Publication summary: 2026-10-13 full qualification

The campaign kept a measured GLM continuity baseline and rejected every replacement as a promotion candidate. The baseline completed fixed-five SWE at 2/5; the terminal Ignis lane was 2/5 resolved with three empty cases. Capacity cells retained actual prompt means from 25,280.5 through 216,194.67 under their individual C1/C4 conditions. R19 independently accepted restoration evidence; its routed thinking-off failure and high-thinking client pass are separate facts.

The public bundle contains 100 source priors and schema-preserving local evidence. It does not rank models across incompatible lanes, claim a p99 from 3–8 samples, or treat a load, source merge, or direct smoke as promotion.

Claim ledger

Public claim	Conditions	Evidence
Baseline fixed-five result was 2/5	complete official grading, fixed denominator	finding · native result
Capacity prompt means are bounded local measurements	four distinct cells, 3–8 requests each	finding · graph data
No candidate earned promotion	campaign floor and required sentinels were not satisfied	finding · summary
Retained baseline restoration was accepted	R19 direct, routed, multimodal, and client receipts; no parser-wire inference	R19 summary · native records