Skip to content

Offline edge preparation evidence

complete-evidence.tar.gz contains 870 indexed files, including both complete 401-module source trees, frozen tools and result schemas, acquisition receipts, all twelve full calibration results, measurements and logs, and packaging source. The archive is 3,471,025 bytes. It contains no raw image or scan files.

inventory.json pins the archive and index bytes. summary.json is the original formal-cohort summary, with these median ratios:

  • Frame preparation: 3.880811x.
  • Complete estimation API, including all preparation, fits and controls: 2.594888x.
  • Complete fresh measurement worker, additionally including decoding and startup: 1.509598x.

The formal source is 73b155344a85b1caa6e659639be8e6bdf2de6abd. All six complete results and prepared inputs agree. The predeclared 2x estimation gate passes. The initial cohort is retained separately, including its original CRLF-related Git audit failure and corrected post hoc validation. No observations are pooled.

Extract the tracked archive into a new directory and replay the maintained CPU-only auditor:

python -m tarfile -e docs/assets/offline_edge_preparation/complete-evidence.tar.gz outputs/offline-edge-replay
python tools/benchmark_offline_edge_preparation.py --audit outputs/offline-edge-replay/formal --verify-git
python tools/benchmark_offline_edge_preparation.py --audit outputs/offline-edge-replay/initial --verify-git

Git verification requires the recorded source commits locally. Omit --verify-git for schema, byte, import, timing and complete-result replay without Git history. Neither mode imports Calibrex or runs archived code. The timings are spent-data performance observations without calibration accuracy or SOTA claims. See the measurement scope and API.