Skip to content

Bench foundation: bounded NODE recording, IMU validity and review handoff - #10

Merged
overlordxrz-source merged 9 commits into
mainfrom
review/bench-foundation-20261024
Oct 25, 2026
Merged

overlordxrz-source merged 9 commits into
mainfrom
review/bench-foundation-20261024

Conversation

@overlordxrz-source

@overlordxrz-source overlordxrz-source commented Oct 24, 2026 •

Copy link
Copy Markdown
Owner

Adds the focused IMU quaternion boundary and bounded NODE raw-recording → existing Observation JSONL → WORLD replay pipeline. No UI application, team/GNSS/radar implementation, CI workflow, firmware or frozen Alpha evidence changes are included.

Independent review remediation — ready for re-review

OpenCode requested changes on 2b54410508118ae4e2c75d0f162467ada98dba28, recorded at 760a3f6e6c98b46578f53cce712c06b97266b003. The review was imported verbatim. New head: 8a84369941c74d5f884c9ec6cc33d10395223597. Regression-first commit: f99db9a10448cbd86e7af4e834885133af0d39a8; implementation: 57c8be07a5e728808c998224f1712cf79b459dbb.

All five requested fixes:

  • Checked raw/receipt write results and persisted byte/count/end/hash reconciliation; short or inconsistent evidence cannot end COMPLETE or exit 0. Equal-size corruption is also caught.
  • Direct KeyboardInterrupt/BaseException publishes terminal CANCELLED/FAILED, closes resources and re-raises the original exception. Stop requests during finalization cannot succeed.
  • Reversed receipt completion/stop clocks are rejected and persisted receipts are independently checked.
  • Nonrepresentable quaternion integers raise ValueError without consuming driver state.
  • Explicit acquisition completion, clean-first-run quality result and scientific acceptance. Gaps, malformed/recovered/spliced/unsupported records and clock discontinuities fail the clean gate.

Compatibility: a fully preserved but dirty capture can retain COMPLETE acquisition artifacts while clean_first_run=FAIL; CLI now exits 2 for that case. Scientific acceptance remains NOT_EVALUATED. Status fields are additive; Observation v1 and JSONL/manifest interfaces remain unchanged. Diagnostic artifacts may survive late cancellation; file existence is not an acceptance signal.

Validation

  • Initial 18 reconstructed regressions all failed on the reviewed code; two supplemental edge controls separately failed before their fixes.
  • 59 focused passed; 304 offline passed, 1 hardware deselected; architecture PASS. CI on exact final head passed.
  • All five targeted disposable mutations detected with expected test failures. Existing adversarial/ownership controls retained.
  • Actual source: 2,788,217 identical bytes / 681 reads / 7,290 observations, zero rejected/unaccounted bytes. Expected/persisted raw and receipt hashes match; raw SHA-256 f0323be3d79af1bd5f29fd21b483323c90eeb7be3f90c8e89c980e92d8432b2c unchanged.
  • Unchanged WORLD v2 4e1c920 consumes NODE output; first/middle/last full envelopes and browser jumps verified. No physical serial port opened.

Review entry points and physical gate

Per-defect remediation, commands, mutations and compatibility · canonical branch handoff · prepared ESP32 protocol.

Noncooperating serial consumers may bypass ownership requests; cooperative exclusion is tested but physical USB/alias behavior and already-open readers are unverified. Operator/device evidence and specific owner authorization remain mandatory. EXP-NODE-ESP32-001 remains NOT AUTHORIZED / NOT RUN. No merge, flash, reset, physical acquisition, human trial or Alpha evidence change. PR #7–#9 required follow-ups remain separate. Independent OpenCode re-review is still required.

@overlordxrz-source
overlordxrz-source merged commit d3add95 into main Oct 25, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant