# Direct human-template quota feasibility — diagnostic only

| Class | Existing | Gallery / pose templates | Compatible targets | Added coverage from pose | Accepted | Selected novel | Attainable | Gap |
|---|---:|---:|---:|---:|---:|---:|---:|---:|
| 1 | 48 | 27 / 5 | 155 | 0 | 0 | 0 | 48 | 2 |
| 2 | 5 | 5 / 2 | 229 | 0 | 0 | 0 | 5 | 45 |
| 9 | 38 | 20 / 8 | 128 | 14 | 0 | 0 | 38 | 12 |
| 10 | 10 | 8 / 5 | 194 | 158 | 0 | 0 | 10 | 40 |
| 15 | 39 | 37 / 29 | 309 | 0 | 0 | 0 | 39 | 11 |

Snapshot: 97 saved gallery edits; 52 original pose-human templates. Actual gate-passing candidates and complete canonical provenance are in `accepted_candidates.jsonl`; quota-selected novel frames are in `selected_novel_matches.jsonl`.

Compatibility coverage is an upper bound, not evidence of passing image/vertex/beam gates. Group/split/policy/view details are in `coverage_by_group_split_policy_view.jsonl` and `template_groups.jsonl`.

All eligible targets were scanned. Concurrent template/eligibility changes: False. Results are diagnostic drafts; no publication verification, GT, training eligibility, or production writes.

## Exhaustive scan conclusion

The fixed current human-template union cannot reach 50 in any requested class under the existing gates. There were 4,214 eligible targets, 1,015 compatible targets, 12,415 registration attempts, and zero accepted novel matches. The cumulative count remains 140, with a shortfall of 110. Both candidate JSONL files are therefore empty; canonical provenance for all 149 direct human templates is in `human_templates.jsonl`.

The live pose store has exactly 52 current polygons with more than four vertices; all 52 qualify through `annotation_inventory`, and their target keys are entirely disjoint from the 97 gallery saves. Of these pose templates, 49 belong to the five requested classes; the other three belong to classes 5, 12, and 14. Both live stores retained the same event counts at scan completion.

## Additional compatibility from original pose templates

- Class 9: 14 additional train targets, all with unknown view. Eight use policy `2023_7_3|(264, 352)|primary`, in groups `2023_7_3/scene/300mm` (6) and `2023_7_3/scene/300mm/2000mm` (2). Six use `2023_7_4|(264, 352)|primary`, across five `185mm`/`300mm` groups.
- Class 10: 158 additional train targets under `2023_9_26|(264, 352)|primary`, across 14 collection groups in `hollow`, `hollow_yellow`, and `yellow` at 405mm. Views: unknown 41, oblique candidate 20, narrow-face candidate 97.
- Classes 1, 2, 15: additional templates offer new polygon variants but no additional compatible target identities.

Union-compatible targets by split (train / val / test): class 1 = 138 / 0 / 17; class 2 = 229 / 0 / 0; class 9 = 14 / 114 / 0; class 10 = 159 / 35 / 0; class 15 = 309 / 0 / 0.

Attempt rejection totals: unreliable vertex tracking 11,482; excessive affine deformation 900; vertex refinement too large 27; insufficient image correspondence 3; local beam boundary review required 3. These are attempt counts, not distinct target counts.
