The name
Work is queued and the name returned without waiting for the run. Runs go to one worker, so a second run waits for the first — but its request still gets its own name straight away, and that run already reports as running.
The answer afterwards
A run that raises is kept as failed, and the words of what went wrong are given back with the name.
A name never given out is reported unknown. No record is ever removed, so a name stays answerable for as long as the service runs.
How the figures come across
A finished run's counts come back as numbers and its amounts as text, so nothing is rounded on the way out. A figure that is missing comes back as nothing at all, rather than as a nought or as the word.
The distribution figures are the exception. They pass through as numbers, already rounded where they were worked out.
Related
Source evidence
---
row: P4-40
baseline: 49aa659
created: 2026-10-03 20:55 UTC
modified: 2026-10-04 11:43 UTC
evidence:
- claim: "Work is queued and the name returned without waiting for the run"
source: "`replay_api.py` 150-155"
- claim: "Runs go to one worker, so a second run waits for the first"
source: "`replay_api.py` 105 and 153"
- claim: "A queued run already reports as running"
source: "`replay_api.py` 150-151, the three words at 39-41"
- claim: "The name is answered once, and the run's state only when asked for again"
source: "`replay_api.py` 155 against 157-166"
- claim: "A run that raises is kept as failed, and the words of what went wrong are given back"
source: "`replay_api.py` 108-112 and 284-285"
- claim: "A name never given out is reported unknown, and no record is ever removed"
source: "`replay_api.py` 160-164 against 112 and 116"
- claim: "A finished run's counts come back as numbers and its amounts as text"
source: "`replay_api.py` 272-281 into `metrics_serialization.py` 27-50 and 57-68"
- claim: "A figure that is missing comes back as nothing rather than as text"
source: "`metrics_serialization.py` 18-20"
- claim: "The distribution figures pass through as numbers, rounded where they were made"
source: "`metrics_serialization.py` 44-45, rounded at `percentiles.py` 62-64 and `snapshot_result.py` 93-103 and 116-123"
---