Semantic and Epistemic Governance: Ten-Run Results on the Basic Level 1 Tests

Basic Level 1 empirical preprint v0.2.0 (not peer reviewed). Paper 2. An ontology amendment can satisfy its visible goal while expanding a protected permission; an evidence review can reach a plausible conclusion from claims it is not authorized to use. We evaluate six hosted model configurations on this project's basic Level 1 tests: 48 semantic development cases and 96 epistemic packets, with ten planned judgments per case. The reported dataset is explicitly post hoc amended: the original experiment supplied 8,630 of 8,640 planned responses, and ten later fresh generations filled its unknown-outcome slots without replacing any received answer. The incomplete original cohort and all unknown attempts remain preserved separately. Under strict typed bare-JSON scoring, amended ten-pass means range from 0.625 to 100.000 percent for semantic tasks and from 1.250 to 99.792 percent for epistemic tasks. Sole-outer-fence sensitivity changes some results substantially, including Haiku epistemic correctness from 1.250 to 72.917 percent, so formatting failures are separated from completed task errors. We report every round, round variation, per-case consistency, generation intervals for the fixed test set and separate unit-resampling intervals. The 48 semantic cases form twelve construction groups; the 96 epistemic packets are eight views of each of twelve units. Repeated generations add no independent task units. These are descriptive measurements on inspected development material, conditional on the supplied governance contracts, provider behavior and later follow-up timing. They do not establish held-out generalization, real-source trust, equal compute or autonomous-agent containment. Files: reviewed manuscript PDF; scientific supplement containing exact cases/scorers, saved final projections, original and amended full analyses, attempt/usage accounting, all seven CSV tables, figures and offline replay; SHA-256 checksums. The original incomplete record and ten later fresh generations remain distinct. Full native HTTP, reasoning and partial/error bodies remain private. Timothy O'Brien is the sole author and an independent researcher without institutional affiliation. No external funding or competing interests. AI assistance is disclosed in the manuscript; automated checks do not constitute external scientific peer review. Manuscript and author-generated data/documentation: CC BY 4.0. Source code, scripts, configuration and tests are separately licensed under the MIT License in LICENSE-CODE. Third-party sources retain their rights. The deposit preserves the exact GitHub v0.2.0-preprint artifacts: https://github.com/prosessleder/semantic-epistemic-governance-level1/releases/tag/v0.2.0-preprint . Cite the DOI assigned to this Zenodo record; citation text inside the preserved source ZIP predates the DOI.