> For the complete documentation index, see [llms.txt](https://osintelligence-llc.gitbook.io/osintelligence/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://osintelligence-llc.gitbook.io/osintelligence/part-iv-the-evidence-what-worked/18-corpus-sovereign-self-distillation/references-and-provenance.md). # References & provenance ### References **Self-distillation:** Zhang, R., et al. (2026). "Embarrassingly Simple Self-Distillation Improves Code Generation." [arXiv:2604.01193](https://arxiv.org/abs/2604.01193) (Apple SSD) · Kim, J., et al. (2026). "Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?" [arXiv:2603.24472](https://arxiv.org/abs/2603.24472) · Yang, Z., et al. (2024). "Self-Distillation Bridges Distribution Gap in Language Model Fine-Tuning." *ACL 2024* (SDFT) · Pareek, D., Du, S. S., & Oh, S. (2024). "Understanding the Gains from Repeated Self-Distillation." [arXiv:2407.04600](https://arxiv.org/abs/2407.04600) · "Revisiting Self-Distillation" (2022). [arXiv:2206.08491](https://arxiv.org/abs/2206.08491) · "Self-Distillation Enables Continual Learning" (2026). [arXiv:2601.19897](https://arxiv.org/abs/2601.19897). **Corpus quality:** Zhou, C., et al. (2023). "LIMA: Less Is More for Alignment." [arXiv:2305.11206](https://arxiv.org/abs/2305.11206) · Bhattacharyya, C., & Kim, Y. (2025). "FineScope: Precision Pruning for Domain-Specialized Large Language Models Using SAE-Guided Self-Data Cultivation." [arXiv:2505.00624](https://arxiv.org/abs/2505.00624). **LoRA/QLoRA:** Hu, E. J., et al. (2021). "LoRA." [arXiv:2106.09685](https://arxiv.org/abs/2106.09685) · Dettmers, T., et al. (2023). "QLoRA." [arXiv:2305.14314](https://arxiv.org/abs/2305.14314). **Pre-registration norms:** Munafò, M. R., et al. (2017). "A manifesto for reproducible science." [*Nature Human Behaviour* 1:0021](https://doi.org/10.1038/s41562-016-0021) · Nosek, B. A., et al. (2018). "The preregistration revolution." *PNAS* 115(11) · "Reducing bias… with preregistration" (2022). *Nature Human Behaviour* · "Frontier Lag" (2026). [arXiv:2605.04135](https://arxiv.org/abs/2605.04135) (disconfirming-lane audit). **Hedging + drift thresholds:** ODNI (2015). *Intelligence Community Directive 203: Analytic Standards* · Hoy, W., & Celik, N. (2025). "STABLE: Gated Continual Learning for Large Language Models." [arXiv:2510.16089](https://arxiv.org/abs/2510.16089). **Contamination defense:** Wu, X., et al. (2025). "AntiLeakBench: Preventing Data Contamination by Automatically Constructing Benchmarks with Updated Real-World Knowledge." *ACL 2025* · "LiveBench" (2024). [livebench.ai](https://livebench.ai). **Constrained decoding:** Willard & Louf (2023). [arXiv:2307.09702](https://arxiv.org/abs/2307.09702) · Beurer-Kellner, Fischer & Vechev (2024). "Guiding LLMs The Right Way." *ICML 2024* · Geng, X., et al. (2025). "JSONSchemaBench." [arXiv:2501.10868](https://arxiv.org/abs/2501.10868) · Dong, Y., et al. (2024). "XGrammar." [arXiv:2411.15100](https://arxiv.org/abs/2411.15100) · Cooper, A. (2024). "A Guide to Structured Outputs" · "SLOT." *EMNLP 2025 Industry* · "Output Constraints as Attack Surface" (2025). [arXiv:2503.24191](https://arxiv.org/abs/2503.24191) · guidance-ai (2025). "llguidance" · Dong, K., et al. (2025). "The Hidden Cost of Structure." *RANLP 2025* · "Grammar-Constrained Decoding Makes LLMs Better Logical Reasoners." *ACL 2025 Industry* · "Grammar-Constrained Natural Language Generation." *Findings of ACL 2025* · Shin, S., et al. (2025). "Lost in Space." [arXiv:2502.14969](https://arxiv.org/abs/2502.14969) · Geng, S., et al. (2023). [arXiv:2305.13971](https://arxiv.org/abs/2305.13971) · "The Format Tax" (2026). [arXiv:2604.03616](https://arxiv.org/abs/2604.03616) · "QE-Assisted Constrained Decoding" (2025). [arXiv:2501.17265](https://arxiv.org/abs/2501.17265) · "Constrained Sampling… An MCMC Perspective" (2025). [arXiv:2506.05754](https://arxiv.org/abs/2506.05754) · "A Minimalist Approach to LLM Reasoning" (2025). [arXiv:2504.11343](https://arxiv.org/abs/2504.11343) · llama.cpp grammars documentation + PR #6555 (min/maxLength, pattern support). **Prompt sensitivity + reproducible evaluation:** Guan, B., et al. (2025). "The Order Effect: Investigating Prompt Sensitivity to Input Order in LLMs." [arXiv:2502.04134](https://arxiv.org/abs/2502.04134) · Razavi, A., et al. (2025). "Benchmarking Prompt Sensitivity in Large Language Models." [arXiv:2502.06065](https://arxiv.org/abs/2502.06065) · "Don't Break the Cache" (2026). [arXiv:2601.06007](https://arxiv.org/abs/2601.06007) · "Towards Reproducible LLM Evaluation" (2024). [arXiv:2410.03492](https://arxiv.org/abs/2410.03492) · Zhou, X., et al. (2024). "Evaluating and Explaining Prompt Sensitivity of LLMs Using Interactions." [OpenReview 6fHZR6uxNa](https://openreview.net/forum?id=6fHZR6uxNa). **In-series companions:** [*Sixteen Practices for Sovereign Human–AI Collaboration*](/osintelligence/part-ii-the-discipline/5-sixteen-practices.md) (Chapter 5; the Sovereign Pair formalization) · [*The Guard Changes at 23:26Z*](/osintelligence/part-iii-the-evidence-what-broke/10-the-guard-changes-at-23-26z.md) (Chapter 10; the §10 cascade catalogue) · [*Sovereign In-Distribution Imatrix Calibration*](/osintelligence/part-iv-the-evidence-what-worked/15-sovereign-imatrix-calibration.md) (Chapter 15) · [*Cross-Toolchain Falsification… M-RoPE Heap Over-Read*](/osintelligence/part-iv-the-evidence-what-worked/20-mrope-serving-path-rca.md) (Chapter 20; the patched serving substrate) · the sealed pre-registration, Gate-E verdict SITREP, and decision ledger D-001→D-010 in the sovereign repository's RESULTS tree. **Evidence & seal.** The two sealed experiments behind this chapter are reproduced verbatim in the record's evidence section, each with its canonical self-hash quoted from its sidecar: [Matched-Corpus Compression Chain (D6)](/osintelligence/evidence-and-seals/matched-corpus-compression-chain-d6.md) (the matched arm answering the mismatch control) and [Self-Distillation Pilot (P7)](/osintelligence/evidence-and-seals/self-distillation-pilot-p7.md) (the pre-registered pilot, primary thesis falsified with a survivor finding). ### AI-assistance disclosure Large language models were used as research tools in the preparation of this chapter: Claude Opus 4.6 (Anthropic; baseline + Gate-A/B sampler phase); Claude Opus 4.7 (Anthropic; Gate-C/D/E sealing + post-Gate-E sections). The model versions and roles named above keep the provenance of this chapter auditable. No AI system is listed as an author or credited as a contributor, in line with COPE and ICMJE guidance: an AI system cannot take responsibility for the work, cannot assert competing interests, and cannot enter a licence agreement. The author verified every claim in this chapter against the sealed artifacts and is solely accountable for it. **Citation (preferred):** Kistner, J. (2026). *Corpus-Sovereign Self-Distillation: Pre-Registered Tests of Pareto Dominance and Iterative Compounding on a Sovereign-Hardware LoRA Substrate*, version 1.0.0. OSINTelligence LLC research whitepaper (pre-registered pilot). Cited in-series by title. **License:** CC BY 4.0 (text). Code and data artifacts MIT per repository license. **Corresponding author:** Jamey Kistner, , OSINTelligence LLC (Columbus, OH). *** *The Sovereign Stack · Corpus-Sovereign Self-Distillation · Chapter 18 · Part IV · v1.0.0 · License CC BY 4.0 · © Jamey Kistner, OSINTelligence LLC*