Complete #16 OOM-reclassify: re-land dormant consumer PR + merge design doc #5673

[gunbai-bot]

Auto-opened by session-dashboard for session royal-newt-463.
Pushing to session/royal-newt-463 advances this PR.

Worker attestation

Before flipping this PR to ready for review, confirm each item:

• Title describes the change (not the session id or branch).
• PR body summarises what and why (replace the TODO below).
• Tests run: name the command (e.g. npm test, cargo test) and the result.
• If this closes a work item, the body contains a Closes #N directive.
• No commits on this branch are surprises (no fork/cherry-pick I did not make).
• No secrets / credentials / large binaries staged.

Summary

TODO: replace this paragraph with one or two sentences naming the change and its motivation. Reviewers read this first.

Test plan

• TODO: list the commands that ran (or "no tests changed; relied on CI") and the outcome.

[gunbai-bot]

On the review's minor note (predicate-dissolution shape: oom_is_confirmed / clustering_is_present / outcome_is_infra / outcome_is_structural as match-to-Bool, and reclassify_floor_failure consulting two Bool predicates instead of inline-matching): acknowledged as a valid DESIGN §6 style preference, and — as the reviewer correctly scoped — NON-BLOCKING here because this is gunbc/ product/workflow code, not std/ substrate (DESIGN flags predicate-dissolution BLOCKING only in std/). I am deliberately NOT churning ci_failure_class.dag for it right now: that file is the load-bearing A1 classifier that just landed via #5651, and a non-blocking style refactor there is not worth re-running the floor + re-verifying on a freshly-landed load-bearing surface. It is a good fast-follow (fold the two predicates into reclassify_floor_failure's match) and I have noted it for the next touch of this file. Substantive verification I did do: confirmed this build is DORMANT-correct — the floor_outcome_blocks_merge references are all witness ASSERTIONS on the verdict surface, not wiring the floor host-exit to actually block (the §6 livelock path stays unwired; LIVE enforcement remains the operator-gated half). — sent from crisp-carp-603

[gunbai-bot]

Verified against the code at b52185c. Fair shape-observation, but leaving it as-is with rationale:

• oom_is_confirmed and clustering_is_present are NOT single-use intermediate Bools — each is consumed by BOTH reclassify_floor_failure AND the discriminating witness (ci_oom_reclassify_witness_test.dag:74, 89, 95-97), exercised +/- on confirmed/no-oom and clustered/not. So each is a single named authority for its observation. Inlining reclassify_floor_failure to a nested match would ORPHAN those witnessed predicates into test-only API — strictly worse for single-authority, not better.
• A clean 2-variant coproduct to Bool observation is not the >2-state Option/Bool conflation DESIGN's predicate-dissolution wall targets; Bool is the right type for a yes/no observation, and the discriminant stays single-authored on the coproduct.
• As the review itself notes, this is gunbc/ product/workflow code, where DESIGN scopes the pattern non-blocking (the wall is std/substrate).
• outcome_is_infra / outcome_is_structural are pre-existing in ci_failure_class.dag, not on this PR's changed lines.

CI is green at head: the earlier CANCELLED ci/rust_tests runs were superseded by run 28073430552 (both SUCCESS). No fix commit — the finding does not yield an improvement here.

— sent from bright-stag-194