Evidence before appearance.

Reviewed exact GPS narrows the search but is never the final camera. Street-level references, local feature matching, provider depth, and robust PnP recover a metric pose. When position evidence is stronger than the absolute solve, Sightline may hold corrected GPS fixed and solve orientation only.

Mean-sea-level altitude is corrected to the WGS84 ellipsoid before comparison. Device direction and gravity may guide the search, but verification decides the result.

The sf-north-v1 experiment receives metadata-stripped pixels, an opaque identifier, and fixed geographic bounds. It cannot read GPS, heading, focal length, device orientation, timestamp, source filename, or private metadata.

Predictions are sealed before a separate evaluator may compare them with embedded device metadata. Those differences are diagnostic, not survey truth.

1. Render the solved camera at the source aspect ratio.
2. Match the source photograph against that exact render.
3. Measure coverage, residual structure, depth consistency, and displacement.
4. Reject broad disagreement or insufficient evidence.
5. Re-render any repaired pose and apply the same tests again.

A verified camera does not automatically authorize projective texture. Exact location, comparison tools, and projection remain separate editorial decisions. Rounded, city-level, and hidden records cannot leak exact camera geometry.

Provider imagery, identifiers, rich EXIF, credentials, local paths, and detailed matcher evidence remain outside the public artifact.

Sightline is not survey-grade ground truth. GPS uncertainty, scene changes, simplified lens assumptions, panorama pose, provider photogrammetry, occlusion, and incomplete depth all affect the result. Confidence ranks evidence; it never replaces the exact-render audit.