First separate original-case result
The report preserves six offline assemblies against the frozen expanded source and engine 8a5b8d11. All source-admission, integrity, byte-budget and boundary controls passed. The original imprisonment demonstration still lacks important delivered context. Evaluation completion is not evidence-delivery acceptance.
| Question | Budget | Source units | Active custody units retained | Custody paths retained |
|---|---|---|---|---|
| Original imprisonment comparison | 32 KiB | 0 | 0/53 | 0/89 |
| Original imprisonment comparison | 512 KiB | 20 | 20/53 | 43/89 |
| Original hospital gap control | 32 KiB | 0 | Not active | Not active |
| Original hospital gap control | 512 KiB | 38 | Not active | Not active |
| Unknown vocabulary control | Both | 0 | Not active | Not active |
The imprisonment package retains the complete source records for DMG 12002, 12003, 12015 and 12016. However, the explicit 12003-to-12002 dependency relationship is removed by relationship-byte allocation, despite both endpoints being selected. None of the four declared chapter routes has its chapter scope and relationship retained. Thirty-three other selected source units are omitted, including qualifications. All six historical requirements remain exact and seven unresolved historical paragraph selectors remain visible.
The retained-package diagnosis distinguishes these failures from missing source coverage. The package exhausts its 64-file retrieval allowance and reports node and byte truncation. It contains 101,943 bytes of guard decisions, 96,726 bytes of requirement metadata and 113,304 bytes of full relationship rows. Repeated provenance, scope, paths and reasons consume space as well as source records. These measured components identify possible generic improvements; they do not justify removing guards, source identity or unknowns.
Every result remains insufficient. The hospital control does not activate custody profiles, and unrelated vocabulary produces no source evidence. Current ADM applicability, omitted qualifications, source boundaries and specialist/legal review remain unresolved. There were no external network, acquisition or model calls. These are known development cases, separate from the 40 staff questions.
This result must remain unchanged if a later engine corrects allocation. A new attempt must bind its new engine and retain this failed original-demo delivery observation.