Change Validation¶
Use the smallest honest validation set that proves the changed scientific intent.
Validation Flow¶
flowchart LR
change["changed nav surface"]
local["local family tests"]
reference["reference-backed proof"]
downstream["dependent estimator or product proof"]
result["documented validation result"]
change --> local
local --> reference
reference --> downstream
downstream --> result
Minimum Validation By Change Type¶
| change type | minimum validation |
|---|---|
| parser change | owning format-family tests plus one reference-backed product test when applicable |
| orbit or time change | relevant reference tests plus one downstream estimator check when the state feeds solvers |
| correction change | correction-family tests plus one dependent position, PPP, or RTK test |
| estimator change | local solver-family tests plus narrow integration tests that prove public behavior |
| refusal or integrity change | refusal-path tests and evidence inspection for the rejected condition |
Bad Validation Patterns¶
- only running a broad integration test for a low-level change
- only running unit-like tests for a public solver behavior change
- accepting green tests without checking whether the affected scientific family was actually exercised
Review Checks¶
- Which scientific claim changed?
- Which proof exercises that claim directly?
- What refusal or uncertainty evidence protects readers from over-trusting a solution?