Repository navigation
Keep ML feature values paired with their nearest timestamps - #87
Conversation
|
Integrated as The earlier quiet Python run remains an unaccepted timeout. Diagnostic pytest completed in296.94s under the unchanged300s deadline; PR71 retains suite-budget ownership. Source integration does not establish physical-camera or installed-estate adoption. Actions-hold readback: Jacob's hold at HAMON143 comment6067592767 was posted at19:36:07 UTC, before our20:10:08 merge. This team discovered and read that hold only during the subsequent current-ownership recovery. The merge had already started two push runs,37837421592 and37837418703; fresh readback shows both completed successfully, with neither queued nor running. No cancellation occurred, and this does not undo completed execution or its potential charges. Future Actions-triggering writes are held; native work continues. Native final custody is |
Correct nearest-feature values
An ML bundle currently passes retained feature row order to
numpy.searchsorted. Real radar extraction can preserve embedded session timestamps in an order different from MCAP log order. In the retained MCAP → radar extractor → native feature writer → public Job API witness, the last window selects energy 4 instead of the nearest sample's energy 9.The loader now checks for a timestamp inversion and stably applies one permutation to both integer timestamps and feature values. The runtime change is four added lines. Ordered data, duplicate relative order and the existing later-time equal-distance choice retain their behavior. Stored inputs, feature writers, label aggregation, validity, center generation, metadata, schema and the unrelated internal self-digest are unchanged.
Closes #86.
Source and current ownership
Received composition
33c6fb3409970941521bd37d236ddda4d3bed581, tree0426265ad9d834f205bcece47eb6d69c94672905, is the genuine composition of qualified contributiona70e2683and current maina2c89957. Native equivalentb733bcdbhas the identical tree. All 1,157 unrelated main leaves remain exact, including the newly landed source picker and saved-job parameter comparison. The complete current progress document remains an unchanged suffix after the owned addition. All 96 Python library leaves were identical between the received parent and current main; no additional runtime replay was needed for these desktop/evidence changes.Reviewed producer SHA-256
ed67d94ded834d63599797bbd8efb7b26f125bd216f52bcb7f01febb928922ce; paired receiver7a5e226afb9c0f49eadf6d4c25b8bc22575f1fdd04afdac854cf8aa4adee58be.Qualification
chatgpt:/root/mac_executionapproved the exact four-line patch, full producer, all 17 receivers and documentation. It independently restored the original bytes and verified all 43 input pairs and 17 output identities, without replay.The initial receiver's four 2 ns windows were below the schema minimum and are retained as a separate receiver defect. Corrected final tests use valid durations and include exact adjacent nanoseconds above 2^53. Original failures and both final source maps remain intact.
Evidence and limits include a verified 265,323-byte capsule with 296 regular members, exact source/test/runtime/dependency provenance, complete paired inputs/outputs, raw process/JUnit/logs, independent review and current-main preservation receipt. Earlier native results retain their original source bindings. The complete published composition must pass the existing supported Windows Python and C++/native workflow before integration; no hardware, installed-desktop or scientific-validity claim is made.
Exact diagnostic successor
Current head
050f41170a656b6a6177770b3d21749ef5263b36, treea0d17ec8df857b295321fef89e04a831fe51e8d6, is a normal guarded successor of33c6fb34. Native equivalent66d657c2has the same tree. It changes only the existing CI wrapper’s-qargument to-vvand adds--durations=20, plus three uniquely scoped evidence files. The complete test selection, 300-second deadline, JUnit/process/source checks and acceptance rules remain exact. Product and maintained receiver hashes above are unchanged. Root independently reviewed the complete 272-byte diagnostic patch before publication.The original run 37829510696 is retained as an unaccepted Python gate: it reached the 300-second timeout with null process exit/counts/JUnit and no assertion failure shown. Its quiet progress is not a pass claim or proof of the active case. The separate C++ job passed 30 CTests and six native integration cases. The preceding current-main control passed 608 tests plus six existing missing-daemon skips in 227.34s on the same runner image/dependency versions; that control does not qualify this head.
The diagnostic packet retains the complete original job logs, API artifact metadata, receipts and exact original/candidate wrapper. Its 68,167-byte capsule contains 13 verified members, SHA-256
847566053aae5c9623fb3832b09fa18ad0ab3459e56eb319f2495db90f41b2ec. Artifact ZIPs were not retrieved; the prior CDN limitation remains respected. One fresh supported diagnostic gate is now required. No quiet rerun, deadline increase, deselection or relaxed acceptance was used.