SIQX publishes its complete moneyline record — including the losing stretches. The model is held to two standards: win rate above the market’s implied probability, and positive expected value. We do not sell a pick product until it clears them.
Two cohorts are closed below. v13 finished 19–16 over 35 resolved picks — above the market, but on team aggregates later found to be defective. On Aug 24, 2026 a question about a single team’s profile exposed the flaw: team scores counted a bench bat and an everyday regular equally, and an injured starter the same as one making every turn. Per the live-forward rule a fix that changes pick selection resets the record, so v14 starts from zero. No paid recommendation product is for sale, and none will be until the live record demonstrates a real, positive-EV edge.
v13 finished above the market — 54.3% against 45.2% implied — but the Phase-1 logistic coefficient stayed negative throughout, meaning a larger behavioral edge did not predict a higher win rate within the sample. The cohort also ran on team aggregates that counted every qualified player equally regardless of whether he was playing. That was found on Aug 24, 2026 through a public question about one team’s rotation, not through the pick record, and corrected in the direction the evidence pointed rather than the direction that would improve results. The record is closed exactly as it stands and is not pooled with v14.
v11 ran with four count-split components silently duplicated — a Statcast filter the API quietly ignored. Once corrected, the hitter–xwOBA correlation re-derived from +0.573 to +0.444 and the pitcher correlation from −0.526 to −0.708. On September 9 a second score-side defect was corrected — the historical FIQ had been a hard-hit proxy, an outcome measure inside a behavioral index — and the hitter correlation re-derived again, from +0.444 to +0.31 on 1,207 closed-season hitter-seasons. Pitchers were unaffected. Each downward revision removed an outcome leak; neither tuned anything. Per the live-forward rule, a methodology change resets the record: v11 is closed exactly as it stands, and every pick from June 10 forward is graded as v12. v12 has no cohort of its own because it resolved no picks: a pitcher-shrinkage change on June 14 made it v13 four days later, and that change altered no pick selection, so the record carried forward unbroken. The cohort therefore runs v11, then v13.
A pick is only good if it wins more often than the odds said it would. Beating your own coin flips is meaningless; beating the market’s is the whole game. v14 tracking has just begun — conclusions wait for 100+ picks.
Was each pick +EV at the price we posted — not an idealized number? EV is graded at opening odds, the price a follower could actually have taken.
CLV is captured opportunistically as a sanity check — never required, never used to grade the core edge. Live in-game odds are excluded so they can’t flatter it.
Components, weights, and thresholds are frozen. Nothing is ever tuned against the pick log. If the methodology changes, a new version is declared and the record restarts from zero.
PPD games are auto-voided after two days and excluded from every calculation — never quietly counted as wins.
No volume-padding, no chasing. When no underdog clears the behavioral-edge threshold, there is no pick — and we say so.