Retry timed-out SMT queries with perturbed seeds; report slowest VCs - #21
Open
larskuhtz wants to merge 1 commit into
Open
Retry timed-out SMT queries with perturbed seeds; report slowest VCs#21larskuhtz wants to merge 1 commit into
larskuhtz wants to merge 1 commit into
Conversation
…slowest VCs Large sweeps are tail-dominated, and part of the tail is seed luck: cvc5's e-matching on Veil's VC shapes can diverge under one random seed and finish in seconds under another, so a query that timed out once may be provable cheaply on a second attempt. * Retry-on-timeout (veil.smt.retries, default 1): each VC gets retry dischargers that only fire after an earlier attempt *timed out* (never after sat, genuine unknown, or non-timeout errors). Attempt k re-runs the discharge tactic with the solver seed set to k (veil.smt.seed -> cvc5 seed/sat-random-seed) and a short budget (veil.smt.retryTimeout, default 120 s) -- timed-out-but-provable queries either finish fast under a fresh seed or not at all. The perturbed options are baked into the proof term via `set_option ... in`, so witness (re-)elaboration at #gen_theorems time replays the exact configuration that succeeded. Retried successes are reported as "(retry k, seed k)" so flakiness stays visible. * Slowest-VCs report (veil.report.slowVCs, default 10): check commands now list the slowest discharge attempts (including failed ones, which burn the full timeout), flagging any above veil.report.nearTimeoutPercent of veil.smt.timeout as near-timeout -- these are the divergence candidates one model change away from a timeout. Attempts under veil.report.slowVCsMinMs (default 5 s) are not reported, so output stays deterministic for fast specifications (#guard_msgs). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
larskuhtz
force-pushed
the
port/smt-retry
branch
from
August 21, 2026 22:32
94ebbca to
ab78509
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.
Large sweeps are tail-dominated, and part of the tail is seed luck: cvc5's e-matching on Veil's VC shapes can diverge under one random seed and finish in seconds under another, so a query that timed out once may be provable cheaply on a second attempt.
Retry-on-timeout (veil.smt.retries, default 1): each VC gets retry dischargers that only fire after an earlier attempt timed out (never after sat, genuine unknown, or non-timeout errors). Attempt k re-runs the discharge tactic with the solver seed set to k (veil.smt.seed -> cvc5 seed/sat-random-seed) and a short budget (veil.smt.retryTimeout, default 120 s) -- timed-out-but-provable queries either finish fast under a fresh seed or not at all. The perturbed options are baked into the proof term via
set_option ... in, so witness (re-)elaboration at #gen_theorems time replays the exact configuration that succeeded. Retried successes are reported as "(retry k, seed k)" so flakiness stays visible.Slowest-VCs report (veil.report.slowVCs, default 10): check commands now list the slowest discharge attempts (including failed ones, which burn the full timeout), flagging any above veil.report.nearTimeoutPercent of veil.smt.timeout as near-timeout -- these are the divergence candidates one model change away from a timeout. Attempts under veil.report.slowVCsMinMs (default 5 s) are not reported, so output stays deterministic for fast specifications (#guard_msgs).