Scopes

Research · methodology note · Sept 21, 2026

The MLB binding artifact: what we caught and removed

The sports scorecard flags a game when Kalshi's price diverges from the de-vigged sharp-book fair value by at least six points. Between July 18 and August 24, 2026 it recorded about 226 such flags, all MLB. On review, almost all of them were not real divergences: a bug bound a game's fair value to the wrong game or the wrong snapshot (a first-pitch time and game-date matching error), which inflated the measured gap. We fixed the binding on August 24, and the inflated gaps disappeared the same day.

How we know it was measurement, not the market

The full snapshot distribution settles it. The typical MLB game always tracked the sharp book to within about two points: the median gap was ~2pt before the fix and stayed there after. What changed was the tail. The 90th-percentile gap, the part that produces flags, collapsed from 9 to 13 points down to about one point overnight when the fix shipped, not gradually. A market that is genuinely getting more efficient tightens slowly; a cliff on the exact day of a code change is a measurement artifact. Trading volume actually rose over the same period, so this was not markets thinning out either.

WindowMedian gapp90 gapAvg volumeFlags
Jul 18 to Aug 24 (pre-fix)~2 pt9 to 13 pt45 to 72k~40 / wk
Aug 24 onward (post-fix)~0.4 pt~1 pt86 to 110k0

The flags themselves confirm it: across 223 resolved pre-fix flags the record was 109 wins to 114 losses, a coin flip, with no edge that survives the sample size. Real divergences would have shown a durable win rate above the market's implied line. These did not.

Correction, November 1, 2026. Two statements above are wrong; the conclusion stands.The median did not stay at two points. Recomputed from every recorded MLB snapshot, the median gap was 1.95 points before the fix and 0.42 points after it (Aug 24 to Sep 21), as the table shows; the 90th percentile fell from 11.35 to 1.21 points. The whole distribution tightened on the day of the fix, not only its tail. A change of that size on the day of a code change is still a measurement artifact.The 109 to 114 record was scored by the same bug. The binding fix of August 24 noted that the scorecard's sports resolver had the same flaw, matching a flag to the wrong game, and that already-resolved flags would be corrected; that correction was never made. Rechecked against Kalshi's own settlement of each flag's game, 111 of the 226 recorded results disagree with it. Scored correctly, the fair value was right on 96 of the 226 flags and the market on 130 (42.5%). A fair value bound to the wrong game should lose to the market, and it did. These flags remain withheld from scoring, so no scorecard figure changes; their recorded resolutions are left as they are and the discrepancy is disclosed here.

What we changed

The scorecard is append-only, so we do not delete or rewrite a flag once it is recorded. Instead we extended the same validity gate that already withholds in-play flags: a sports flag now counts toward the public record only if it was pregame and recorded on or after the binding fix. The 226 pre-fix MLB flags remain in the underlying record, marked not-scored, so the correction is auditable rather than hidden. The live sports scorecard now reflects only flags measured with correct game-binding.

Cite this study

Free to cite with attribution to Scopes. For the citation format, boilerplate, and media contact, see Press & citation.