> 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/evidence-and-seals.md).

# Evidence & seals

Every results chapter in the monograph traces to pre-registered, sealed experiments: hypotheses hashed and dated before the run, verdicts reported against them in full.

Pointers to the sealed artifacts, hash manifests, and the assembled monograph of record (OSF, with DOI) land here as they ship.

*The sealing discipline itself is open source: the MethodSeal package (skill, enforcement hook, sealing routine, and independent verifier) is free to download and use.*

***

[**Verifying a Seal**](/osintelligence/evidence-and-seals/verifying-a-seal.md) · Start here: how a seal is made and how anyone can check one, the canonical-prefix self-hash, the sidecar, the freeze gate, and the deviation discipline, with a step-by-step check any reader can run.

**The sealed experiments,** each reproduced with its frozen pre-registration or report of record and its self-hash:

[**Pruned-122B Quality Gate (D7)**](/osintelligence/evidence-and-seals/pruned-122b-quality-gate-d7.md) · Behind Chapter 17. A pruned 122B mixture-of-experts, pre-registered before the run, PASS on all three hypotheses.

[**35B Prune Experiment (P3, mismatch control)**](/osintelligence/evidence-and-seals/35b-prune-experiment-p3-mismatch-control.md) · Behind Chapter 16. A pre-registered 35B expert-prune that returned a composite FAIL and became the mismatch control.

[**Matched-Corpus Compression Chain (D6)**](/osintelligence/evidence-and-seals/matched-corpus-compression-chain-d6.md) · The matched-corpus arm that answers the 35B mismatch control: the same chain with the sovereign corpus at every layer, a composite FAIL by reframe rather than by data.

[**Self-Distillation Pilot (P7)**](/osintelligence/evidence-and-seals/self-distillation-pilot-p7.md) · Behind Chapter 18. A thesis falsified on its primary gate that left a stronger survivor finding, and the run whose irreproducible hash created this record's reproduce-don't-recompute rule.

[**APEX Imatrix Calibration (P5)**](/osintelligence/evidence-and-seals/apex-imatrix-calibration-p5.md) · Behind Chapter 15. Sovereign-imatrix quantization that made the 35B Architect smaller, faster, and quality-indistinguishable from its baseline, all four validation gates green.

[**IOC Classifier Micro-Agent (P4)**](/osintelligence/evidence-and-seals/ioc-classifier-micro-agent-p4.md) · Behind Chapter 13. The first sovereign in-weights micro-agent, a 0.8B LoRA IOC classifier that beat a prompted 2B baseline 100 percent to 22 percent on a held-out probe.

[**CTI-NER A/B (P6)**](/osintelligence/evidence-and-seals/cti-ner-a-b-p6.md) · Behind Chapter 14. A sovereign 0.8B CTI-NER LoRA that beat the 9B orchestrator 2.79x on relaxed entity-F1 at 11x smaller, trained with no external teacher.

[**Watcher L4 Cascade Classifier (cycle22)**](/osintelligence/evidence-and-seals/watcher-l4-cascade-classifier-cycle22.md) · Behind Chapter 19. A pre-registered sovereign 2B LoRA replacing a rule-fallback path, carried by an H1-H6 hypothesis battery.

[**Watcher Gen-2 Literal-Statement Classifier (cycle25)**](/osintelligence/evidence-and-seals/watcher-gen-2-literal-statement-classifier-cycle25.md) · The pre-registered Watcher retrain that replaced class-ID output with a literal operator-voice infraction statement, tested on a labeled-negative specificity battery.

[**mRoPE Serving-Path RCA (P7.5 / P7.6)**](/osintelligence/evidence-and-seals/mrope-serving-path-rca-p7.5-p7.6.md) · Behind Chapter 20. A Blackwell-CUDA hypothesis falsified by a Windows crash log, the real defect isolated to llama.cpp core and fixed by an upstream cherry-pick.