NBA Injury Report PDF Extractor

Turn a supplied NBA-format injury-report PDF into player rows with the reported status, reason, team, game date and source page. Use the output in sports research, availability dashboards or an existing data pipeline. This is an independent tool, not affiliated with the NBA. It does not diagnose injuries or authenticate reports.

Quick start

For a labelled example without downloading anything:

{ "demo": true }

For a real historical public NBA document:

{
  "pdfUrl": "https://ak-static.cms.nba.com/referee/injury/Injury-Report_2026-02-05_10_30AM.pdf",
  "maxPlayers": 100
}

The example report is historical, not today's injury list. Supply the report URL you want processed. Set a finite Max total charge before running.

If your PDF has no public URL, supply its standard padded base64 bytes through pdfBase64 instead. Do not include a data-URL prefix, whitespace or pdfUrl too. maxPlayers is optional and defaults to 1,000. Demo mode accepts only demo:true.

Results

Each injury_report_entry contains the source's player name (surname first), team, matchup, game date, participation status, reason and report timestamp. sourcePages and sourceRows identify where the entry was extracted; the PDF's SHA-256 identifies its exact bytes. Names and reasons are not translated or reclassified. Multi-line reasons and continuations across pages are preserved.

team_report_status means the source explicitly said NOT YET SUBMITTED. It is not evidence that a team has no injuries. These rows are unbilled.

parse_diagnostic explains an input, download, parsing or budget problem. If any extraction ambiguity remains, all player rows from that PDF are withheld. The run fails with a diagnostic; partial source text is not delivered as valid data.

The Dataset contains separate player and diagnostic rows, not a single nested report. Filter on recordType and isDemo before adding rows to a production feed.

Charging and limits

The unit is one delivered real player row, using the price shown in Pricing. Demo rows, team submission statuses and diagnostics do not request result charges. A platform start charge may still apply.

The Actor checks the remaining budget before downloading and before every paid row. It stops at the player or spending limit and identifies remaining rows. If a write or charge acknowledgement is lost, it stops rather than retrying. In a failed run, a row or charge may already exist: inspect the Dataset and run charges before starting again. Automatic resurrection is not supported; a new run is independent and may charge for the same document again.

Supported input is one digital PDF, up to 10 MiB, 20 pages and 1,000 physical table rows, in the NBA seven-column injury report layout. Minimum run memory is 1,024 MB, covering the Node process and the separately limited Python worker. Source download has a 20-second overall deadline; parsing runs under a separate resource limit. Public HTTPS on port 443 is supported, with up to three redirects. Private network addresses, cookie/login flows and compressed HTTP response bodies are rejected. The recorded URL omits any query parameters; use the content hash for provenance.

Scanned PDFs, OCR, arbitrary table layouts and archive crawling are not included. The game time has an ET label but no AM/PM, so no UTC start time is guessed. The report timestamp's timezone is not specified in the supported source layout. Text-layer extraction does not certify visual authenticity or current player health.

Report a problem in the Actor's Issues tab with the run ID and a public sample when possible. Avoid sharing private documents or signed access URLs in public.