Grandissimo Path Lab

Win probabilities

Per game: the two providers, their blend at the current slider, and your number (win % of the first-named team). A decided game is fixed and locked. Effective is what the number in force is worth that far out with the future-week discount on: each simulated season draws the true line around it (team strength drifts ≈0.75 pts/√week, plus ≈2.4 pts of game noise — calibrated on ~1,300 look-ahead vs closing lines), which regresses far-out favourites toward 50% and makes a team's future games move together. Hand-edited games get the same treatment (your week-16 number is a forecast too).

Field ownership

Two modes (“Your edits”, top bar). Default — this week locked, later weeks reference: as below. Lock all my edits: every leg you edited is played exactly as entered — your shares become whole opponents among whoever is alive in each simulated season, no regression — and every win % you typed is treated as the truth (no future-week discount on that game).

Current leg — locked: your read is the field, exactly: shares are converted to whole opponents (largest remainder) and every simulated world has exactly those picks, assigned only to opponents for whom the team is legal. Later legs — reference: your shares tilt the fitted field rather than replace it. Per opponent, the model's own pick distribution is taken at today's lines and at the simulated season's kickoff lines; multipliers fitted at today's lines reproduce your shares in expectation, and the picks are drawn from the season's distribution × those multipliers — so the field still moves with the lines in every season exactly as the fitted model does, and your entry gets no edge from “knowing” a field that cannot see line movement (that edge was what inflated every EV by ~$100k before this build). With the future-week discount on, a read h weeks out keeps weight 1/(1+(h/half-life)²) and the rest follows the fitted model over all teams. Shares are percentages of the whole field: if your list sums to less than 100, the remainder (100 − sum) plays the fitted model — a partial read never forces the whole field onto the teams you named. List shares summing to 100 when you mean the entire field. Stale reads are guarded: a listed team outside the field's pick floor on today's lines (win % under the floor, or too far behind the week's best) has no share to tilt, so it plays as an unlisted team — the model picks it only in seasons where its line moves inside the floor; the leg header flags such teams, and after a run shows how much of the read sat outside the floor on simulated lines. An opponent who has already used every team you listed takes its best legal favourite (counted as a forced pick). Legs you leave blank follow the fitted field model under the pick floor — no slop. “Model (rest)” deliberately hands a share back to the fitted model.