Repository navigation
docs: scope Windows memory evidence and qualify refusal claims (MR-18) - #37
Conversation
Co-Authored-By: OpenAI Codex <noreply@openai.com>
|
Direct self-review at27322c407f478ac4902bd82b1f547f394641e706/treea72f8ae2ff85c4a4e9ad85f36dfa80ce23b89032, base9da4127003a90842ddafeb03a6f07e7c5877f34c. Full two-file diff read against existing MR-17 threshold/nonzero rule, ledger55 observation and source OnJobMessage/Outcome paths. No unresolved findings. Descendant usage aggregates into the attempt, but a descendant's lower cap is not its own cap; a packet alone cannot attribute a limit. Detail must state own-job evidence and unproven allocation refusal, preserving raw exit when own evidence absent. ADR date amended, no RPC/enum/schema/feature activation. Real evidence distinguishes Core3 detail failures/2 controls from a shared-real-port nested hard packet10 attribution failure with own peak117784576<notification268435456, and from three earlier default-forwarding delivery prerequisite timeouts. No timeout counted as the defect, no manually posted packet/fake signal. Full native final-head Windows3/12s (20261026-040704-27322c407f47-win) and Linux3/44s (20261026-120657-27322c407f47-linux) passed; all six results and both wrappers0 read, required statuses success, fresh main/head/tree unchanged. Primary scope docs https://learn.microsoft.com/en-us/windows/win32/api/winnt/ns-winnt-jobobject_associate_completion_port and https://learn.microsoft.com/en-us/windows/win32/procthread/nested-jobs. OpenAI Codex alone under user instruction, no independent reviewer claimed. Merge before implementation. |
Specify own-attempt-job notification/peak or hard-limit peak evidence before latching memory_limit_hit from a Windows memory packet. Descendant usage aggregates into the attempt, but its lower cap is not the attempt cap. Preserve the existing nonzero/evidence normalization while making result detail state own-job scope and unproven allocation refusal; memory causation is not established. No-own-evidence nested messages preserve raw normal exit classification. AD-0005/date and04 section7 aligned; no schema/enum/RPC/package migration, enforcement feature, credential operation or later-feature activation.
Native Windows baseline: unchanged classification logic plus read-only provenance/composition seams. Coreab04206 gives3 detail failures+2 controls/no skips14.663 s. Real nested job on real shared completion port atcc6ddba delivers kernel hard packet10, nested peak110678016 B, own attempt peak117784576<own notification268435456, yet old own hit=True:1 discriminating failure/no skips4.292 s. Three default-forwarding probes had delivery prerequisite timeouts and are not attribution failures/passes; separate/raw traces retained. Shared-port composition is explicit, no manually posted packet or fake data.
Reviewed head27322c407f478ac4902bd82b1f547f394641e706/treea72f8ae2ff85c4a4e9ad85f36dfa80ce23b89032, base9da4127003a90842ddafeb03a6f07e7c5877f34c. Author zero design errors,protoc25/both schemas,generated C# AOT-compatible zero warnings,47 tool tests pass,diff clean. Merged as
973934c658f62967be4b6cc40d0526d877ee4383after full native Windows3 steps/12 s and Linux3 steps/44 s passed. All6 results,both wrappers0 and required exact-head success statuses read with fresh refs. Direct review and gate.Merged before implementation#58, now merged as
a4c1b2bf1b8afd0112fb490aa590c3ac8199ac96after complete final-head native Windows13/1140 s and Linux16/1382 s. Plan record#70 merged asf9d386c2daa55131f1136cc71f4d0225b323f46bafter its own fresh native gates. OpenAI Codex worked/reviewed alone under the owner's instruction; no independent reviewer claimed. MR-18 complete; STOP here. Remaining68:64 open,3 deferred,1 blocked. MR-19 unstarted; MR-16 and MR-07/11/15 unchanged.Generated by OpenAI Codex.