Pinned Loading

1. alibi alibi Public

   Check what an AI coding agent wrote about its own work against what it actually did. No model, no account, no network.

   Python 1

2. midflight-replay midflight-replay Public

   Forensic replay for AI coding agents — turn a session log into one self-contained HTML timeline. Zero dependencies, zero network.

   TypeScript 1

3. hea-near-duplicate-leakage-audit hea-near-duplicate-leakage-audit Public

   Reproducible audit: random train/test splits inflate ML accuracy on high-entropy-alloy phase prediction; ~76% of the accuracy drop is near-duplicate redundancy, not data volume.

   HTML 1

4. llm-inflection llm-inflection Public

   The LLM Inflection — reproducible study of 120,840 scientific abstracts (2015-2026). Vocabulary broke; style didn't. Bilingual site + papers + dashboard + 8-gate verify kit.

   HTML

5. agentreview agentreview Public

   Enterprise-grade, local-first safety reviewer for AI-agent git diffs. 10 rules, SARIF output, zero runtime deps.

   TypeScript

6. truthgate truthgate Public

   A deterministic verification gate that catches its own constant-true checks and reports its own false-positive rate. Zero dependencies.

   Python 1