Week 1 · Kickoff Night

Same rules,
same $20,
fresh prompts.

Season 2 opens with a rebuilt prompt system that gives every model repo access, whichever way that model can reach it. ChatGPT (via the Codex CLI) and Claude read the local filesystem. Gemini reads the same repo through GitHub's raw URLs. All three review their own past picks, run open-web research, and cite every source. Picks land before kickoff and get graded against the ESPN box score.

See this week's matchups Read the methodology

Weeks

Regular season

Full schedule →

Week 1

This week's slate

Season snapshot

Where we stand

Model page →

Games Predicted

Recorded model responses

Bets Locked

Fresh prompt system live

Picks Won / Graded

0 / 0

Reasoning reviewed separately

Combined ROI

N/A

Original Claude bet prices missing

Feed

Methodology updates

All updates →

2026-09-09

Prompt system reset for Season 2

Three lanes now. ChatGPT (via Codex CLI) and Claude get local repo access. Gemini gets pointed at raw GitHub URLs. Every model does open-web research and cites every source. Fair-comparison baseline preserved.

Iteration

Rubric

Reason grades separate from outcome

A right pick on a vibe scores low. A wrong pick that named a real factor scores high.

Grading

Season 1

2025 baseline stays public

15 games, 86 bets, 11 corrections. The corrections are the point.