Xeon model qualification artifact bundle

Qualification closed with no promotion. The matched comparison stopped at candidate A1: although A1 passed preflight (25/25), long-tool validation at actual 110760 tokens, and Oracle 18/18, its SWE result was 2/5 on the campaign full-attempt denominator and 2/4 in native official grading (resolve rate 0.5). This is below the 4/5 quality gate. B1, B2, and A2 were not run, and no comparative speed claim is available.

The exact r11 tier was restored and readmitted. It passed direct 25/25 and routed 7/7 functional preflights; a fresh Pi read-tool round trip returned the exact unseen fixture. Both tiers were ready and admitting, the candidate was unloaded, and the retained router/model identities were unchanged. Promotion is false.

The retained GLM contract evidence covers the stated preflight, Oracle, vision8, bounded context, and functional C4 cells. It does not establish a coding improvement or speed improvement over r11. The closed inventory hashes all retained public files; Git delivery and CI remain pull-request work, not benchmark evidence.