MIESC 5.4.x Release Notes¶

Current line: 5.4.3 (2026-05-16). Full per-version detail in ../CHANGELOG.md; this is the user-facing summary of the 5.4 line.

Highlights¶

miesc evaluate command group: corpus (precision/recall/F1/timing ground-truth eval), ablation (per-layer contribution), compare (statistical run comparison), info (dataset stats).
Reproducible experiment config (--config experiment.yaml), JSONL export for ML pipelines, per-layer timing, auto-generated experiment cards (version/platform/seed/ command), 80+ category-normalization aliases.
Research Guide (docs/guides/RESEARCH.md) and a copy-paste CI workflow for automated benchmark evaluation.

Expanded find→fix→verify documentation: before/after patch listings, Foundry exploit test generation, formal-verification (CHC) methodology, threats-to-validity, and compilation-rate analysis.

Cardano (Plutus / Aiken), Algorand (TEAL / PyTeal), and Stellar (Soroban) analyzers alongside EVM and Starknet/Cairo, routed by miesc analyze.

PyPI (pip install miesc) and multi-arch + full Docker images on GHCR.
CI: MyPy gating, coverage threshold, benchmark-regression checks, weekly dependency scanning, Sigstore release signing, SBOM.
Documentation: SUPPORT, GOVERNANCE, MAINTENANCE, ARCHITECTURE, COMMUNITY, mutation- testing and signed-commits guides.

SmartBugs-curated: 95.8% recall (Paper 1 reproducible profile; a v-next baseline raising this to 99.3% with deterministic detectors is pending PDF/freeze finalization).
EVMBench local high-severity extraction: 92.5% recall (4-provider ensemble).
11 real DeFi exploits: 81.8% recall (κ=0.77).

pip install --upgrade miesc          # 5.4.3
# or Docker:
docker pull ghcr.io/fboiero/miesc:5.4.3

Release-notes discipline for the 5.x line was restored in 2026-06; earlier per-release notes (v4.x) are in archive/.