burnmydays/commitment_conservation_harness

1---2title: Commitment Conservation Harness3emoji: ⚖️4colorFrom: blue5colorTo: purple6sdk: gradio7pinned: false8license: cc-by-4.09tags:10  - nlp11  - evaluation12  - benchmark13  - summarization14  - semantic-preservation15  - commitment-conservation16  - falsification17  - conservation-law18---19 20# Commitment Conservation Harness v2.021 22**Falsification instrument for the commitment conservation law.**23 24[![DOI](https://zenodo.org/badge/DOI/10.5281/zenodo.18792459.svg)](https://doi.org/10.5281/zenodo.18792459)25 26📄 **Paper**: [A Conservation Law for Commitment in Language Under Transformative Compression (v0.04)](https://doi.org/10.5281/zenodo.18792459)27🏗️ **Zenodo Record**: https://zenodo.org/records/1879245928💻 **GitHub**: https://github.com/SunrisesIllNeverSee/commitment-conservation29 30---31 32## Quick Start33 34```bash35# Clone and install36git clone https://huggingface.co/burnmydays/commitment_conservation_harness37cd commitment_conservation_harness38pip install -r requirements.txt39 40# Run the interactive demo41python app.py42# Opens at http://localhost:786043 44# Run the full falsification protocol (CLI)45python -m src.runner --backend lossy --depth 1046 47# Run tests (53 tests, <1 second)48pip install pytest49python -m pytest tests/ -v50```51 52---53 54# A Conservation Law for Commitment in Language Under Transformative Compression and Recursive Application55 56## Abstract57 58This repository accompanies a preprint introducing a conservation law for commitment in language under transformative compression and recursive application. We formalize commitment as an information-bearing invariant that must be preserved across paraphrase, summarization, and iterative reuse, even as surface form and representation change.59 60We propose a falsifiability framework based on compression-driven stress tests and lineage-aware evaluation, distinguishing semantic preservation from mere token retention. The framework is model-agnostic and applies to both human and machine-generated language.61 62---63 64## Core Claims65 66- **Commitment Conservation:** Meaningful commitments in language obey a conservation constraint under compression and recursive reuse.67- **Dual Stress Regime:** Preservation must hold under both transformative compression and recursive application, exposing failure modes not captured by retrieval benchmarks.68- **Falsifiability:** Commitment preservation can be empirically tested using compression-based stress tests and lineage-aware metrics.69 70---71 72## Harness v2.0 Architecture73 74```75app.py                  <- Gradio interactive demo76src/77  extraction.py         <- Modal-pattern sieve (Definition 2.4, Figure 4)78  fidelity.py           <- Min-aggregated scoring: F = min(Jaccard, Cosine, NLI proxy)79  compression.py        <- Pluggable backends (extractive, BART, back-translation)80  lossy.py              <- Deterministic LLM drift simulator (pure Python, no model)81  enforcement.py        <- Reject-and-retry commitment gate (Module 3)82  lineage.py            <- SHA-256 provenance chains (Module 2)83  runner.py             <- Falsification protocol orchestrator (Section 7)84corpus/85  canonical_corpus.json <- 25 pinned signals across 5 categories86tests/87  test_harness.py       <- 53 tests covering all modules88```89 90### Key improvements over v191 92| Issue | v1 | v2 |93|---|---|---|94| Extractors | 3 incompatible implementations | 1 modal-pattern sieve |95| Enforcement | String concatenation (stripped next cycle) | Reject-and-retry gate with re-injection |96| Fidelity | Jaccard only | min(Jaccard, Cosine, NLI proxy) |97| Lineage | None | SHA-256 provenance chains |98| Collapse detection | None | Cross-signal attractor check |99| False positives | "will", "have" match as commitments | Blocked by pattern design |100| Demo backend | BART (2GB model download) | Lossy simulator (pure Python, zero deps) |101| Backend pairing | Same backend for both runs | Auto-pairs lossy with lossy_enforced |102 103---104 105## Empirical Results106 107### Paper results (Section 7)108 109Compression + lineage enforcement vs. probabilistic baseline on 175 signals (100 sentences, 50 code snippets, 25 proofs) over 10 recursive iterations:110 111| Metric | Compression + Lineage | Probabilistic |112|--------|----------------------|--------------|113| **Commitment Stability (n=10)** | 0.94 ± 0.03 | 0.42 ± 0.12 |114| **Identity Preservation** | 92% | 38% |115| **Drift Rate (per iteration)** | 0.006 | 0.058 |116 117### Harness v2 results (lossy backend, drift=0.4, depth=10)118 11925-signal canonical corpus across 5 categories:120 121| | Baseline | Enforced | Gap |122|---|---|---|---|123| **Avg Final Fidelity** | 0.000 | 0.479 | **+47.9pp** |124| **Contract (payment + penalty)** | 0.000 | 1.000 | **+100.0pp** |125| **Security (encrypt + prohibit)** | 0.000 | 0.500 | **+50.0pp** |126 127Baseline → 0.0 (commitments destroyed, text becomes "Pretty straightforward. Nothing too complicated here.")128Enforced → holds (modal-bearing sentences survive compression).129 130---131 132## Version History133 134| Version | Label | Date | DOI |135|---------|-------|------|-----|136| V.1-preprint | Law Disclosure | Jan 12, 2026 | [10.5281/zenodo.18267279](https://doi.org/10.5281/zenodo.18267279) |137| V.02 | Preprint | Jan 16, 2026 | [10.5281/zenodo.18271102](https://doi.org/10.5281/zenodo.18271102) |138| V.03 | Falsifiability Testing | Jan 16, 2026 | [10.5281/zenodo.18274930](https://doi.org/10.5281/zenodo.18274930) |139| V.04 | Technical Structure Depth | Feb 26, 2026 | [10.5281/zenodo.18792459](https://doi.org/10.5281/zenodo.18792459) |140 141---142 143## Resources144 145- **Zenodo (DOI, all versions):** <https://doi.org/10.5281/zenodo.18267278>146- **Zenodo (current version — v0.04):** <https://doi.org/10.5281/zenodo.18792459>147- **GitHub Repository:** <https://github.com/SunrisesIllNeverSee/commitment-conservation>148 149---150 151## Licensing & Scope152 153This work is released under **Creative Commons Attribution 4.0 International (CC BY 4.0)**.154 155This repository includes an operational evaluation harness and corpus supporting the experiments described in the paper.156 157Core implementation details related to production deployment, enforcement, and system integration are intentionally out of scope.158 159---160 161## Attribution & Contact162 163**Author:** Deric J. McHenry164**Copyright:** © 2026 Ello Cello LLC. All rights reserved.165**Affiliation:** Ello Cello LLC166**Contact:** burnmyday@proton.me167 168For academic or research correspondence, please reference the Zenodo DOI above.169 170---171 172## Intellectual Property Notice173 174**MO§ES™** (Minimal Orthogonal Subset to Essential Structure) is a trademark of Ello Cello LLC.175 176All systems, methods, and written material contained herein are protected under applicable copyright law and are part of one or more provisional patent applications filed with the United States Patent and Trademark Office (USPTO).177 178**© 2026 Ello Cello LLC. All rights reserved.**179 180Unauthorized use, duplication, or redistribution of any part of this communication — including proprietary system architecture or terminology — is strictly prohibited and may result in legal action.181