← Back
Optimization Lab Experiment Builder Base Model #45

Create Projection Experiment

Launch a controlled BettingOdds optimization run using lane-aware presets, bounded parameter spaces, and research-grade candidate generation.
Dashboard Experiments Models Leaderboard

Research Lane

Start with a research lane. The lane preloads the target metric, default search mode, candidate budget, and parameter families for the experiment.
Lane-first Preset-driven Editable
Lane Guidance
Minutes Calibration focuses on role stability, redistribution, and core minutes weighting.

Experiment Setup

Define the research target, date scope, and experiment metadata for this optimization run.
Research-grade Versioned Deterministic
Range Guidance
Use shorter windows for tighter recent calibration, season windows for broader stability checks, and custom windows only when you want a controlled historical research slice.

Search Strategy

Choose how the optimization lab explores candidate models around the current champion.
StructuredAdaptive
Used by Random, Adaptive, and Structured Adaptive as the total candidate budget.
Number of historical top models used to infer useful search direction.
Legacy Adaptive only. Narrows bounds around prior winner regions while preserving exploration.
Mode Summary
Premium deterministic search using local probes, pair interactions, and bounded refinement around the champion.
Recommended Usage
Best for elite research runs where you want smarter convergence, cleaner auditability, and stronger search discipline.

Parameter Search Space

Review and edit the legal search envelope for each parameter. Structured Adaptive will stay inside these bounds while exploring around the active champion.
4 parameter(s) Champion-centered Bounded
Parameter-Space Guidance
Keep ranges local and intentional. Elite optimization converges faster when parameters are tuned in bounded neighborhoods rather than broad global sweeps.
RollingMinutesWeight
PlayerBuild
Decimal 2 dp
RelevanceMinutesWeight
PlayerBuild
Decimal 2 dp
RedistributionStarterBoost
PlayerBuild
Decimal 2 dp
RedistributionMinAvailability
PlayerBuild
Decimal 2 dp

Launch Summary

Live
Active Base Model
structured-0010-0003
Research Lane
Minutes
Target Metric
MinutesMae
Search Mode
StructuredAdaptive
Parameters
4
Grid Estimate
2058
Candidate Budget
24
Execution Interpretation
Premium structured search allocates candidate budget intentionally across probe, interaction, and refinement phases.

Lane Decision Support

Minutes Calibration
Best first test lane. Optimizes minutes weighting and redistribution behavior that usually cascades into better points, PRA, and total projections.
Redistribution / PRA
Focuses on role replacement and redistribution logic that often affects player-stat downstream accuracy.
Team Factor / Fair Spread
Best for spread-side calibration when team-factor weighting and clamp behavior need refinement.
Fair Total Calibration
Best for total-line tuning using offense, defense, and team-factor calibration neighborhoods.

Mode Decision Support

Grid
Best when parameter count is low and exhaustive coverage is still computationally manageable.
Random
Best for broader bounded exploration when full combinatorial coverage would be too expensive.
Adaptive (Legacy)
History-informed but still built around legacy randomized search. Keep for backward compatibility and comparison.
Structured Adaptive
Premium mode. Uses deterministic local probes, interaction exploration, and bounded refinement around the champion.

Actions

Pre-Launch Check
Confirm the lane, target metric, date range, and parameter bounds before launch. All generated candidates will be cloned from the currently active base model.
Cancel