eye 58 · model_key WF970 · bridge 17
Keyed by (brand, model_key, eye_size), not this frame’s numeric id -- a `resolve_all` recompute can recreate the frame row and null out a generation’s `frame_id` pointer, but its natural-key identity (and so its attribution here) survives. The same `geometry_digest` across two rows means the same shape was regenerated, not that anything actually changed.
| created_at (UTC) | n | stdev_max (mm) | geometry_digest | output | flags | captures |
|---|---|---|---|---|---|---|
| 2026-08-13 01:54:43 UTC | 2 | 0.5201 | 9d2b9735d73d | WF970 58-17.2.DAT | surplus-angles-truncated | 1 |
Solid line: mean shape. Shaded bands: +/-1sigma (dark) and +/-2sigma (light) from frame_stats.profile_mm. Outer ring: deviation heat map, calm-to-hot, bucketed at 3.60° -- reflects the frame's latest generation only.
Display registration only: each trace is independently re-centered on its own bounding box and left eyes are mirrored to right orientation for visual comparison -- this is not the pipeline's own registration.register() output and performs no rotation alignment or resampling.
As captured (L eyes unmirrored) -- unlike the ensemble overlay above, which always mirrors a left eye into right-eye orientation for comparison (F7).
Shape-level flags (has-excluded-captures, no-bridge, high-divergence-ensemble, ...) live on a generation, not on the frame itself -- shown below as the LATEST generation’s flags, not this frame’s own.
latest generation flags:surplus-angles-truncated
Exclusion decisions are CLI-only (human-only doctrine).
| id | capture_time | residual (mm) | flags | excluded |
|---|---|---|---|---|
| 11 | Apr 14, 2026, 09:19 PM UTC | 0.1694 | parse angle-count-inconsistent resolution fram-question-markmodel-from-filename alignmentnone | — |
Advisory only (DECISIONS #55) -- a score never changes whether a capture is averaged; that stays a human CLI decision. Reviews attach to a specific capture, not the frame itself -- open a capture above to add one.
No reviews yet.