head-prio-gold — examflow review prioritizer (Laya fine-tune)

Fine-tuned convaiinnovations/laya (Apache-2.0) for one job: prioritize review-queue items among review-first, review-normal, review-last — in a single encoder pass (~ms on GPU).

Training (all $0: Kaggle T4 x2)

• 2,000 gold construction-truth cases (incl. hidden-risk traps), 0 eval overlap
• Full fine-tune, 3 epochs, lr 2e-5, batch 8, bf16; option order shuffled per sample + 3 instruction variants (anti-prior-collapse)
• Held-out synthetic eval (n=150): 0.9800 vs heuristic 0.820 (+16pp)

Scope & limits (read before use)

• SYNTHETIC distribution: proves the loop, not real-world accuracy.
• Confidence is temp-uncalibrated until per-head refit (base checkpoint ships invalid temperatures — refit before trusting it).
• Never final-judge duty: dispatcher/signal layer only, abstain below tau.
• Safe format: model.safetensors (no pickle, no code execution on load).

Load

from laya import Agent
agent = Agent(model_id_or_path="ngdghfdc/head-prio-gold")
out = agent.predict(state, {"q": {"type": "choice",
    "instructions": "Pick the best action.",
    "criteria": {o: o for o in ["review-first", "review-normal", "review-last"]}}})