[JacobStephens2]

Summary

docs/experiment-results.md         format 1 fields, meanings and version policy
src/evals/experiment/{store,reading}.rs
                                  identify the JSON view with format: 1
tests/experiments/result_format.rs compare real CLI/results with the document
docs/guide.md + site/              link and explain the contract

Closes #1483. Documents Trial results, Scores and rechecks, and suite Verdicts
with their Bars. No new export.

Evidence

• Before: cargo test --test experiments result_format -- --nocapture
  failed with Experiment.format: missing documented field.
  After: the same test passed with format: 1; it checks real fake-Harness
  output, retained Trial files, rechecks and Session contents. Mutations prove
  nested added, removed and retyped Score fields are rejected.

• cargo check --tests passed.

• cargo test --test experiments -- --test-threads=3 passed: 85 tests,
  one ignored (155.17s).

• RUST_TEST_THREADS=3 CARGO_BUILD_JOBS=3 /usr/bin/time -p scripts/affected-tests main
  passed all 51 site checks and 1,541 Rust unit tests, then stopped on two
  ExecutableFileBusy (Text file busy) errors in the unchanged fixture's
  binary warmup, before Experiment execution. Selector wall time: 769.59s,
  including compilation (89s). Both exact serial retries passed:

  cargo test --test experiments a_plan_limit_pause_spends_its_retry_budget_even_if_the_repository_changed -- --exact --test-threads=1
  cargo test --test experiments distinct_suite_task_pairs_keep_distinct_trial_results -- --exact --test-threads=1

• CI
  passed for the exact head a97f77fb: Linux's full cargo nextest run and
  cargo test, formatting/clippy, macOS, site and Console checks. The remaining
  local continuation was stopped after these equivalent full-suite checks
  passed; no second local full-suite run was needed.

Test seams

• Experiment CLI JSON: experiment, experiments list/show, recheck and
  retained Session reads, using fake Harnesses.
• Retained per-Trial result JSON files under the Experiment directory.

Review coverage

Independent Standards and Spec reviews against main: 0 findings on each axis.
Both reviewers read all eight changed files, including generated
site/docs/index.html. Unread changed files: none. Generated-output exclusions:
none. Reports and final files-read lists are retained in
.thirdshift-review-YWv285/{standards.md,spec.md}, with counts in summary.json.

Unaddressed findings

• Standards: none.
• Spec: none.
• Security: none.

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

Documentation, a test guard and an additive JSON format marker; no retained
result migration.

Blast Radius: readers

Experiment JSON consumers receive the new top-level format field. Existing
fields and scoring behavior stay the same.

Built with codex codex-cli 0.164.0 · gpt-6.1-sol · xhigh