Compare deterministic auto-scores (tests, lint, coverage) against manual scores in .mek/drift-baseline.json.
Divergence rule (ScoreRift-compatible): flag when
|auto - manual_as_float| > divergence_threshold(default 0.15), ANDconfidence ≥ min_confidence(default 0.5).
Also flag any dimension where auto < floor (ratchet — regardless of confidence).
/mek-drift init— seed the baseline.- Open the baseline; fill in
manual(letter grade) for the dimensions you care about. - Work normally. Stop hook checks for divergences after PR-sized changes.
/mek-drift status— on-demand check./mek-drift acceptor/mek-drift ack— resolve a flagged dimension.
Only python is shipped. Other presets (JS, Go, Rust) are v0.2.0.