Add a self-learning replay analyzer that builds a SQLite corpus and generates per-player coaching notes
-
Add a self-learning replay analyzer that builds a SQLite corpus and generates per-player coaching notesIntroduce a corpus database and learning pipeline that ingests processed replay analyses, computes percentile baselines per metric/team size, and emits per-player critique notes that show up alongside replay details in the UI.
-
Add a web 'Pro Replay Analyzer' SPA to upload, browse and review analyzed replaysShip a new React-based web UI (Upload, Replays list, Replay detail, Baselines pages) that lets users drop .replay/.zip files, follow parsing progress, view per-player critique cards, radar comparisons and baseline visualisations.
-
Add API endpoints to upload, batch-upload, list, view and delete replays and to read/recompute baselinesExpose endpoints for single replay upload, batch upload (accepts .replay and .zip), listing ingested replays, fetching full replay+performances+notes, deleting a replay (which rebuilds baselines), and fetching or forcing recompute of baselines.
-
Add batch upload with .zip support and streamed NDJSON progress eventsSupport uploading multiple files or a .zip in one request and stream structured NDJSON progress events (parsing, zip expansion, ingest/duplicate/failure events) so the front-end shows live per-file status.
-
Show a 'still learning' message and suppress unreliable notes when baselines are under-sampledCritique generation now checks sample counts and will suppress automated percentile notes for buckets with too few samples, showing a "still learning" notice in the UI until enough replays are ingested.
-
Improve parser error messages to include rrrocket stderr/stdout and file sizeWhen rrrocket fails during parse the server now returns a clearer error that includes rrrocket stderr (or stdout), exit info and the replay file size to make failures easier to diagnose from the UI or API responses.
4 internal changes in this release