Rework player award ratings, swap weak photos, and rewrite copy for a more human tone

Stop retrying a failed local LLM extractor on every scenario request

Fix flaky overdue-results test by pinning to the frozen snapshot fixture

Fix result-check automation: skip scoreless FINISHED matches, decouple overdue gate

Redesign Player Awards as EA FC-style player cards

Add Player Awards section (Golden Ball/Boot/Glove, Best Young Player)

Decouple result-check tests from live mutable snapshot

Finalize forecast tracking and result automation

Finalize forecast integrity and evaluation gates

Remove scalar tuning runtime dependency

Show Dixon-Coles top scorelines

Calibrate forecasts and add market blend gate

Tune recency and add ensemble evaluation

Publish hashed pre-kickoff forecast proof

Finalize public release and live forecast proof tooling

Fix forecast view toggle and performance scorecard

Complete forecasting release gates and official bracket

Fit and validate World Cup forecast model

Lighten the Space shell

Add test verifying inlined training prompts match runtime

Phase 0: submission-critical work

Reject zero-weight model hallucinations

Use fast small model and real World Cup Elo ratings

Recover honestly from missed scenario factors

Keep extraction within CPU Space limits

Use chat template for scenario extraction

Build live World Cup 2026 forecaster

Harden dataset evaluation and expose runtime fallback

Build Underdog Lab forecasting hackathon app