Eval report — step-002200

• entries evaluated: 4535
• full-parse exact match: 0.0818
• mean token confidence: 0.8063

Per-component F1

tag	precision	recall	f1	support
country	0.2143	0.2082	0.2112	245
region	0.3430	0.1298	0.1883	3205
locality	0.2478	0.3053	0.2736	3357
dependent_locality	0.0050	0.1000	0.0096	40
postcode	0.8324	0.5916	0.6916	2980
subregion	0.0000	0.0000	0.0000	0
cedex	0.0000	0.0000	0.0000	1
venue	0.3765	0.4015	0.3886	1101
street	0.3559	0.2616	0.3016	2928
house_number	0.7446	0.8335	0.7866	1742

Calibration (confidence bucket → accuracy)

bucket	n	accuracy
0.0–0.1	0	0.0000
0.1–0.2	21	0.2381
0.2–0.3	666	0.2913
0.3–0.4	2416	0.3266
0.4–0.5	4308	0.3152
0.5–0.6	4777	0.3383
0.6–0.7	4943	0.3569
0.7–0.8	5534	0.3825
0.8–0.9	8066	0.4363
0.9–1.0	30517	0.6440