--- title: reeval history, show, compare, checkout description: Read recorded runs, diff two of them and reproduce one. --- > ## Documentation Index > Fetch the complete documentation index at: https://docs.reeval.ai/llms.txt > Use this file to discover all available pages before exploring further. Four commands read the history: recorded runs, plus unrecorded scratch runs where noted. ```bash npx reeval history # pass/fail per eval across runs npx reeval show latest --md # one run as Markdown npx reeval compare # latest vs its nearest recorded ancestor npx reeval checkout latest # a worktree at that run's commit ``` ## reeval history [//]: # (generated:cli-help:history) ```text reeval history [filter] Pass/fail per recorded eval over recent runs --limit --all (include unrecorded runs) --json ``` [//]: # (/generated:cli-help:history) The filter is a case-insensitive substring of the eval id. Runs go oldest to newest, left to right. ```text 2 runs, oldest → newest: 20260926T173922Z_7e5376d_ada-lovelace_31fh (7e5376d) … 20260926T173923Z_7e5376d_ada-lovelace_t2ea (7e5376d) ✓✓ 2/2 evals/hello.eval.ts › greet module › creates a tested greet module ✓✓ 2/2 evals/hello.eval.ts › greet module › adds a default name in a follow-up turn … ✓ pass ✗ fail ! infra error ↓ skipped ○ not gated · not run ¦ eval or stack changed ``` `¦` marks a run where the eval's fingerprint changed from the last run that had the same cell, so a filtered run and the full run after it don't count as a change. `○` marks a study's run (every cell `gate: 'none'`), which the pass rate leaves out; `--json` gives its outcome as `not_gated`. ## reeval show [//]: # (generated:cli-help:show) ```text reeval show [run] One run: result blocks, --md (summary.md) or --json (the record) ``` [//]: # (/generated:cli-help:show) The default output is a state table (commit, dirty patch, lockfiles, images, harnesses, environments, judge, agent context, packages, runner), then the result blocks and tally `reeval run` ended with. `--md` prints the run's `summary.md`; `--json` prints the record. A run with an `insights.json` (from `reeval run` or [`reeval analyze --insights`](/reference/cli/analyze#insights)) prints its sentences in the blocks, as text and `--md`. A run that raised [remarks](/reference/cli/run#remarks), recorded or not, ends with the concerns it printed. ## reeval compare [//]: # (generated:cli-help:compare) ```text reeval compare [base] [head] Diff two runs; base defaults to head's nearest recorded ancestor --md --json --fail-on-regression -v, --verbose Also each cell's lift and effort, base → head, and the stack changes ``` [//]: # (/generated:cli-help:compare) - **Defaults:** head is `latest`. Base is the nearest recorded, complete run on head's git ancestry, falling back to the previous run by time. - **Blocks:** one per eval that changed, in file order: `regressed` (both run ids and each check that flipped, `pass → fail` with its message, or `not run`), `fixed` (the checks that pass now), `can't compare` (what changed, and each side's id and result), `infra error`, `new` and `removed`. Runs of unchanged evals collapse to `+N same`. The last line is the tally, regressed first, zeros left out. - **Can't compare** means a fingerprint part or the grader changed, and reeval names it: `changed evals/helpers.ts`. - **Not run:** a check head never reached reads `not run`, never `fail`, and a check neither side reached is never a flip. - **Unfinished runs:** when base or head was interrupted or stopped at its duration budget, a note above the tally says so, since some of its evals didn't run to the end. - **Cancelled evals:** an eval that failed only because Ctrl-C cancelled its attempts reads `can't compare`, with that side's result `cancelled`, never `regressed` or `fixed`. In `--json` its row is `drift` with `cancelled: ["head"]` (or `"base"`). - **Filtered runs:** when base and head ran different cells of an eval, the eval is classified on the cells both ran, and the rest are listed as `cells added` or `cells removed`. With no cell in common, it can't compare them. - `-v` adds each cell's lift and effort, base → head, and every stack change. The default output has no statistics. - `--md` prints a `### reeval compare: head vs base` heading and the same blocks in a text block, for a PR comment. `--json` returns `{ base, head, changes, rows }`. - **No earlier recorded run:** `No earlier recorded run to compare with, so every eval is new.`, then head's result blocks and `N new evals`; `--json` has `base: null`. That's not an error. - `--fail-on-regression` exits 1 when any eval regressed, or when a `new` or `can't compare` eval fails, since no comparable earlier run vouches for it. A changed eval that passes, or one whose head was cancelled, doesn't count. Without the flag, `reeval compare` exits 0. ```text regressed greets in the house style base 20261020T231407Z_ea2a649_ada-lovelace_8h1n head 20261020T231408Z_b2c98b4_ada-lovelace_fypq house greeting pass → fail expected command to succeed, exit code 1 sign-off not run +1 more passed +2 same … 1 regressed · 1 fixed · 1 can't compare · 1 new · 1 removed · 4 same ``` ## reeval checkout [//]: # (generated:cli-help:checkout) ```text reeval checkout [run] [dir] A git worktree at the run's commit, with its worktree.patch applied (default dir: .reeval/checkouts/) ``` [//]: # (/generated:cli-help:checkout) It runs `git worktree add --detach` at the run's commit, then applies the run's `worktree.patch` if it was dirty. Remove the worktree with `git worktree remove `. The default directory is inside `.reeval/`. Deleting `.reeval/` leaves a stale worktree entry until you run `git worktree prune`. ## Related - [History](/guide/history) - [Fingerprints and drift](/concepts/fingerprints) - [The run record](/reference/record)