mirror of
https://github.com/Rulez777/FramePicoTracker.git
synced 2026-10-17 15:01:16 +02:00
38 lines
4.8 KiB
Markdown
38 lines
4.8 KiB
Markdown
# Reproduce evidence on your own inputs
|
||
|
||
These are preparation recipes and acceptance criteria, not executable authorization. Obtain appropriate source/assets first, preserve exact source/config/argv hashes and use fresh human readiness for actual device actions. Do not replay a consumed original invocation.
|
||
|
||
## Version and patch audit, entirely offline
|
||
|
||
Acquire the exact stock container/application, verify hashes and matching segments, then verify every sparse diff forward and reverse in memory. Refuse different versions, output inside the handoff and source overwrite. Document image/candidate status independently of offline reproducibility. Build your own local decompilation only if authorized, using correct address mapping; compare exact code-region bytes before transferring a native conclusion to a modified image.
|
||
|
||
## Applied-mode RF check
|
||
|
||
Prepare an exclusive owner with the exact admitted original210 source/module and protected diagnostic10 provenance. Keep controller-off/HMD watchdogs, target/link/nonce/tag checks, final mode/N/pending reply admission and finite lease. Collect distinct request-matched nonces over the intended duration and actual four-barrier teardown. Do not rerun393-v2 unchanged: it already answered this question. A new check must address a changed implementation or unresolved observable.
|
||
|
||
## Camera timing only
|
||
|
||
Use a bounded30s owned camera session with no Roy/RF/LED/IMU/OTA/pose publication. Preserve decoded-frame-to-packet timestamp matching; record RAW/MONO/BOOTTIME and consumption time separately. Establish a current-boot mapping only when enough actual samples support a uniquely causal source domain. The prepared probe's historical criterion used at least40samples, causal fraction≥0.9 for the selected domain and<0.1 for alternatives. Report uncertainty and domains, not a single silently mixed age.
|
||
|
||
## Passive stereo and exposure discriminator
|
||
|
||
Use exact active local calibration, same-exposure bracketing HMD poses, strict250ms candidate age and actual pixel arrays. Record all attempted pairs, not only accepted samples. No LED commands means no causal optical identity; publication remains false even if geometry fits.
|
||
|
||
The unrun profile3 candidate temporarily changes native settings/power under an audited owner, checks actual upper exposure750–850us, records at most30s and restores original profile/power/settings. Refuse start if native settings/source/executable identity changed or cleanup prerequisites are unknown. If measured exposure is wrong, stop; if exposure is correct but markers/fits remain absent, inspect the retained changed data rather than sweeping phase or weakening detectors.
|
||
|
||
## Pixels, detection and geometry replay
|
||
|
||
Decode your own native recording with exact packet metadata, sample frames with a stated stride/denominator, count ordinary threshold180 blobs and separately labelled contrast-approximation blobs, and bind results to the original recording hash. Neither count is native contour or identity proof. For algorithm changes compare identical actual paired arrays and carried seeds in one common HEAD basis, preserve rejection labels/inliers/ambiguity/covariance, and report full-path timing separately from a helper-only timer. A simulated benchmark head/clock must remain labelled simulated.
|
||
|
||
## Causal LED identity
|
||
|
||
Only after applicable authorization and source review, prepare the actual bounded native OFF/ON or dark-pulse episode with independent tags, source publication timing, stable lit brackets, real decoded dark ROI frames and actual restoration. Missing/invalid frames never satisfy darkness. Retain unresolved OFF state on failure and do not clear it from transport ACK alone. Bind the identity to current tracker/link/boot and the admitted optical fit.
|
||
|
||
## IMU and calibration
|
||
|
||
Use actual admitted sensor producer/mode and sequential sample records. Determine counter cadence from actual timestamps/counters, then independently establish physical gyro-to-optical time offset, proper axes/scale/bias and held-out residuals. Require real sufficient stationary/gravity directions and dynamic excitation, and reject gaps under the existing policy. Do not manufacture continuous data by interpolating missing counters. Sparse alternative estimators are diagnostics until independently admitted; a fitted matrix with a negative determinant or weak excitation is a rejection, not an excuse to project a rotation.
|
||
|
||
## End-to-end tracking
|
||
|
||
Require actual current measured world pose, admitted identity/time/covariance and valid PTP2/OpenVR publication. Measure source-to-display latency, stationary jitter, slow/fast movement, gaps, invalidation and bounded occlusion. Then verify all three body roles with both controllers under their actual guard. Preserve a real player-view recording as evidence; old desktop videos, UI registration and test counts cannot substitute.
|