QISA arXiv package (paper-v0.2.4)

arXiv package for QISA Consensus Engine.

Includes:

• main.tex + refs.bib
• ARXIV_COMMENTS.txt (paste into arXiv Comments field)
• COVER_LETTER.txt (short cover letter template)

QISA Consensus Engine v0.2.4

Release v0.2.4

Highlights:

• Trusted Publishing to PyPI via GitHub Actions (OIDC)
• Deterministic traces + external verification tools
• Adversarial benchmarks (paper-grade)

QISA Consensus Engine v0.2.3

Release v0.2.3

Highlights:

• Trusted Publishing to PyPI via GitHub Actions (OIDC)
• Deterministic, externally verifiable traces
• Adversarial benchmarks methodology

v0.2.2

Adversarial benchmarks snapshot + CI artifacts. See benchmarks workflow for reproducibility.

QISA Consensus Engine v0.2.1

Release v0.2.1

Highlights:

• Non-convergence policy + quality metrics
• Paper-grade tools docs (trace export/verify)
• Adversarial benchmarks + methodology

Repro:

• pip install qisa-consensus-engine==0.2.1
• docker build -t qisa-consensus-engine:dev .
• docker run --rm -v ${PWD}:/app qisa-consensus-engine:dev python benchmarks/bench_adversarial.py

v0.2.0 — Non-Convergence Policy + Quality Metrics

v0.2.0 — Non-Convergence Policy + Quality Metrics (Normalized API)

Resumen técnico
Esta versión eleva QISA a un estándar de “reference-grade” para auditoría y reproducibilidad: define política explícita de no-convergencia, normaliza el API de métricas de calidad y cierra brechas de cobertura bajo un gate estricto en CI.

Cambios principales

1. Política explícita de no-convergencia

• Comportamiento definido y rastreable cuando el fixpoint no estabiliza dentro de max_steps.
• Señalización clara del motivo de parada (stop reason) y trazabilidad consistente para análisis posterior.

2. Métricas de calidad — API normalizada

• compute_quality_metrics acepta fuentes de datos normalizadas:
  • preferencia: trace → records → decisions (+ final_state opcional)
• Métricas base deterministas:
  • disagreement_entropy
  • x_target_variance
  • decision_count
  • final_state_keys (si se provee final_state)
• Retrocompatible: llamada sin argumentos retorna {}.

3. Confiabilidad de ingeniería

• CI coverage gate enforced (fail-under 95%).
• Suite de tests ampliada para cubrir rutas no ejercitadas y estabilizar contratos (incluyendo trazas y validación de tamper).
• Estado actual: tests green + coverage local ~99%.

Determinismo y auditabilidad

• Orden estable para serializaciones/métricas (apto para hashing y comparación reproducible).
• verify_trace es compatible con implementaciones que lanzan excepción o retornan bool/None (test cubre ambas).

Compatibilidad

• Cambios compatibles; el ajuste principal es la normalización del API de métricas para consumo consistente.

Cómo validar (local)

• python -m ruff check .
• python -m pytest -q
• python -m coverage run --source=src -m pytest -q
• python -m coverage report -m

v0.1.3

v0.1.3

• CI green
• Coverage artifact published
• Property-based tests: determinism + tamper-evidence

v0.1.2 — Metadata sync (Zenodo)

What changed:

• Zenodo metadata finalized via .zenodo.json
• README updated with DOI badge + citation
• CITATION.cff updated with Zenodo DOI and version alignment

Why:
Ensure Zenodo archives the repository with canonical metadata and consistent citation info.

Reproducibility:

1. python -m pip install -e .[dev]
2. python -m pytest
3. python -m bench.run_bench > bench/results_scenario_v1.json
4. python -m bench.make_results_table

v0.1.1 — Zenodo archival trigger

No code changes. This release exists to trigger Zenodo archival and DOI issuance.

v0.1.0 — Deterministic consensus core (auditable + reproducible)

QISA Consensus Engine — v0.1.0

Scope

First reference release of a deterministic, auditable consensus engine* with explicit fixpoint semantics.

Core guarantees

• Deterministic convergence (fixpoint + idempotence)
• Tamper-evident trace hashing (hash chain per step)
• Verifiable trace export and verification
• Reproducible benchmark harness with pinned results

What is included

• Multi-perspective opinions + deterministic consensus operator
• Trace export (trace_to_json) and verification (verify_trace)
• Benchmark harness (bench.run_bench) with baselines
• Auto-generated comparison table from pinned JSON results
• Tests covering convergence, idempotence, tamper-evidence, and benchmarks

What this is NOT

• Not an LLM
• Not stochastic
• Not dependent on external services

Reproducibility

python -m bench.run_bench > bench/results_scenario_v1.json
python -m bench.make_results_table

Status

Reference-grade core suitable for audited decision systems and for integration under LLL or non-LLM perspectives.