| # Private release validation summary | |
| The figures below were independently rechecked from the local release archive on 2026-08-01. They describe the private dataset and are published as aggregate engineering evidence only. | |
| ## Final blocking validation | |
| - status: pass; | |
| - JSONL rows parsed: 216,868 across 58 files; | |
| - CSV files checked: 6; | |
| - reported errors: 0; | |
| - reported warnings: 0; | |
| - final SFT rows: 5,262; | |
| - sealed evaluation cases: 266; | |
| - preference pairs: 304; | |
| - source-ledger identifiers: unique; | |
| - source references in evaluation and generated records: resolved; | |
| - manifest path and size checks: pass; | |
| - packaged image verification: pass. | |
| ## Important interpretation | |
| The source archive also contains intermediate audit reports created before the final split was rebuilt. Some of those reports record normalized duplicate groups and train/evaluation leakage that the final blocking validator later reports as resolved. | |
| The public toolkit therefore presents the final validation result as the release status while retaining this caveat. It does not claim that every intermediate build passed. | |
| ## Known measured limitation | |
| An inherited citation-precision benchmark measured 98.61 percent against a desired 99 percent target. The result is documented rather than rounded up or presented as passing. | |
| ## What was not independently established | |
| - legal permission to redistribute third-party game assets; | |
| - semantic correctness of every private example; | |
| - exhaustive narrative completeness; | |
| - equivalence between decompiled implementation evidence and authorial intent. | |