Files

4.1 KiB
Raw Permalink Blame History

Ordered continuation plan and acceptance criteria

First: reconstruct the evidence environment

Obtain authorized first-party source and private inputs independently. Verify the exact stock/image/module/source hashes and source schema against the catalogs. Read the current technical chapter plus relevant historical ledger entries, including failed attempts. Check for newer live project work after the packaging snapshot. Do not reinstall diagnostic10, switch profiles or start a radio merely to initialize this handoff.

Optical blocker: one changed discriminator

The last saved393-v2 mode check is complete; do not repeat it. The last passive stereo probe also completed and failed to obtain four consistent points at37.233us. The prepared profile3 probe is a different, justified candidate, but it is not authorized or executed by this handoff.

Before asking for readiness: verify source/config/dependency and argv seals, original native settings hash, actual executable identity, HMD/controller watcher permission, exclusive idle radio owner and literal true MCU restoration fields. Prepare native camera process ownership, settings backup, power/profile restore and interrupted-run cleanup. Define a30sbound, camera-only scope and zero pose publication.

Only after fresh direct human readiness: prove actual upper exposure750–850us under the intended profile, actual packet timestamps/current-boot mapping and live bracketing head poses. Stop if exposure/preconditions are wrong. Collect the full denominator, detected-marker counts, stereo fits/rejections, source ages and cleanup. If marker geometry appears, separately prepare causal identity and freshness validation; do not label unassigned passive fits tracked poses.

If unchanged failures recur without a source-supported correction, stop that experiment and analyze the retained data. Do not lower thresholds, widen age/covariance limits or phase-sweep to manufacture success.

One-tracker integration

Verify actual native mode1/N/pending state with fresh nonce-bound replies; preserve finite lease and ACK/query ordering. Integrate only parity-tested geometry changes into the exact selected pipeline, preserve ambiguity and identity gates, and measure total latency rather than helper-only compute. Resolve detection/profile mismatch explicitly.

For native25 episodes keep independent command tags, actual publication/ACK/application/physical-dark evidence, bounded unresolved OFF state and observed ON restoration. Validate actual current lit/dark frames on the same link/boot. Keep48-of61 and the two-dark/six-lit rules.

Receive actual IMU and demonstrate enough continuous/appropriately sampled excitation for axes, bias and physical time offset. Reject weak/degenerate data and missing-counter interpolation. Confirm held-out residuals and independent gravity directions before enabling fusion/occlusion behavior.

Then prove actual fresh world poses through PTP2/OpenVR with stationary jitter, known physical movements, gaps, latency and invalidation. A displayed model or connected state is insufficient. Capture the actual player/controller path; code compilation and mathematical replay are supporting evidence only.

Three trackers and body roles

Only after one clearly stable tracker: identify and calibrate the other two independently, review any firmware rollout as a separate explicit operation, finish deferred multi-peer RAM/DMA/mailbox ownership/lifecycle admission, validate mode leases and source attribution per peer, map waist/left/right roles and verify simultaneous controller coexistence. Do not transplant a first-tracker result into the other units.

Documentation on every iteration

Record trigger, exact source/image/module/config/argv/hash, Hardware/Static/Recorded/Inferred class, physical conditions, input data provenance, denominator, failure/negative result, literal cleanup barriers, remaining ambiguity and the next specific observable. Append a dated English research entry and update the exact corrected claim in current state/supersession. Keep primary material private; publish only rights-cleared explanatory text and version-bound diffs.