README.md

metadata

title: Claim-level Reproduction - Latent Agentic Substructures
emoji: 🎯
colorFrom: yellow
colorTo: red
sdk: static
pinned: false
tags:
  - trackio
  - trackio-logbook
  - open-experiment
  - icml2026-repro
  - paper-sW8U2TYDMp

Current claim-level evidence

This revision preserves the exact judged 7/12 artifact while adding visible, reproducible evidence for all six claims. Current results are five VERIFIED claims and one FALSIFIED broad paraphrase (Claim 4). This is an evidence forecast only: the live score remains 7/12 until the judge evaluates this revision.

Start at the canonical logbook, then use the visibility matrix, the six current claim pages, and the illustrated report. The current executable code is under reproduction/; raw JSON and controls are under evidence/.

All 17 judged paths remain at their original locations. Exact judged text/SVG copies are also preserved beneath historical/judged-3d065680/; the three binary PNG assets remain hash-identical at their original root paths because the release is intentionally text-only. The old combined verification is labeled Historical rejected baseline and is not the current verifier.

Historical Space description

Repro - Agents, Epistemic Utility, and Unanimity

An open experiment logbook, published with Trackio.