Tried a tool Dusk open sourced today and one result caught me off guard.
Pituitary, built to catch when docs, specs and code stop agreeing. Point it at a repo, it indexes specs and decision records, flags diffs that contradict something already accepted. Ran it against a test diff that clearly broke an existing spec.
Didn't fail the check. Figured that was a bug.
Wasn't one. The tool looks for a rationale comment near the change, WHY, HACK, that kind of marker. If someone already annotated the deviation as deliberate, it routes differently than plain accidental drift.
That's the actual split. A contradiction and a violation aren't the same thing here. Spec says one thing, code says another, still gets logged rather than failed if a human already explained the gap.
Spec written and indexed, code diverges, diff runs through check-doc-drift, contradiction detected, tool searches nearby lines for that marker, deliberate deviations go one path, unexplained ones fail the build.
Nobody checks if that marker still means anything though. Nothing stops someone writing WHY just to quiet the flag, nothing checks if the original reason behind an old one still holds.
What happens to that convention across hundreds of PRs a week, once it's standing between a regression and a quiet pass? 👍
#dusk $DUSK @Dusk
Pituitary, built to catch when docs, specs and code stop agreeing. Point it at a repo, it indexes specs and decision records, flags diffs that contradict something already accepted. Ran it against a test diff that clearly broke an existing spec.
Didn't fail the check. Figured that was a bug.
Wasn't one. The tool looks for a rationale comment near the change, WHY, HACK, that kind of marker. If someone already annotated the deviation as deliberate, it routes differently than plain accidental drift.
That's the actual split. A contradiction and a violation aren't the same thing here. Spec says one thing, code says another, still gets logged rather than failed if a human already explained the gap.
Spec written and indexed, code diverges, diff runs through check-doc-drift, contradiction detected, tool searches nearby lines for that marker, deliberate deviations go one path, unexplained ones fail the build.
Nobody checks if that marker still means anything though. Nothing stops someone writing WHY just to quiet the flag, nothing checks if the original reason behind an old one still holds.
What happens to that convention across hundreds of PRs a week, once it's standing between a regression and a quiet pass? 👍
#dusk $DUSK @Dusk
