Novel statistics
Each one was searched against prior art before it was built, re-derived a second time to try to break it, and published with the confounds that could make it wrong. They are descriptive market science and structure measurements — every artifact carries edge_claimed: false, and none of them is a profit claim.
Per sport, the number of hours before tip by which half of the pre-game absolute line motion is complete.
mlb -- half of all pre-game line motion is complete by then; n=1,307,128 move pairs observed 2026-06-18 to 2026-07-17 (30 days, 30 daily files), final-hour share 0.1813
The 0-1 share of a sport's game clock that stays genuinely contested before the score gap crosses a point-of-no-return (near-median threshold).
mlb -- 95 of 178 games decided at a 3-runs gap
Per team, the win-prob dependence on its single most-indispensable player, under two independent estimators, with a cross-method agreement flag.
DEN -- Nikola Jokić, Elo on/off net-rating delta 23.713; the second estimator names a different player (agreement 1 of 30 teams)
The market's excess resolving power over a scoreboard-only model, per bit of uncertainty still left in the game, by checkpoint.
soccer_intl -- mean over 19 in-game checkpoints; the market's lead rises 0.6393 to 1.7056 as the game runs
Structural market overshoot at large in-game moves, scaled by how far short our model falls of beating the market at max disagreement -- a measured failure to harvest, not an edge.
soccer_intl -- the market overshoots 0.2856 on big in-game moves, yet our model is the closer forecast only 0.2152 of the time at max disagreement (n=4,406)
A published null is a result, not a failure. This stat measures our own inability to convert a real market overshoot. It is on this page at full size because the ones that do not work are the reason to trust the ones that do.
Per team-season points tax from back-to-back congestion: its b2b frequency times the tested b2b performance penalty.
DEN 2025-26 -- the most congested schedule of 90 team-seasons; 17 back-to-backs (frequency 0.218)
How a stat gets published here
The definition is searched against the published literature and the public analytics canon before anything is built. The verdict is written down even when it is unflattering -- INCREMENTAL means someone got most of the way here first, and the card says so.
The stat must be computable from artifacts already committed in this repo. No new scrape, no private feed, no number that cannot be recomputed from the cited source files.
The result is rebuilt a second time, by a different route where one exists, specifically to break it. Load-Bearing Index ships with two independent estimators that disagree on 29 of 30 teams -- that disagreement is printed, not hidden.
Every stat publishes what would make it wrong: sparse tail buckets, excluded sports, absolute-move churn, conflated model misspecification. Confounds are part of the stat, not an appendix.
These are market-science and descriptive measurements. None of them is a profit claim, and each artifact carries the flag saying so.
Nothing on this page is a betting edge. The claim is narrower and harder: these are measurements that are honest about their own limits, and you can recompute every one of them from the artifacts named on the cards.