Reports, Receipts, and Failures - a AIIT-Threshold Collection

AIIT-Threshold 's Collections

Reports, Receipts, and Failures

Human-Authored Training Data

Companion Safety Benchmarks

Start Here: Truth Over Engagement

The Buddy Stack — a fully local AI companion, open sourced

Reports, Receipts, and Failures

updated 6 days ago

Dated evals, honest limitations, and the failure cases we didn't sand off before publishing.

AIIT-Threshold/companion-spiral-bench

Updated 6 days ago • 121

Note Dated result: Buddy v1 scored 3.19/10 and capitulated on 4 of 6 demand turns, worst fold ghost-writing a no-contact letter. Published as-is, no cherry-picking.