Measurement / Observation dependence

Repeated ticks are not repeated evidence.

This exhibit makes the dependence behind in-game calibration measurements visible. It reads residual_autocorrelation.json -> sports.<sport>.autocorr_values.{model, market}[], retaining the model and market arrays as separate populations.

What the measure says

Lag-1 residual autocorrelation compares each within-game signed residual with the one immediately before it. A value near 1 means consecutive ticks carry almost the same information: the probability path changes smoothly while the terminal outcome stays fixed.

The site reports ticks and games because they answer different questions. Ticks describe how often the system was observed; games describe the independent units behind a calibration measurement. This page does not infer an exact effective sample size. See the effective-sample-size finding and the calibration page for their published context.

Eligibility floors: at least 10 rows per game and residual variance of at least 1e-9. Exclusions are shown by side below.

Reading trail

Read first: Read the effective sample size finding before treating repeated ticks as separate evidence.

Next question: Which time and probability cells carry the least support behind their gap?

Read the analysis: How many games actually sit behind a bin (sources regenerated)

MLB

Published corpus: 27,351 ticks across 178 candidate series.

MLB within-game distributions

Each dot is one eligible game-side series. Model and market are separate populations; dots are not paired.

Model: 173 eligible series

Median 0.9713; above 0.9 89.6%

Market: 173 eligible series

Median 0.9645; above 0.9 86.1%

public/data/showcase/residual_autocorrelation.jsonSource as of 2026-09-16n not published

International soccer

Published corpus: 4,265 ticks across 27 candidate series.

International soccer within-game distributions

Each dot is one eligible game-side series. Model and market are separate populations; dots are not paired.

Model: 26 eligible series

Median 0.9663; above 0.9 84.6%

Market: 26 eligible series

Median 0.9517; above 0.9 76.9%

public/data/showcase/residual_autocorrelation.jsonSource as of 2026-09-16n not published

Published-series summary

Formula: median is the middle value after sorting each side's published array; share above 0.9 = count(rho > 0.9) / eligible series. Each side remains a separate population.

Sport	Side	Eligible series	Median autocorr	Share above 0.9	Skipped
MLB	Model	173	0.9713	89.6%	low rows 4; flat residuals 1
MLB	Market	173	0.9645	86.1%	low rows 4; flat residuals 1
International soccer	Model	26	0.9663	84.6%	low rows 1; flat residuals 0
International soccer	Market	26	0.9517	76.9%	low rows 1; flat residuals 0