> ## Documentation Index
> Fetch the complete documentation index at: https://docs.obversa.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Read a Record

> Prints a run's record as a page a person scans, or as JSON for a page of your own.

`obversa-record` prints a run's record as Markdown you scan in a minute:
each step's runs, what they cost, and where a step ran again. The record
itself stays JSON lines for machines. Use the command after a run, or on
one that was killed, to see what happened without opening the file. To
watch a run while it works, use [the run's page](/driving/monitor).

Point it at the file a run wrote with `recordTo`:

```bash Terminal theme={null}
npx obversa-record records/public-docs-writer.jsonl
```

`--json` prints the same structure as JSON for a page of your own. Both
come from `@obversa/runtime`: `summarizeRecord(events)` returns the
structure and `renderRecord(events)` the Markdown, so a host can print a
record without the command.

## Layout

* **The header** carries the run's name, when it started and ended, its
  outcome, and the tokens in and out, with how many were read from cache. A
  run that was killed says so: no end recorded.
* **A section per step**, in the order the steps first ran. A step that ran
  again after a review has a numbered run for each time, and the findings
  that made it run again sit above that run: which step returned them, why,
  and the count against the step's limit.
* **Each run** shows its outcome and how long it took, the first line of
  what the step said, each model call with its tokens, and the tool calls
  folded to one line: the tool, how many times, and the files or commands
  it touched when the engine reports them.
* **A pause** is its own line, with the question a person is being asked,
  or the reason a limit stopped the run.
* **Anything else** in the record prints as a line with its kind, so
  nothing is hidden.

The output below is the header and the first step of a run of the writer,
reader and judge team over a docs page. The run was killed during the
writer's third round:

```text Output theme={null}
# public-docs-writer

- Started 2026-09-27 16:09:13 UTC. No end recorded.
- Tokens: 1,911,281 in, 17,845 out, 1,761,466 read from cache.
- Monitor: http://127.0.0.1:54312/

## write

### Run 1: pass, 3m 0s, started 16:09:13

Done. The page now reads as you would explain it to a colleague. Key changes:

- claude-sonnet-4-5-20250929: 802,624 in, 7,903 out
- Read ×3, Edit ×3, bash ×1

Kickback from route, 1 of 6: round 1: the judge says another round is worth it (0.63); 6 block, 4 should, 0 nit

### Run 2: pass, 1m 22s, started 16:13:25

Fixed all 10 findings. The page now uses words a person would say: reject instead of fail, approve i…

- claude-sonnet-4-5-20250929: 791,430 in, 2,876 out
- Read ×1, Edit ×6

Kickback from route, 2 of 6: round 2: the judge says another round is worth it (0.63); 3 block, 7 should, 0 nit

### Run 3: no end recorded, started 16:16:10

- Read ×1, Edit ×4
```

The header says no end was recorded, and the third run of `write` has a
start and no end: that is where the run was killed. The two lines that
begin `Kickback` are the route returning the draft to the writer, with the
judge's reason and the count against the writer's limit of six.

## Next steps

* [The record](/concepts/record): the event log this page is read from, and
  what a resumed run repeats.
* [Watch a run in the browser](/driving/monitor): the same lines, live,
  while the run works.
* [Runtime](/packages/runtime): `renderRecord` and `summarizeRecord` for a
  page of your own.