probe: the mask-algebra tail is worth replacing; the avx512vl gate is not needed; which replacement wins is NOT yet resolved - #315

Merged

AdaWorldAPI merged 13 commits into

claude/mask-algebra-tail-probe

Sep 16, 2026

Commits on Sep 16, 2026

probe: does the tail descent pay for the mask-ALGEBRA ops, end to end?
claude
committed
05722d4
probe: measure against the REAL mask_and, with a measured noise floor
claude
committed
a88d57c
probe: declare mask_algebra_tail_probe as std-gated
claude
committed
34bf8d0
probe: document the four timing/gate helpers
claude
committed
2936d20
probe: the tail needs a fixed TRIP COUNT, not intrinsics
claude
committed
6aaadc3
blackboard (15): the tail wanted a trip count; cancel the U64x4/U64x2 build
claude
committed
fc009d1
probe: answer the codegen bar the padded tail was chosen for, on aarch64
claude
committed
fa4532a
probe: the "pooled over all passes" line was reading ONE pass
claude
committed
101d7c9
probe: address four codex findings; withdraw the D-vs-F claim
claude
committed
3853fc5
blackboard (16): correct (15) — its timing headline came from a buggy binary
claude
committed
797b0d8
probe: the (S - D) median was pooled and unfiltered, labelled per-pass
claude
committed
525f052
probe: mark the (S - D) median when it is below its own resolution
claude
committed
5385ebf
probe: wrong unit, an unsupported bound, and a checked ternlog claim
claude
committed
80b80e0