⚖️ Commitment Conservation — Falsification Instrument

Paste text with commitments. Watch what survives recursive compression.

Baseline systems lose commitments through modal softening, quantity erosion, and conversational drift.
Enforcement systems preserve them. This instrument measures the gap.

📄 A Conservation Law for Commitment in Language Under Transformative Compression and Recursive Application — D.J.M., Ello Cello LLC

Fidelity Curves

How It Works

Extract: Modal-pattern sieve identifies commitments (obligations, prohibitions, constraints)
Compress: Text is recursively compressed through a lossy channel simulating LLM behavior
Measure: Fidelity scored as min(Jaccard, Cosine, NLI) — all three must pass
Compare: Baseline (no awareness) vs Enforced (commitment-preserving selection)

The lossy backend simulates real LLM drift: modal softening (must → should → maybe), quantity erosion ($100 → "the amount"), and sentence dropping. Deterministic and seeded for reproducibility. For results with real models, run the harness locally with --backend bart.

This is a measurement instrument, not a product demo. Paste your own contracts, API specs, medical protocols, legal clauses — anything with commitments — and see whether they survive.