Current sparse frames do not support an independent metric temporal check

Frozen be471af7, audited2026-10-30. See protocol, plan and available calibration inventory, all12 image pairs, schema and adjacent-frame local availability.

Do not build a temporal depth filter on the current sparse triplets. Only1/12 adjacent archived image pairs meets the fixed image-only directional-support screen. Independent metric camera motion is absent from the declared local input set: OXTS provides IMU poses, and the available IMU-to-camera chain would consume the target camera/LiDAR extrinsic. Monocular essential-matrix translation has unknown scale. Neither is silently presented as independent metric camera motion.

The image intervals are3.728–11.590 seconds, much longer than adjacent capture frames. Same-index camera/OXTS timestamp differences reach12.238ms; these are availability/timing statistics, not validated interpolation or synchronization. All selected OXTS packets exist. No independent direct camera/IMU calibration or metric camera-trajectory artifact was configured in the audited directories; this is a local inventory finding, not a claim that no such evidence can exist.

Drive	Frame pair	Interval(s)	Cross-check matches	Essential inliers	Cheirality inliers	Image cells	Screen
2011_09_26_0005	frame-38→frame-76	3.926	583	32	24	[4, 6]	False
2011_09_26_0005	frame-76→frame-114	3.925	540	30	16	[8, 3]	False
2011_09_26_0009	frame-111→frame-223	11.590	555	26	15	[8, 7]	False
2011_09_26_0009	frame-223→frame-334	11.496	568	27	20	[7, 5]	False
2011_09_26_0011	frame-58→frame-116	6.007	605	33	19	[4, 6]	False
2011_09_26_0011	frame-116→frame-174	6.007	641	189	179	[10, 10]	True
2011_09_26_0013	frame-35→frame-71	3.728	516	27	12	[2, 4]	False
2011_09_26_0013	frame-71→frame-107	3.728	594	25	9	[2, 5]	False
2011_09_26_0015	frame-74→frame-148	7.666	518	48	38	[4, 6]	False
2011_09_26_0015	frame-148→frame-222	7.666	519	47	28	[3, 4]	False
2011_09_28_0002	frame-93→frame-187	9.746	1070	884	0	[0, 0]	False
2011_09_28_0002	frame-187→frame-281	9.746	1354	1267	0	[0, 0]	False

The screen requires30 cheirality-supported correspondences spanning6/16 cells in both images. Passing0011/frame116→174 is not a validated camera trajectory or recovered calibration. The0002 pairs have884/1267 essential inliers but zero cheirality support, illustrating that many image matches alone do not establish metric motion; this could reflect low parallax or degeneracy, not absence of shared scene content. Likewise failed ORB support does not prove geometric overlap is zero. These are fixed-algorithm readiness measures, not ground-truth overlap.

A file-existence inventory of ±1 capture neighbors around all18 selected frames finds0/36 complete local stereo pairs. No neighboring image contents were read or new data downloaded by this audit. The existing data therefore cannot be silently reinterpreted as short-baseline temporal coverage.

Three focused tests, repository Ruff and strict documentation build pass. Checks validate report/schema, all12 pairs and all frozen input/source hashes. Tests check blank-image zero support, camera-geometry validity and analytic parameter parity. No reference camera/LiDAR extrinsic enters ORB/motion estimation; available calibration files are inventoried, not composed into a selection transform. No extrinsic fits, temporal masks, ICP reruns or accuracy verdicts change.

Concrete next prerequisite

Before another estimator, freeze a bounded short-baseline acquisition/preparation plan for ±1 stereo neighbors within these same six already-spent recordings. Pin archive identity, exact member names/timestamps, byte limits and provenance before fetching; do not broaden to fresh recordings or test holdouts. Then audit image support again on those short intervals and independently validate the metric motion source. Stereo visual odometry may provide metric scale from camera baseline, but shares stereo errors with the measurements under test and cannot be called independent ground truth. Any temporal gate must disclose that coupling and pass controlled-motion tests first. Do not bridge the current limitation using target extrinsics, arbitrary motion scale, long-gap interpolation or selected good pairs.