[JacobStephens2]

Summary

Closes #1283. One retained Experiment reading module captures the facts used by CLI text/JSON, Console list/detail and Session-ID selection. Commands, retained formats and displayed accounting stay the same.

- View -> independent results, progress, Session evidence and usage walks
+ View -> one capture -> progress, terminal Scores, evidence, minutes, Voids, usage

Trace writing and export discovery keep their separate lifetime. The domain-to-code index now points to the reading module; the hand-written site descriptions remain accurate.

Evidence

• Before: view assembly separately walked Trials for progress, evidence and usage, reread pending records per legacy attempt and checked terminal files again for running Trials. The new capture regression initially failed because the captured reading interface did not exist.
  After: six reading-interface tests verify counted reads, controlled pending/active/result publication, absence reuse, pending/terminal distinctions, legacy attempt indices, converted-log identity, span-only usage and malformed/error handling.
• cargo check --bin thirdshift: passed.
• cargo test --bin thirdshift evals::experiment::reading::tests: 6 passed.
• cargo test --test experiments: 73 passed, 1 fixture entry point ignored; includes usage, retries/Resume, paused minutes, rechecks and Trace export.
• cargo test --test console experiments::: 3 passed, including CLI/Console pairing and Session-ID evidence selection.
• cargo clippy --bin thirdshift -- -D warnings, cargo fmt --check, git diff --check: passed.
• /usr/bin/time -p scripts/affected-tests main selected cargo nextest run. The local run stopped after 900/2,815 tests: 899 passed, one failed, 60 skipped, and 1,915 not run after fail-fast. run_health::silent_watches_only_session_trace_records expected silent but observed ok at tests/run_health.rs:189. Wall time was 801.24 s, including 1m 05s compilation.
• The exact rerun, cargo test --test run_health silent_watches_only_session_trace_records -- --exact, passed (1 passed, 0 failed, 0.86 s). No source change was needed for that result.
• The full Linux CI job passed cargo test, cargo clippy --all-targets -- -D warnings, formatting and Python tooling on exact head 035c09903f89808653879d67be1fad210da67871. This supplies the complete selected Rust-suite validation; successful CI is reused as docs/testing.md requires. Site, Console browser and macOS-gate checks passed too.

Test seams

• Retained Experiment reading: complete view through the private filesystem adapter, including counted capture and publication before projection.
• Existing Experiment CLI/Console and Session-ID interfaces, Arm usage and Trace export regressions.

Review

Independent Standards and Spec reviews used git diff main...HEAD, including the follow-up at 035c0990.

• Standards: one optional Duplicated Code heuristic addressed by sharing Stored::apply_rechecks; no hard violations or outstanding findings.
• Spec: no findings.
• Reports with final files-read lists: .thirdshift-review-tlqX9O/standards.md and .thirdshift-review-tlqX9O/spec.md (retained with this Session).
• Unread changed files: none under either axis. Generated-output exclusions: none.

Unaddressed findings

• Standards: none.
• Spec: none.
• Security: none. Guidance-mode review of git diff ee7fe1134d8a98471f0a6498da6ea35b7fbefe80...HEAD (new src/evals/experiment/reading.rs, narrowed evidence.rs/usage.rs, store.rs delegating to one capture). No SECURITY.md or threat model in repo; judged under the skill's resource-and-file-handling and resource-exhaustion classes. Evidence-session canonical-confinement checks are preserved and covered by the escaped-symlink case in src/evals/experiment/reading/tests.rs (6 reading tests pass). No introduced finding demonstrated with a proof-of-concept, so no fix and no unaddressed introduced finding.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

Local retained-reader refactor with no stored-format migration.

Blast Radius: Experiments

CLI and Console retained Experiment projections share this reader. Trial execution, mutation-side reads, locks, grading rules and Trace ancestry/export are unchanged.

Built with codex · gpt-6.1-sol · xhigh