Release notes

Current public release

The release manifest could not yet be read.

v1.1 — evidence-corrected successor

• The league reference is now frozen entirely before the target period, and paired intervals find the short windows have higher error. Season-to-date is not distinguishable from that reference.
• The forward test also uses one cutoff-fitted 24-state RE transform on both the pre-cutoff history and post-cutoff target; the descriptive full-season RE matrix is excluded from that test.
• Each value first averages offensive run value per PA inside an appearance, then gives every appearance equal weight. Lower is better for the pitcher. Windows include both starts and relief outings and disclose the exact mix. Majority role selects only the window lengths: 5 and 10 trailing appearances for a starter majority, or 10 and 20 for a reliever majority; the trailing stream is never filtered by role. PULSE v3 corrects this pitcher contract while preserving every mapped pitcher value and the complete v2 hitter/forward evidence.
• Play parsing consumes the complete event grammar, and game finalization rejects a truncated scheduled half while preserving valid walkoffs, skipped home halves, extras, and shortened games.
• Builds verify every local raw byte against its admitted receipt before writing, stop at the first failed dependency, and no longer let pytest rewrite or reseal evidence.
• The API v2 contract carries release, source-tier, as-of, coverage, capture, model-id, and model-hash lineage. Game projections preserve every literal game field and disclose the literal source hash.
• Invalid query fallbacks remain visibly disclosed, canonical-host metadata is explicit, and the public version and methodology copy match the product that is actually live.
• The baserunning v2 tournament correction did not promote an estimator. Active CROWN v3 now states “single-identifiable-runner standalone terminal SB/CS/PO/POCS only; multi-runner, unidentified, advancement, and strict compound secondaries remain unallocated.” Historical CROWN v2 remains unchanged; v3 corrects only that disclosure and changes no numerical value. A raw-derived lane audit finds 150 fitted unallocated rows (129 SB, 17 PO, 4 error-negated CS), or 158 corpus-wide after eight censored SB rows. These counts are not artifact fields and authorize no allocation. A future primary-only counterfactual remains deferred research.
• biographies_2025_v1 is restored at its historical byte identity; it was not silently regenerated against the corrected successor models.
• The 2025 event-policy totals are 3,439 SB and 981 CS; official-gamelog credits are 3,440 SB and 989 CS across exactly nine clean games. One SB annotation is embedded on a balk advance and is not reclassified as a separate steal; eight error-negated CS/POCS attempts are not charged when the out does not stand.
• Pitching innings are outs-authoritative baseball notation: the digit after the point is 0, 1, or 2 outs, never a decimal fraction. Paul Skenes' 563 outs display as 187.2; ERA remains 1.97.

Exact rebuilt analytical identities

• historical pulse_2025_v2 — 45349378d1b4286ccc2c77d97cfd8222ad7a43ef56cb90d33f8dc4579d9f7779
• active pulse_2025_v3 — bcf5b1ee35c0c8867dc04f39fcee5db8634a4e730eb2255a273b67d276c67a35
• baserunning_impact_rate_discrete_actual_2025_v2 — b233bbbb735bc7cd4b43bb01079f53c93fad9152ddc3b61e26e367672a820370
• historical crown_actual_partial_2025_v2 — d20353e92b5a11a56b6175bdf7d2402baa16b743a9d0ed5a26c6603e615a0b4a
• active disclosure-only crown_actual_partial_2025_v3 — a7af999e05eecbed6e506e4fc34a2104dfbd7aa99a15bf8278783428a8657a24
• restored biographies_2025_v1 — 563aab423d8361ae3b913b2ba30075a787bbb9bb21b26d6a38889dd42c40ca9c
• historical season_stats_2025_v1 — ac14115b79304d13763e79ff73a89d1311f1dd5c55d09e980531eb54559aeea4
• historical disclosure-only season_stats_2025_v2 — 47b909ca2ab2ca0e1f22da756261c4ded6c6caf89d9577eba93b7ebe8504bc18
• active representation/disclosure season_stats_2025_v3 — 3419d28661a8aaac935515b8a522c9969f790fddbb2802802097a4fd4139e048

v1.0 — immutable historical release

v1.0-hardball-signal remains fixed at 76b246f286082c3f6817c065e00e71a9b4e17b7c. The successor does not move that tag or rewrite its artifacts. Corrections are additive and versioned so historical claims remain auditable.