EvidenceTree Uses Anytime-Valid E-values to Keep Agents from Cheating Their Way to Discoveries
pierrelux · x · 2026-10-15
EvidenceTree targets the gap where agents are good at exploring but bad at falsifying and scientific rigor. It prevents systems from cheating or taking shortcuts to claim discoveries by carefully gating access to data and confirmatory evidence, and keeping an auditable evidence trail behind every claimed discovery.
The approach uses e-values, which are anytime-valid — well suited to agents that can inspect accumulating evidence to decide what to test next or when to stop, with evidence from successive experiments combined by simple multiplication. The team also built an MCP server that plugs into agentic harnesses (demoed with Codex) to handle the statistical bookkeeping via the e-value machinery.
The company's first paper will be presented at the NeurIPS workshop on e-values in Paris this December.
Related event: CarbonForge Uses E-values to Stop Agents from Faking Discoveries(3 posts)→
More from coding & agent
- Single HTML file with Three.js syncs all your MIDIs to AI video — techhalla · 2026-10-23
- Vercel ships 'vercel deploy --temporary' for ephemeral agent-built deployments — cramforce · 2026-10-23
- Running multiple agents out of one Notion workspace as a shared task board — nbaschez · 2026-10-23
- We tested 12 LLM classifiers — the new classifier hype is LARP — KlausCodes · 2026-10-23
- Dev builds ESP32 desk display showing Claude Code and Codex agent status via local Go helper — tonylook · 2026-10-23
- Let your agent read its own traces: AgentsView + history tool debugging workflow — dbreunig · 2026-10-23