Break is available: it finds where your agent fails before your users do. Read the launch
Prove is available: it tells you whether a change to your agent changed how often it fails. Read the launch
Poisson Labs

Changelog · Prove · 2026-11-04

Prove 0.2.0: diff reads in plain words, and --full prints what it printed before

0.2.0

diff’s default output changed: it reads in plain words, and --full prints what diff printed before. The default is the verdict block, then one table of the conditions in the order of their axis values, each with its change and the smallest rise and fall it could have detected, and what the comparison left out, in words. The machine sections (the per-event tables by hash, the coverage table, the omnibus, the fault lines and the bound) moved behind --full. A tool that read them from the text passes --full or reads --json, which did not change.

Saved comparison ids change at this minor. An id hashes the tool’s major and minor version, so the same comparison saved again under 0.2.0 gets a new id. Comparisons saved by 0.1.x stay readable.

inspect prints identical errors as one traceback with a count. The count is of the episodes that have the error, and each listed episode is on a line of its own under it; an error none of the listed episodes has is counted in one line. --json is unchanged.

The dashboard’s comparison page shows its heatmaps on the first screen. They sit between the verdict’s detail lines and what the verdict rests on; the differences and limits charts stay after the conditions.

The limits line counts the rows with no limit by direction. In diff, check, report, the dashboard and the terminal app, a row with no limit on its fall is no longer said to have none on its rise; “no change of any size” is said only of a row with no limit either way.

The look file gains the banner section the Poisson Labs tools share. Its static UI’s type rule also lets labels (table headers, detail labels, captions) take the system’s sans.

MATH ASINFRASTRUCTURE.

Talk to us