[ submind.us · v0.2 · archived 2026-05-12 ]
SubMind: vanguard predictive engine
Two honest tools, shipping shortly.
The forecasting engine was the wrong product. The machinery under it was right. v0.2 has been archived. Two successor projects are being extracted from the salvage.
NEMESIS Leaderboard submind.us
Anti-leakage benchmark for LLM forecasters. 100 historical events at exact 0.500 base rate. No-info Brier floor is 0.250 — skill becomes unambiguous. Submit a model, get graded.
HonestCI github → pypi
Your LLM features need CI too. Drop-in regression detection: prompt drift, citation integrity, narrative-vs-status consistency, calibration shift, file tamper, adversarial counterfactuals.
v0.2 → archive
├── 35,420 LoC reviewed
├── 6 components salvaged
├── 20-check audit pipeline → honestci
└── NEMESIS-100 holdout → leaderboard