thalarch-teacherlisted

# Thalarch Teacher The Teacher Loop makes the model **learn from evaluated outcomes without pretending the base model weights changed**. It is a bounded artifact-feedback protocol: `student candidate → independent judgment → targeted revision → cold verification → experience`. ## 1. When to use Use when: - a meaningful implementation has multiple ways to be subtly wrong; - the user explicitly wants maximum-quality review; - a design/UI result needs independent acceptance judgment; - a prompt/skill/agent/router change is being evaluated for general usefulness; - repeated failures suggest the model needs a reusable correction pattern; - a stable benchmark/evaluator exists. Do not invoke for every trivial edit. Mechanical compiler fixes do not need classroom theatre. ## 2. Independence contract The judge receives: - user/repository acceptance criteria; - relevant final artifact/diff; - current evidence/results; - explicit scope and exclusions; - known required invariants. Do **not** give the judge the producer's persuasive chain-of-thought. The judge should reconstruct the acceptance test from the contract and artifact. When the host cannot provide an independent context/agent, perform a staged cold review and mark the lack of true context independence when material. ## 3. Hard gates before scores A candidate cannot pass if any required hard gate fails: - acceptance criterion violated; - scope/external-action boundary violated; - regression in must-preserve behavi