Changelog

0.12.0 — 2026-08-24

Standalone skill pack

• Made all generic analytics and modeling skills usable without sports_ds, repository-relative paths, or an end-to-end pipeline.
• Added sports-ds-bridge as the explicit optional path from toolkit loaders, core APIs, CLI commands, and pipelines to portable user-owned artifacts.
• Added validated fold-metric, held-out-prediction, calibration, Elo, simulation, visualization, and reporting handoffs.
• Added isolated helper, boundary, negative-case, and cross-skill smoke tests.

Toolkit and release engineering

• Eliminated pandas DataFrame-fragmentation warnings in player-form feature construction while preserving exact output behavior.
• Added offline CI across Python 3.10–3.13 and scheduled/manual live-data CI.
• Unified the repository, plugin, and toolkit release version at 0.12.0.

0.11.1

MLB player path to parity

• Parallel boxscore fetch + resume cache; optional full panel parquet cache
• Richer batter features: OPS/ISO/K%/BB% form, rest days, batting-order slot, opp starter K/9
• Lean proven feature set + starter filter + shrunk player baseline + hist-GBR/ridge/blend ladder
• Live dense 2023→2024 walk-forward beats constant on fantasy points (and TB/hits/PA)
• mlb-player-pipeline supports --target fantasy_points|total_bases|hits|plate_appearances

0.11.0

Player paths (NBA + MLB)

• NBA player panel via bulk load_nba_player_boxscore + nba-player-eda / nba-player-pipeline
• MLB batter panel via cached per-game mlb_boxscore + mlb-player-eda / mlb-player-pipeline (--max-games for smokes)
• Shared player form engine + generic pipelines/player_model.py (NFL refactored onto it)
• Sport-specific feature defaults for NBA/MLB fantasy proxies
• Unit tests for NBA/MLB player form/pipeline; optional live player integration tests

0.10.0

Package depth

• Richer team form features: EWMA, rest days, home/away split form, calendar week, offense-vs-defense proxies
• RICH_WIN_FEATURE_COLS + sports-ds nfl-win-rich (logistic / hist-gbm / form+Elo ensemble ladder)
• Form+Elo probability ensemble (sports_ds.models.ensemble)
• NFL player-level path: load_player_game_panel, player form features, sports-ds nfl-player-pipeline / nfl-player-eda
• Feature registry expanded for rich + player features
• Unit tests for rich features, ensemble, player form; optional live rich/player integration tests

0.9.1

Skill depth + tests

• Full operator-manual pass across skill pack
• Heavier offline integration tests (25 passing at tag)

0.9.0

Package depth

• Generic team-margin and team-Elo engines (pipelines/team_margin.py, team_elo.py)
• NBA/MLB margin + Elo CLI: nba-margin-pipeline, nba-elo, mlb-margin-pipeline, mlb-elo
• sports-ds feature-registry
• calibrate / leakage-audit accept --sport nfl|nba|mlb
• Feature registry module + unit tests (13 passing)

Skill depth

• Raised environment-setup, sports-visualization, model-card, anti-slop, results-reporting, experiment-log, nflreadpy, model-interpretation to package-aware manuals
• Multi-sport plot scripts (--sport) + walk-forward metric plotter

0.8.0

Multi-sport package paths

• Shared team-game panel helpers (sports_ds.data.panel)
• Generic team-win walk-forward engine (sports_ds.pipelines.team_win)
• NBA / MLB / NHL loaders + *-eda / *-win-pipeline CLI commands
• NHL corrupt-score dump detection (skips unusable seasons like constant 2-3 scores)
• Docs/skills updated for multi-sport package paths

0.7.0

Package

• Add sports_ds.ratings as-of Elo
• Add sports_ds.metrics (brier, log-loss, ECE/calibration)
• Add sports_ds.audit.leakage package leakage audit
• Add margin regressors + nfl-margin-pipeline
• Add nfl-elo walk-forward baseline pipeline
• Add NBA loader + nba-eda / nba-win-pipeline (requires [multi])
• Expand CLI: calibrate, leakage-audit, margin, elo, nba commands

Docs

• Product charter (docs/product-charter.md)
• Agent runbook (docs/agent-runbook.md)
• Architecture + README synced to real package surface

Tests

• Unit tests for features, splits, Elo as-of, metrics, leakage audit, regressors

0.6.0

• Initial public sports_ds + skill pack with NFL win pipeline and deep skill manuals