Skip to content

QuorumGrad v0.1.0

Latest

Choose a tag to compare

@omar07ibrahim omar07ibrahim released this 10 Aug 08:01

QuorumGrad v0.1.0 is the first portfolio-grade release of a deterministic fixed-point robust-aggregation receipt laboratory.

Highlights

  • bounded strict JSON contract for 3–31 clients and 1–16 coordinates;
  • coordinate-wise trimmed mean and median with explicit low / included / high witnesses;
  • exact reduced rational arithmetic with no floating-point aggregation or verification;
  • independent O(n²) pairwise-rank verifier with no shared analyzer, engine, or ledger implementation;
  • canonical round, coordinate, 18-entry ledger, and complete receipt digests;
  • zero third-party runtime dependencies;
  • offline aggregate, verify, inspect, and verified HTML report CLI;
  • owner-only 0600 receipt/report output files with exclusive creation;
  • 88 tests, 96.57% line coverage, strict mypy and Ruff;
  • clean-wheel validation on exact CPython 3.11.15 through 3.14.6;
  • real desktop/mobile/CLI screenshots, four SVG diagrams, and a three-frame GIF;
  • source-bound, digest-pinned, networkless Chromium evidence generation;
  • CodeQL security-extended, Dependabot, private vulnerability reporting, and protected main.

Reference result

The checked-in synthetic round contains 11 clients, 6 coordinates, and 66 fixed-point update cells. Declared trim 2 produces 42 included cells and 24 explicit exclusion witnesses. Independent pairwise-rank replay confirms every public receipt field.

Coordinate commitment:

de93deb1e874406cbd897ad8671c2c6cf401d1cb9afd5c2751a55e54552d779a

Receipt SHA-256:

245a270d423e8b20e9bdf24125fe67a0546772850d54303ad97f32d267170ddf

Install

python -m venv .venv
. .venv/bin/activate
python -m pip install .

Claim boundary

The fixture is synthetic. This release proves deterministic aggregation, exact arithmetic, and independent replay for the declared bounded round. It does not identify malicious clients, guarantee convergence, preserve privacy, provide secure aggregation, certify a federated-learning deployment, validate submitted updates, or defeat every poisoning strategy.