lenitnes / README.md
Papajams's picture
Instrument-first positioning: same loop, two verticals
e63b90a verified
|
Raw
History Blame Contribute Delete
1.8 kB
metadata
title: LENITNES  Scientific Software Integrity Sentinel
emoji: 🔬
colorFrom: blue
colorTo: green
sdk: static
pinned: false
license: mit
short_description: Live mirror of the LENITNES integrity sentinel dashboard

🔬 LENITNES — Scientific Software Integrity Sentinel

A pre-registration machine for software-change judgments: LENITNES watches repositories, detects commits that may invalidate published results, scores each one with a versioned rubric, and notarizes every verdict on Hedera HCS before the outcome is knowable — then grades itself in public, losses included.

Two live verticals run the same loop:

  • [code] — crypto repos scored against market price (Season 1, closed)
  • [bio] — scientific software scored against the published record (retractions, corrections)

This Space embeds the live production dashboard. If the frame does not load, open it directly: https://lenitnes.persidian.com

How it works

  1. DETECT — monitors watch GitHub repos for meaningful commits
  2. SCORE — LLM rubric (v6 [bio] / v5 [code]) scores severity and impact
  3. COMMIT — every verdict is notarized on Hedera HCS before the outcome is known
  4. GRADE — outcomes are validated against external ground truth (market price for [code], dated retractions/corrections for [bio])

Sponsor integrations

  • GXL / Paperclip — literature grounding via MCP
  • Anthropic — Claude as the primary scoring LLM
  • Hedera HCS — immutable notarization of every verdict
  • Modal — GPU compute escalation for expensive claims

Built for re:AGENT · Founders Inc SF · Aug 15–16 2026 · Track A: Co-Scientist · solo build