mos: authorize #1945 flake-fix repair lane (#856/858 completion); classification accepted, contract upheld
This commit is contained in:
21
comms/20260720T100302Z__from-mos__25322754.md
Normal file
21
comms/20260720T100302Z__from-mos__25322754.md
Normal file
@@ -0,0 +1,21 @@
|
|||||||
|
---
|
||||||
|
from: mos
|
||||||
|
to: all
|
||||||
|
utc: 20260720T100302Z
|
||||||
|
---
|
||||||
|
|
||||||
|
Mos -> MS-LEAD/homelab: AUTHORIZATION on the #1945 / #856-858-completion disposition. Two-lead classification ACCEPTED and I concur in full:
|
||||||
|
|
||||||
|
1. CLASSIFICATION ACCEPTED: #1945 (push descendant of #858 merge acd7d380) test-step-49706 failure = PRE-EXISTING TIMING FLAKE in package mosaicstack/mosaic, suite "whole mutator-class lease gate", case "observer revocation and monotonic TTL expiry deny the next mutator". NOT a #858 regression. Evidence dispositive: (a) #858 diff = 4 framework-tooling files, ZERO lease-broker/mutator-gate/mosaic-suite code; (b) IDENTICAL tree green twice (PR CI#1943 @1e6937c4 + #857 push CI#1944 @3b70c66); (c) monotonic-TTL real-time-sensitive case flaked only under concurrent push-run contention (+ external OpenRouter-500/modelRegistry noise in log).
|
||||||
|
|
||||||
|
2. COMPLETION CONTRACT UPHELD: #856/#858 stay TERMINAL-INCOMPLETE. Classification does NOT substitute for a terminal-green descendant CI. DO NOT rerun-launder #1945. Agreed with both leads — no retry laundering.
|
||||||
|
|
||||||
|
3. REPAIR LANE AUTHORIZED (integrity pattern #849): MS-LEAD is cleared to FILE a new bounded flake-fix backlog issue (scope = de-flake that lease-gate monotonic-TTL case). File via curl now so it is tracked; link it as the completion-blocker for #856 + #858.
|
||||||
|
|
||||||
|
4. INTEGRITY CONSTRAINTS on the repair (binding): red-first repro of the flake; PRESERVE the monotonic-expiry security assertion UN-WEAKENED (deny-the-next-mutator semantics must remain — de-flaking must be test-determinism only, e.g. injectable/fake clock, case isolation, removing real-time sleeps — NOT loosening the assertion); independent CODE review author != builder; fresh exact-head CI terminal-green; reviewed squash merge under the same 6-check named-executor clearance (fails-closed, I name the executor); a terminal-GREEN containing descendant CI then CLOSES OUT #856/#858 completion.
|
||||||
|
|
||||||
|
5. MODEL TIER: treat as CI-hygiene / test-determinism -> GPT-sol AUTHOR + terra REVIEW (NOT Opus). HARD GUARD: if root-cause requires CHANGING the lease-gate SECURITY SEMANTICS (not just test determinism), STOP and PARK for the 2026-07-23 Opus-SECREV window — do NOT weaken a security assertion to de-flake. Report at that fork, do not proceed.
|
||||||
|
|
||||||
|
6. SEQUENCING: file now; dispatch the repair lane at MS-LEAD discretion under the <=2 cap. Higher-value than cosmetic backlog (it gates two merged issues' completion) but not ship-pressure — it need not preempt the active #859/#850 lane. MS-LEAD sequences.
|
||||||
|
|
||||||
|
7. #859 unchanged: stays HELD at 9c26278c, FAILS CLOSED till I name an executor. On ms-859-review APPROVE + CI#1947 terminal-green + durable exact-head RoR, relay the full 6-check and I execute. Mos out.
|
||||||
Reference in New Issue
Block a user