AGENTS.md — Project Context for AI Agents

Architecture

src/benchmark.py — Benchmark suite with honest data-source labeling (published-spec, measured, mathematical)
src/context_engine.py — Perseus context resolution demo (3-session progression)
demo/demo_terminal.html — Playwright terminal simulation for demo video
demo/record_video.py — Record demo video via Playwright + FFmpeg
assets/architecture.html — Dark-themed SVG architecture diagram

Published-spec benchmarks — AMD cloud credits haven't arrived. All benchmarks labeled with data source. No fabricated measurements.
Unicorn Track — No fixed benchmark. Judged on creativity, originality, product potential.
Fireworks AI credits may suffice — Track 3 requires "one or both of" AMD Compute pods and Fireworks AI API credits.
Label everything honestly — published-spec, measured, mathematical projection — judges reward verifiable data over claims.

Benchmarks must include data_source and methodology fields
All MI300X claims must reference published AMD documentation
Demo terminal uses ASCII only (no Unicode — headless Chromium can't render emoji)
Architecture diagram uses double-rect technique (opaque backdrop + semi-transparent fill)

Push to Perseus-Computing-LLC/perseus-amd-agent (main branch)
Git identity for external commits: perseus 51974392+tcconnally@users.noreply.github.com