AMD Developer Hackathon ACT II / Track 1

Accuracy first. Remote tokens only when needed.

ProofRoute combines embedded FunctionGemma assessment, proof-carrying deterministic solvers, selective Gemma E2B inference and evaluator-authorized Fireworks fallback.

Final scorecard Pareto calibration Fuzz report Neural audit Fireworks arena Release decision README SUBMISSION

Why this wins

The hard part is not calling the strongest model. The hard part is knowing when not to call it.

4 guarded stages

Live score

Our last scored evaluator baseline completed successfully and became the immutable control for every later change.

84.2% / 4,198 tokens

Submission shape

Public, immutable, no-download image with audited OCI labels and a compact rollback tag.

2.94 GB

Pull the submitted image

docker pull --platform linux/amd64 \
  ghcr.io/rvbernucci/track1-token-router:v3.9.0-dual-functiongemma

This public tag exposes a linux/amd64 manifest and passed the complete release workflow.

Expected route

solver_arithmetic, because code can answer this safely.

Expected answer

42, with final validation preserving the strict numeric format.

Remote spend

0 Fireworks tokens. The remote model is held for tasks where risk justifies it.

Run the submitted container

Create the official input file, then pass the three evaluator variables from your shell.

mkdir -p input output
printf '%s\n' '[{"task_id":"t1","prompt":"What is 6 * 7? Return only the number."}]' > input/tasks.json

docker run --rm --platform linux/amd64 \
  --memory=4g --cpus=2 \
  -e FIREWORKS_API_KEY \
  -e FIREWORKS_BASE_URL \
  -e ALLOWED_MODELS \
  -v "$PWD/input:/input:ro" \
  -v "$PWD/output:/output" \
  ghcr.io/rvbernucci/track1-token-router:v3.9.0-dual-functiongemma

cat output/results.json

Calibration status

The runtime remains fail-closed. The generic E2B answer contract improved accuracy from 62.50% to 84.38% on a balanced 64-task ablation. We then evaluated 4,400 local answers: mechanical evidence handled 2,621 rows, while GLM-5.2 and Codex/Antigravity agreed on 96.7% of the 1,779 semantic cases. The resulting neural challenger reached 95.95% protected precision at 8.41% coverage, but remains research evidence until it proves end-to-end dominance over the shipped gate.

Answer Contract Engine

Models receive only the raw prompt, never the evaluator's JSON wrapper. ProofRoute validates the returned content and reconstructs results.json deterministically. A controlled ablation reduced Fireworks input tokens by 51.9% while preserving byte-identical Kimi answers.

Run source checks

These commands validate the repository; they are separate from running the submitted image.

python3 -m pip install -e .
scripts/offline_release_check.sh
python3 scripts/export_public_report.py --check

Verified: 758 automated tests passed, with one environment-dependent skip.