Build your Edge

The four ways through

• Lab — ask the question first

  Pick a catalogue question and read its sample, baseline and lift before you build anything. Answers are precomputed, so the number you see is the number that was graded.

  64 answers · Test a football situation against graded games.

• Bench — assemble the model

  Snap bricks together into a model, see what each one contributes, and backtest it walk-forward against the same settlement ledger the house is graded on.

  1008 bricks · Connect typed bricks into your own model.

• My models — what you own

  Every model you have built, with its record, its gate reason in words, and whether it is currently clearing breakeven.

  398 cards · Follow the walk-forward record of every model.

• Remixes — start from someone else

  Remix a published model, change one thing, and keep the lineage so the two records can be compared honestly.

  3 lineage edges · Fork a public model with attribution intact.

Sharp money

lens - family A - proven - remixed —

• Feed

  = its Signal becomes a column the learner trains on (Signal -> as feature -> Frame).

  Features joined

  —

  pregame-close watermark - never a postgame column

• Filter

  = only games where its verdict clears the bar reach the learner (Signal -> >= x -> keep rows).

  Verdict threshold

  —

  Slate shrinks N -> M games

• Module

  = it votes in the ensemble with a weight (Signal -> Ensemble). Proven lenses only - the Ensemble port enforces it.

  Weight

  —

  others rescale to 1.00

States