- What it runs on
- Your whole case library, nightly
- What it decides
- Drift from the accepted baseline
- Cost of being wrong
- A regression found at review, not at commit
What it checksCriterion
- 01Drift from baselineper-case tolerance you set
The quantity you care about, compared with the run you signed off.
- 02Silent change of regimesame inputs, different physics
A case that still converges, into a different answer than last month.
- 03Environment movedsolver build, decomposition, libraries
The run did not change. What ran it did.
These are criteria, not results. Nothing here has been measured — the product is not built.
What comes back
- The cause, and the line of solver output it came from
- The change, as a diff you can reverse
- Physics checks, measured against their limits
- A record somebody else can open on a link
Targets
- Every product in the suite
- Your CI
- The same evidence record
Named as intent. No adapter for any of these exists today.
Standing
In development, targeted at October 2026. It is the layer over the other four rather than a physics of its own: the same watch-diagnose-verify loop, run on a schedule over a library of cases instead of once over one.