feat(verification): discriminated-union spec + deadline runner + VERIFIERS registry#27
Draft
pradeepvrd wants to merge 1 commit into
Draft
feat(verification): discriminated-union spec + deadline runner + VERIFIERS registry#27pradeepvrd wants to merge 1 commit into
pradeepvrd wants to merge 1 commit into
Conversation
pradeepvrd
commented
Jun 20, 2026
pradeepvrd
force-pushed
the
submit/1-models-loop
branch
from
39c87ff to
88c9939
Compare
pradeepvrd
force-pushed
the
submit/2-verification
branch
from
6592b55 to
a510df6
Compare
pradeepvrd
force-pushed
the
submit/1-models-loop
branch
from
88c9939 to
2827366
Compare
pradeepvrd
force-pushed
the
submit/2-verification
branch
2 times, most recently
from
9500fba to
2da6431
Compare
pradeepvrd
force-pushed
the
submit/1-models-loop
branch
from
2827366 to
49f0968
Compare
…FIERS registry Verification used to live in `pkg/agents/verifier/` (`verifier.py`, `spec.py`, hand-maintained leaf checks); this migrates it into `devops_bench/verification/` with a registry-driven spec, a deadline-bounded runner, and a `VERIFIERS` registry. **Behavior changes** - Spec parsing is registry-driven (`VERIFIERS.register()`): new verifiers register themselves and appear in the schema without editing a central discriminated `Union`. - The runner computes one `time.monotonic()` deadline shared across the whole spec tree: `sequence` consumes it serially and fail-fasts; `parallel` hands each child the full remaining budget and ANDs the results. - `timeout_sec` is a float clamped to the deadline (was an int); a leaf short-circuits when under ~1s of budget remains instead of issuing a doomed `kubectl wait`. - Compound results nest typed `VerificationResult` children (`children` + leaf `raw`) instead of dicts/lists; nodes require an explicit `type` (no compound inference from bare lists/dicts). **Bugs fixed** - `parallel` checks used to block until the timeout even when all children finished early; they now return as soon as the children complete. - Per-child timeout rebudgeting (`initial − elapsed`) let repeated children drift and extend the overall budget; the shared monotonic deadline removes that.
pradeepvrd
force-pushed
the
submit/2-verification
branch
from
2da6431 to
415e316
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Verification used to live in
pkg/agents/verifier/(verifier.py,spec.py, hand-maintained leaf checks); this migrates it intodevops_bench/verification/with a registry-driven spec, a deadline-bounded runner, and aVERIFIERSregistry.Behavior changes
VERIFIERS.register()): new verifiers register themselves and appear in the schema without editing a central discriminatedUnion.time.monotonic()deadline shared across the whole spec tree:sequenceconsumes it serially and fail-fasts;parallelhands each child the full remaining budget and ANDs the results.timeout_secis a float clamped to the deadline (was an int); a leaf short-circuits when under ~1s of budget remains instead of issuing a doomedkubectl wait.VerificationResultchildren (children+ leafraw) instead of dicts/lists; nodes require an explicittype(no compound inference from bare lists/dicts).Bugs fixed
parallelchecks used to block until the timeout even when all children finished early; they now return as soon as the children complete.initial − elapsed) let repeated children drift and extend the overall budget; the shared monotonic deadline removes that.