Findings / Effective Sample Size
How independent is our data, really?
This historical ledger estimates row dependence with lag-1 residual autocorrelation. Its published anchor is a proxy, not a verified count of distinct games.
Support label: this ledger is the joined-corpus measurement, revision 1. It was taken on the MLB and international soccer in-game corpus before that corpus was re-segmented into one stored file per game, and its recomposition on the segment-clean corpus is pending, so every row below describes the joined corpus rather than the published revision 2 population.
The AR(1) estimate approximates effective rows from the recorded residual rho. The published anchor takes min(AR(1) estimate, stored series count). The stored series field counts (game_id, side) pairs. Multiple sides and mixed-game input paths mean this count cannot establish how many distinct games were observed.
| Corpus | Rows | Stored series | Residual autocorr (rho) | Effective sample (AR1 proxy) | Published anchor (proxy) | Implied interval-width factor |
|---|---|---|---|---|---|---|
| MLB | 78,986 | 227 | 0.9809market 0.9799 | 761 | 227 | 18.7xAR1 estimate 10.2x |
| Soccer (Intl) | 9,003 | 51 | 0.9624market 0.9489 | 172 | 51 | 13.3xAR1 estimate 7.2x |
Scroll horizontally for all columns.
How to read this — the historical MLB snapshot has 78,986 rows and a published anchor proxy of 227, capped by the stored series count. The last column is an implied interval-width factor under that proxy; it is not an instruction to scale a published interval. Intervals on the regenerated corpus are re-estimated by cluster bootstrap, never widened from an earlier number.
Formula
AR(1) is a first-order approximation of a smooth residual path, not a full independence model. This descriptive-only snapshot makes no advantage or return claim.
These historical proxy values describe the joined corpus only. Distinct-game counts and replacement intervals require the pending segment-clean recomposition.