Data and sibling repositories

The forecasting pipeline uses three sibling repositories. Their newer files and acquisition code can support updates; a new collection system is not needed for every input. This inventory was checked against local records on October 30, 2026. Dates below describe the inspected files, not a promise of complete coverage.

Source Verified local contents Reuse
player_value, local directory war Men’s raw Cricsheet: 897 Tests through October 22, 2026; 2,607 ODIs and 3,646 T20Is through October 25. Player UUIDs and match lineups are present. International appearance history and stable IDs. Its fetch command downloads each format with checksums and retrieval receipts.
side_order 11,905 international match results, 1877–October 25, 2026, with match IDs, teams, start and completion dates. Match calendar and independent coverage checks. Its weekly workflow provides the update/publication pattern. Results alone cannot identify which players appeared.
ipl_auction 1,188 raw IPL matches, 789 players, April 2008–April 12, 2026; 19 matches after 2025. Auction records cover seasons 2008–2026 and include nationality and specialty. IPL performance, affiliation evidence and reviewed name aliases. Raw IPL data are newer than its cached season statistics and WAR, which end in 2023.
python-espncricinfo Player client exposes date of birth, role, batting/bowling styles and ESPN player IDs; match client exposes scorecards. Enrich player profiles and investigate missing scorecards. This is acquisition code, not an already collected, complete domestic player panel.
get_cricket_data, toss_up, test_selection Historical match summaries and a documented source-feasibility audit. The older scraper uses Python 2. Historical fixture discovery and design evidence. The inspected summary tables lack the domestic player-performance denominator.

What the current forecast actually consumes

national_selection.data.import_local reads raw IPL matches from ipl_auction, raw international matches from war, and the match calendar from side_order. It records source-file hashes and exports linkage and coverage audits. The benchmark import defaults to December 31, 2025. The monthly refresh passes a newer endpoint from the validated calendar and writes separate current inputs, leaving the benchmark unchanged.

The international archives were retrieved on October 27, 2026. Their local 2026 records add 21 men’s Tests, 100 ODIs and 535 T20Is beyond the frozen analysis period. The IPL cache is only part of the 2026 season. Advancing the forecast to October using that cache would omit later IPL performances.

Cricsheet’s official downloads include an IPL JSON archive using the same player UUID convention. The downloader in player_value accepts Test, ODI and T20I. The monthly refresh uses that same archive-validation pattern for all three formats and IPL. Its raw and prepared international tables are ignored locally rather than tracked in Git, so a fresh GitHub checkout must run acquisition or receive a versioned data artifact.

Reuse contract for monthly updates

  1. Refresh international archives with the existing player_value downloader. Use all available participant history for prior-cap exclusions, rather than only the recent window prepared for player-strength estimation.
  2. Consume side_order’s dated calendar and compare match IDs by country and format. A missing fixture remains visible; a completed calendar window does not make incomplete player records complete.
  3. Refresh the IPL archive, then use this repo’s player-match accounting. Keep Cricsheet UUIDs as the player key. Auction P0001-style IDs are assigned from a sorted name list and should not be treated as durable cross-refresh IDs.
  4. Preserve the frozen backtest inputs and reports. Store each new forecast with its own cutoff, input hashes, model policy and annual fit date. Score only horizons with completed follow-up, retaining coverage qualifications.
  5. Publish validated output tables, then render documentation from those tables.

Country and role evidence need observation dates. The current conservative rule makes a season’s nationality available in January of the following year. An updated pipeline must not silently backdate 2026 auction evidence. Age should be computed from birth date at each cutoff, not copied from a current profile’s age. Current player-value ratings and full-season WAR are not automatically valid historical Xs: their training and performance windows must end before each cutoff.

Included data and audits

The site publishes current snapshots when available and retains the frozen benchmark. The monthly workflow records its actual source versions and coverage dates. The table above documents the sibling-file inventory before the refresh, not the date of the current forecast.

Affiliation reference used by automation

The local ipl_auction/data/acquisitions/sources/clod/ipl_2008_2026.csv is not tracked at the upstream Git commit inspected for automation. The three fields year, player, and nationality are therefore preserved in this package’s reference/auction_affiliations.csv: 3,385 source rows, without inferred values. The importer retains its conservative January-after-season availability rule. Match feeds update monthly; this affiliation reference needs an explicit update when new source evidence is collected. Unlinked players remain in the audit.