| 2026-10-27 | Stacking-clause spotter (lab routine v1 and v2) against the keyword champion, on a dev set and a held-out set of public excerpts. | Recall and false-alarm rate. Bar: at least 90% recall and under 10% false alarms on both sets. | No winner. Neither routine beat the other and passed both sets. Not shipped. |
| 2026-10-27 | APR estimate methods an instruction can carry out, against a spreadsheet IRR on 30 worked offers. | Absolute error versus the reference schedule. Tolerance 0.5 percentage points as the floor. | No winner. Every method an instruction can carry out failed. Not shipped as an APR. |
| 2026-10-27 | Confession-of-judgment detector against the keyword champion, on 30 dev excerpts and 30 held-out excerpts. | Recall and precision. Output rules: disclaimer, ask a lawyer, no unsourced state rules, no links. | Provisional only. Dev recall 95% versus the keyword champion at 70%. Not shipped. A second human label pass was still required. |
| 2026-10-27 | Advance-fee offer screen (frozen v2) against a keyword baseline, on 30 held-out texts. | Held-out recall and false-alarm rate. Bar: recall over 90% and false alarms under 5%. | No winner. Held-out recall 86.7% (13/15) and false alarms 13.3% (2/15). Not shipped. |
| 2026-10-27 | Five-question grant screen against the tells list in the grant-fee guide, on 40 mixed items. | Accuracy, and zero scam items scored legitimate. Bar: at least 90% and no scam scored legitimate. | No winner. Both arms failed the bar. Not shipped. |
| 2026-10-30 | California AB 2116 licensing note against the California row on the state-disclosure guide. | How many of 6 owner questions the page can answer. Champion had 0 hits for 2116, license, and 2028. | Challenger, 6/6 versus 2/6. Not published. The live guide was left unchanged. |
| 2026-10-30 | 1West Small Business Health Index Q3 2026 against four skill-grade market-data criteria. | Pass count on primary-and-dated, whose pool it measures, whether it is a vendor release, and whether the method stayed comparable. | Champion. The index passed 1 of 4. Not added. |
| 2026-10-28 | Polymarket negative-risk market-making paper test against the trades gate. | IC proxy t at least 3 and Sharpe at least 1. | No winner. HOLD. IC proxy t -0.81, Sharpe -0.14. Not added. |
| 2026-10-30 | OpenEvolve against the SPY seed strategy. Lab only. | Sharpe. Seed Sharpe 0.789. | No winner. Seed not beaten in 12 iterations. Not added. |