[nicolasdiolez]

The checked-in staircase capture plan disabled EXIF intrinsics even though the measured pilot enabled them. Set the example to exif_intrinsics: true and record the fresh 16/31-capture comparison performed on merged dev (3bba6be, after #29).

• Add a dated report and structured measurements: all 96 common JPEGs and 16 common masks per consumer profile match by SHA-256. Postshot fails at 16 captures and trains 118/186 views to 5,000 steps at 31; the dense project was saved and reopened. RealityScan's largest component grows from 8 to 58 views, while both cases retain 12 components.
• Update the roadmap and changelog, and link the follow-up from the historical reports without changing their original measurements. Prioritize consumer-result import, mask association/persistence and controlled repeatability checks. Application defaults and diagnostic limits remain unchanged.

Validation: the existing historical-staircase capture-plan test passed (1 test); Ruff, scripts/check_release.py, JSON/local-link checks and git diff --check passed. The full test suite was not rerun for this documentation/example change.

Limits: one fresh run per arm; consumer counts vary despite identical input bytes and reported alignment settings. Five unmatched-mask warnings persist on Postshot reopen. This does not qualify final geometry, held-out splat quality, Windows extraction or a distributed binary. Media, native projects and private evidence are not included.