Repository navigation
2.0: an async memo over two held verdict sources asks the same lane question twice #4399
Description
What happens
Take an async memo m1 = ask(1 + m0() + m0b()) over two verdict memos, m0 = ask(1 + latest(s) + s()) and m0b = ask(10 * (latest(t) + t())). In the held two-action and two-verdict probes (each write made inside an action that holds), m1:15 is asked twice on the same input. On next it was asked 3–4 times. #4397 removed every other duplicate in the probe matrix, but these 6 are left.
A question asked twice on the same input is a duplicate request. Each question should be asked once.
Repros: review-4397/zz-probe.test.ts (the held rows) and probe-h-wip.txt (project repro folder). Related: #4398.
Activity
ryansolid commentedon Oct 29, 2026
I traced this on #4393's head. #4393 fixes the two-verdict held rows. The two-actions held rows (split, value, latest) still ask
m1:15twice:m1:15is asked in A's verdict lane, which is held under A's transaction, overm0 = 4andm0b = 10.m0bthen asks 20 on mainline, because B'st = 1is visible in B's own transaction.m1re-derives in B's world. It readsm0 = 4, andm0b = 10, since20is still in flight. So it asksm1:15again.It is the same question asked in two seats, and F-9 says it should be asked once. Telling the two passes apart needs their inputs compared, which is a per-pass check. So this becomes an
it.failsrow for the reader-seat carve, landing in a tests-only PR after #4393.Generated by Claude Code
test(signals): pin #4402 and #4398 as fixed by #4393; #4399 as a carv…
fix(signals): a memo waiting on a held flight is the hold's work at i…
ryansolid commentedon Oct 31, 2026
Fixed on
nextby #4510 (2ba10bd); it ships in the next RC.Generated by Claude Code
probe(signals): #4544 probed-held bit — a probe of an unheld node ear…