Release review
Two model versions, case by case. Error regressions and unsettled answers flagged; the checks stay with you.
Currentflow-v2.3
Candidateflow-v2.4
Warehouse regression set12 paired cases · mean endpoint error (px) · 12 with trajectories
Not ready: 3 error regressions, 2 unstable cases that pass on error.
Average case error
3.132.69px-13.9%
CaseError change
5 of 12 casesNeeds a decision first
Regression Unstable
aisle-042Reflective floor
Current model3.40px
Candidate5.20px
+1.80px
Error by frame px · lower is betterCandidate peak at frame 8
Current modelCandidate
Error increased by 1.80 px, exceeding your 0.3 px limit.
Refinement trajectory update per iteration · px12 iterations
Current · 4% lateCandidate · 32% late, 3 reversalsreversal: update grew > 5%
Late revision: share of all refinement in the last third of iterations (limit 25%). Reversal: an update larger than the one before it by more than 5% (limit 2).
The candidate made 32% of its refinement in the last third of its iterations and reversed direction 3 times, above your 25% / 2 limits. The current model settled at 4%.The candidate's error rises mid-sequence and its refinement never settles: most of its revisions arrive late. Investigate before release.
Keep this case in your next release check.
Illustrative data · explore the workflowLower is better · equal-weight case averages · stability from label-free trajectories