius: Experiments

Panel estimates, resume and reports share completion facts. Malformed or unreadable saved calls now stop earlier with pair, seat and order context.

Built with codex codex-cli 0.167.0 · gpt-6.1-sol · high