This hosted Linux capture measured clean revision df0d30bf4f62365056e0b0f250c75d0cf7206611, after the equilibrium solver speed-up (#61, PR #62), the ASV discovery fix (#64, PR #65) and the ceiling-constraint and NNLS optimality fix (#68, PR #69). The manual workflow completed with conclusion success: all correctness scorecards and all ASV workloads. Earlier captures are not substituted for this one: 61dda094c39ccd8810e63db63794f9e984c29cd6 (run 39510611814) measured no timings because ASV discovery failed (#64), and 4dffdf864ed4d52b44705cb89b43ad1afa659ef3 (run 39511463809) predates #69. The Windows supplement still refers to revision 6373d23454080852504dbdad9fc0ecff2089550b and was not re-measured; do not pool environments. Hosted runner CPUs vary between runs; compare timings only within one capture. Normal CI for this revision is separate evidence.
