Reproducibility Receipt

Zenodo (concept DOI): https://doi.org/10.5281/zenodo.18267278

Date: January 16, 2026
Status: ✅ Harness Confirmed Operational

Test Execution

All 13 tests passing in test suite:

cd harness
MPLBACKEND=Agg python -m pytest tests/test_full_harness.py -q

Result: 13 passed, 17 warnings in 21.09s

Test Coverage

✅ Commitment extraction (empty/nonempty signals)
✅ Jaccard index (perfect match / zero overlap)
✅ Intersection-based commitment computation
✅ Compression sweep (Prediction 1: σ-invariance)
✅ Recursion drift (Prediction 2: Δ_hard accumulation)
✅ Canonical corpus loading
✅ Transformation pipeline application
✅ Complex signal processing

Key Fixes Applied

Blocking issue: Replaced plt.show() → plt.close() in plotting functions
Test errors: Fixed duplicate function names, removed invalid parameters
File paths: Corrected corpus path from data/ → corpus/
Type checking: Configured .vscode/settings.json to disable strict Pylance checks for research code

Environment

Python: 3.9.6 (virtual environment at .venv/)
Key dependencies: transformers, spacy, matplotlib, pytest
Matplotlib backend: Agg (non-GUI, CI-friendly)

Running Tests

Quick run (recommended):

MPLBACKEND=Agg pytest tests/test_full_harness.py -q

Verbose output:

MPLBACKEND=Agg pytest tests/test_full_harness.py -v

With minimal warnings:

MPLBACKEND=Agg pytest tests/test_full_harness.py -q --disable-warnings

Notes

Tests complete in ~20 seconds (model loading + transformations)
Plots saved to PNG files, no GUI interaction required
Warnings from dependencies (urllib3, matplotlib) are non-critical
Type checking disabled for research flexibility

Harness is research-ready for experimental evaluation.