Harness Analysis
| Candidate | Hermes | ChatGPT | Claude | Gemini | Perplexity | Board |
|---|---|---|---|---|---|---|
| Jared McCain Under 17.5 Points + Rebounds + Assists | — | T2 | T2 | T2 | T2 | Win |
| Victor Wembanyama Under 13.5 Rebounds | — | T3 | T3 | T3 | T3 | Win |
Final Decision
The completed board had only two NBA unders from the same Thunder-Spurs game, so the candidates were not independent diversification for a capped Martingale system. Harness consensus treated McCain as the least-bad Level 0 paper/control candidate and Wembanyama as T3/reject, but no option cleared recovery-grade standards.
Dan explicitly logged no real wager, leaving bankroll and Martingale state unchanged. The visible board swept 2-0: McCain won with 6 PRA and Wembanyama won with 8 rebounds. This should be treated as an opportunity-cost PASS / no-action learning signal, not a retroactive experiment bet.
Candidates
- Rithmm
- 64.2%
- BE
- 53.27%
- DTM
- 23.6%
- Recent
- L10 7/10
- Rithmm
- 68%
- BE
- 54.95%
- DTM
- 29.1%
- Recent
- L10 3/10
Harness Notes
Hermes kept the official action as PASS / SKIP because the two-candidate board was thin, same-game correlated, and not recovery-grade. Dan explicitly logged no real wager, so there was no bankroll or Martingale damage; the paper McCain lean won with 6 PRA.
Correct paper/control lean. ChatGPT treated McCain as playable only as a Level 0 paper starter, not recovery-grade; McCain finished with 6 PRA against the 17.5 line.
Claude's SKIP avoided any bankroll loss and correctly rejected a real Martingale exposure, although the visible board swept 2-0 and the McCain paper lean won.
Gemini's SKIP protected the official experiment from a thin, same-game under cluster, but it was an opportunity-cost miss because both candidates ultimately won.
Perplexity's board-level SKIP avoided official exposure; its forced Level 0 paper ranking put McCain first, and that paper lean won with 6 PRA. Wembanyama was still rejected for system fit despite winning with 8 rebounds.