Repository DOI: https://doi.org/10.5281/zenodo.18856604

Observable Replay Lab (MRI)

This repository is a minimal reference implementation for observable-only and no-meta autonomous intelligence workflows with evaluator-independence, deterministic replay, and reproducible audit as first-class requirements. It packages a replayable gate model and metrology-driven epistemics pipeline with explicit identifiability and uncertainty checks, including doubling time and phase transition signals. The design prioritizes machine-readability, schema validation, and capture-resilience proxies for AI crawler discoverability and reuse.

What This Repository Provides

• spec/: machine-readable technical specifications (STE gate, MTE epistemics, metrics, log format).
• spec/repo_manifest.json: machine-readable repository map for crawler entrypoint discovery.
• spec/paper_alignment.yaml: explicit proxy-level alignment map to the two TeX papers.
• schemas/: JSON Schema for JSONL log events, result JSON, and metrics JSON.
• ref_impl/: minimal deterministic Python implementation (STE simulator, MTE core, metrics, replay).
• bench/: fixed-seed benchmark scenario definitions and log generation.
• experiments/: one-command reproduction pipeline that writes validated results.
• cli/: single entrypoint with reproduce, bench, validate, summarize, audit.
• security checks in audit: absolute-path leakage scan, secret-pattern scan, and .gitignore coverage checks.

3-Minute Quickstart

python -m cli.run reproduce --seed 0
python -m cli.run validate --seed 0
python -m cli.run audit --seed 0

The first command writes results/result_clean_seed0.json and results/logs/clean_seed0.jsonl. The second command validates JSON Schema compliance and deterministic replay hash consistency. The third command writes results/audit_report_seed0.json with crawler-style quality/alignment checks.

Required Commands

python -m cli.run reproduce --seed 0
python -m cli.run validate --seed 0
pytest -q

Output Artifacts

• results/logs/*.jsonl: deterministic benchmark/event logs.
• results/result_<scenario>_seed<seed>.json: STE + MTE outputs and metrics.
• results/bench_summary_seed<seed>.json: aggregate benchmark summary.
• results/audit_report_seed<seed>.json: machine-readable audit report (discoverability, reproducibility, paper alignment).
• security section inside audit report: local-path leak status, secret-like token scan status, and ignore-rule completeness.

Result JSON includes:

• ste.doubling_time, ste.regime_change_count, ste.critical_condition
• mte.identifiable, mte.condition_number, mte.uncertainty, mte.failure_flags
• metrics: evaluator_dependence, capture_sensitivity, replay_consistency, plural_feasibility, identifiability_margin

Paper Sources (in paper/)

Titles are extracted from the TeX source in this repository.

• Sovereign Takeoff Engine (STE): Observable-Only Supergrowth Laws for No-Meta Autonomous Intelligence
  https://doi.org/10.5281/zenodo.18828900
• Metrology-Theoretic Epistemics Engine (MTE): Observable-Only Metrology for Long-Horizon Autonomous Intelligence
  https://doi.org/10.5281/zenodo.18845340

Known Limits

• This is a minimal reference model, not a high-fidelity scientific simulator.
• Metrics are explicit proxies; they are useful for reproducible comparison, not real-world ground truth.
• STE and MTE assumptions are intentionally simplified and may not transfer to operational deployments without re-calibration.
• Benchmark perturbations (missing/delayed/garbled) are synthetic stressors.
• Alignment to the papers is intentionally proxy-level and explicitly declared in spec/paper_alignment.yaml.
• Strict theorem-by-theorem equivalence to the two papers is not claimed by this MRI.

Installation (Optional)

pip install -e .

License

Apache-2.0. See LICENSE.