# SHiRE Academy Stage 3A.4 — Ollama Preload / Same-Retry Resume Fix

The controlled Boolean recovery reached Ollama successfully. Ollama returned HTTP 200, completed the request, and kept `shire-mini-fast:qwen3.5-4b` loaded, but the warm-up checker incorrectly required non-empty generated `response` text. Thinking-capable models may complete with an empty `response`, and model preloading does not require generated text.

This patch changes warm-up to Ollama's supported empty `/api/generate` preload request and treats HTTP 200 plus `done: true` as success. It permits the **same already-authorised qualification retry** to resume once because the qualifier was never started. It does not create a second retry, rerun CadQuery, consume an attempt, certify the skill, change Forge permission, or run during installation.