SensorProof v0.1.0

SensorProof v0.1.0

A deterministic, fault-aware sensor-fusion reference system whose decisions can be independently replayed and verified.

Highlights

• fixed-point [x, y, vx, vy] estimation with exact half-away-from-zero arithmetic;
• SHA-256-derived deterministic sensor noise;
• innovation gating, two-strike fault detection, and bounded quarantine;
• identical-stream ungated baseline for an honest comparison;
• strict and bounded JSON contracts;
• independent transition-by-transition replay verifier;
• run, verify, inspect, and report CLI commands;
• self-contained offline HTML report;
• real, reproducible CLI, browser, diagram, plot, and GIF evidence.

Reproduced result

On the checked-in abrupt synthetic GNSS bias-step scenario:

• robust RMSE: 160 mm;
• ungated baseline RMSE: 8,792 mm;
• RMSE reduction: 98.18%;
• fault observations rejected or quarantined: 38 / 38;
• healthy observations rejected: 2 / 250.

Verification

• 37 tests passed, 94% statement coverage;
• strict mypy, Ruff lint, and Ruff format pass;
• clean wheel build and installation pass;
• CI passes on CPython 3.11.15, 3.12.13, 3.13.14, and 3.14.6;
• all 13 checked-in evidence files regenerate byte-for-byte in the pinned, network-disabled browser container;
• CodeQL analyzed Python and GitHub Actions with 0 open alerts;
• Dependabot and secret-scanning alert queues are empty at release time.

The immutable evidence manifest records source revisions, commands, dimensions, and SHA-256 digests.

Scope boundary

This release demonstrates recovery from an abrupt synthetic GNSS bias step. It does not claim coverage of slow-ramp or correlated faults, production covariance modelling, real-vehicle validation, or safety certification.