revisits
RUNNING · supervisor aliveworking normally · agents on codex
Now
| run | stage | time | epoch | ×1 AUC | Δ vs incumbent |
|---|
| no-dsu | training | 16m 05s | 8/10 | 0.9226 | -0.0013 |
Up next
- no-anatomical-bias prepared
- no-bilateral-smoothing queued
- no-resolution-degradation queued
- plane-contrast-feature-adapters queued
- plane-conditioned-batchnorm queued
- rn18-stochastic-depth queued
- ema-0999 queued
- ema-09995 queued
Results
Curves
Black: control. Each run keeps its own colour; the incumbent is the thickest line and the run training now is dashed. Finished runs are faded; hover or isolate one to read it. Beyond eight runs the oldest turn grey. Click a legend entry to hide a run, double-click to isolate it.
Per label
×4 AUC Δ ×1000 against each run's own reference (the incumbent when it ran); blue up, red down, full colour at ±10. Hover a cell for both AUCs.
Full status
revisits · codex · RUNNING · supervisor alive
Budget 10m 10s–30m 32s training, 34m 26s per experiment · epochs free · fold 0
Inbox: 1 request(s) waiting for the supervisor's next safe point.
Now
training 16m 05s no-dsu, epoch 8/10, val AUC 0.922569 (incumbent 0.923912 at this epoch, -0.0013)
Backlog 8 queued (refill below 3, target 5); forced seeds 2/11 done, research waits for the rest
1. no-anatomical-bias [prepared]
2. no-bilateral-smoothing [queued]
3. no-resolution-degradation [queued]
4. plane-contrast-feature-adapters [queued]
5. plane-conditioned-batchnorm [queued]
6. rn18-stochastic-depth [queued]
7. ema-0999 [queued]
8. ema-09995 [queued]
next candidate: no-anatomical-bias (prepared)
Score
Starting reference AUC: 0.928062 (historical-control)
Incumbent: historical-control | AUC 0.928062 (x1 0.924507)
AUC change since start: +0.000000
Retained experimental gain: +0.000000 (0 retained acceptances; vs historical-control)
Results 2 decided, 0 provisionally accepted
idea delta adoption 80% CI verdict
dense-plane-readouts -0.005248 [-0.007138, -0.003362] reject
plane-contrast-readouts +0.000918 [+0.000096, +0.001703] reject
Cost (API-equivalent, not subscription quota)
completion_reporter/completion 2 jobs n/a
developer 4 jobs n/a
orchestrator/review 4 jobs n/a
total 10 jobs n/a
usage incomplete:
tokens: 0.68M new input, 3.54M cached, 0.04M output
completion reports: 2 reported
Artifacts: /home/claudio/src/knee-abnormality/outputs/autoresearch/revisits