Reliability here means producing medical-extraction responses that satisfy a defined JSON schema. A field-by-field validator distinguishes parseable JSON from responses that meet the full contract and records failure reasons. section 7.3.2
Establish a baseline before changing model weights: test schema compliance and general capabilities. With a strict prompt, the baseline reached 85% compliance on 40 samples; irrelevant inputs were a notable weakness at 58.3%. Treat these results as directional, not statistically robust. section 7.3.2
samples valid_json_pct schema_ok_pct category abbreviations 6 100.0 100.0 clean 8 100.0 100.0 implicit 7 100.0 100.0 irrelevant 12 58.3 58.3 typos 7 100.0 100.0
The category breakdown shows where baseline schema compliance failed. section 7.3.2
After specialization, QLoRA and QDoRA both reached 100% compliance with a minimal instruction, including on irrelevant inputs. This improved reliability while reducing the prompt needed, but came with a moderate decline in general benchmarks, most pronounced on ARC-Easy. section 7.3.5




















