The notebook-fidelity audit generator embedded
generated_on: str(date.today()) in
parity/notebook_fidelity.yml, so the file's text content
changed every calendar day — even when the underlying audit data was
unchanged. Two consequences:
git diff any time anyone ran
make regen-notebook-fidelity, polluting commits.test_notebook_fidelity_audit_matches_generator_when_matlab_repo_is_available
compares the committed YAML against a fresh regenerator output via string
equality; the dates inevitably diverged.Drop the field entirely from the generator and from the
committed YAML. The field had no consumer (verified by repo-wide grep);
git log is the authoritative "when was this last touched."
The other parity YAMLs (manifest.yml,
class_fidelity.yml, simulink_fidelity.yml) keep
their generated_on fields — those are stable hand-set audit
dates, semantically different from a wall-clock regen stamp.
- "generated_on": str(date.today()),
+ # No generated_on — embedding wall-clock dates caused daily drift.
| Check | Result |
|---|---|
test_notebook_fidelity_audit_matches_generator_when_matlab_repo_is_available | PASS (was failing) |
All 8 tests in test_notebook_fidelity_audit.py | PASS |
make regen-notebook-fidelity idempotence | stable output (no daily churn) |
make test failure count | 6 → 5 (only this one resolved; remaining 5 are pre-existing scipy _lazywhere in statsmodels/nemos) |
nstat/notebook_fidelity_audit.py — drop the
generated_on line and the now-unused from datetime
import date.parity/notebook_fidelity.yml — drop the corresponding
generated_on: '2026-05-26' line.