← Back
Optimization Lab
Experiment Details
Completed
Random
diagnostic-pra-mae-test 3
Review experiment performance, inspect ranked challengers, and advance the best model
into validation or promotion workflows.
Status
Completed
Current lifecycle state of this optimization run.
Target Metric
PraMAE
Primary ranking metric used to score challengers.
Candidates
2
Total challengers generated and evaluated for this experiment.
Base Model
v1
Champion baseline used to generate this experiment’s candidates.
Experiment Overview
High-level research context for the current run.
Target: PraMAE
Mode: Random
Research Goal
This run optimizes candidate models against PraMAE,
then ranks challengers against the experiment scoring framework.
Execution Mode
Random run. Candidate generation is sampled from inside the configured parameter envelope.
Experiment Notes
Automatic random-search experiment
Decision Support
GovernanceInterpretation
Use this page to identify the strongest challenger, verify the metric profile, and then move
the model into validation before any final production promotion.
Best Practice
Promotion should remain exceptional. Even when a challenger wins this leaderboard,
validation against the champion should remain the default next step.
Candidate Quality Lens
Prefer challengers that improve the target metric without materially degrading adjacent signals
like spread, total, and box-score accuracy.
Top Challenger
Target: PraMAE
Official Winner
Winning Model
Highest-ranked candidate from this experiment based on the configured target metric.
Model Version
random-0004-0002
Target Metric Value
9.77
Status
Completed
Promotion Posture
Validate first unless governance explicitly approves direct promotion.
Why This Candidate Matters
This challenger finished at the top of the current experiment leaderboard. Treat it as the leading
hypothesis generated by this run, not as an automatic production replacement.
Candidate Leaderboard
Ranked challenger results for this experiment, ordered by PraMAE.
Target: PraMAE
Mode: Random
2 candidate(s)
Leaderboard Reading Guide
Rank is determined by the experiment target metric, but promotion decisions should also consider
secondary error metrics, parameter deltas, and subsequent validation performance.
| # | Model | Status | PraMAE | Spread MAE | Total MAE | PRA MAE | Parameter Delta | Actions |
|---|---|---|---|---|---|---|---|---|
| 1 |
random-0004-0002
Winning candidate
Random candidate generated from base model version 1.
|
Completed | 9.77 | 17.24 | 14.20 | 9.77 |
{"AlphaOff":"0.18","BetaDef":"0.05","RedistributionMinAvailability":"0.82","RedistributionStarterBoost":"1.15"}
|
|
| 2 |
random-0004-0001
Random candidate generated from base model version 1.
|
Completed | 9.80 | 17.24 | 14.20 | 9.80 |
{"AlphaOff":"0.26","BetaDef":"0.06","RedistributionMinAvailability":"0.72","RedistributionStarterBoost":"1.3"}
|