5. A release summary across frames

A release is judged on the benchmark frames, each chosen for one way a calibration can go wrong.

disp2vlm-validate gate

#	Category	Frame	Benchmark condition
1	Coastal subsidence + islands	F08882 Houston	subsiding stations
2	Stable inland	F08886 Oklahoma City	stable stations: |bias| < 1 mm/yr
3	High relief + islands	F16940 Los Angeles	sill drops; tropo mode does not make it worse
4	Northeast / GIA	F08622 New York	vertical pattern vs GNSS
5	Fast basin	F36542 Central Valley	continuous surface; stations outside unaffected; bowl kept
6	Coseismic event	F16941 Ridgecrest	no coseismic step leaks into the calibration
7	Sparse GNSS	F07081 Montana	realistic σ, no artifacts
8	Non-NA plate	F23211 Hawaii, F35991 Puerto Rico	plate; Kīlauea as a deforming area

Run every frame, then summarise

Run each frame as in tutorial 3, with its own gate.yaml for the benchmark condition and the released algorithm as baseline. Then:

disp2vlm-validate summary validation/*/v05 --out release_summary.html --title "cal-disp v0.5"

F08882   v05                      PASS
F08886   v05                      PASS
...
wrote release_summary.html

release_summary.html has one row per frame (verdict, DD sill against raw and baseline with intervals, velocity statistics) and the requirements traceability across frames.

Reading a release verdict

• A frame passes only if every gate rule and its benchmark condition hold.
• Notes are not failures, but they belong in the VnV report: a stable median bias above 1 mm/yr, or a missed reported target.
• When a frame fails for a known, local reason, the release can still go out with a documented waiver. That is a decision for the CalVal review, not for the tool.

The v0.5 release went through exactly this: see the v0.5 CalVal report and the baselines.