You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
t27c frontier --audit: rerun a sample of reused verdicts (glue for verdict_audit.t27) #8556
The glue for specs/verified/verdict_audit.t27 (#8542, merged in #8543).
t27c frontier --audit <round> reruns the REUSE specs that audit_pick(spec hash word, round) selects, about 1 in 16 per round, with test-report in process. It prints agrees or POISONED for each spec and exits 1 when any verdict disagrees with its rerun (audit_result). The seal lookup that decide used inline becomes one seal_of function that both share.
Hand-written lines are paid for by deleting the dead one-shot scripts/gen_w367.py.
The PR also adds the t27b ledger row that specs/compiler/zig_shadow_initializer.t27 (#8418) left UNLISTED. CI measured it as pass.
Acceptance criteria
t27c frontier --audit N prints one line per audited spec and exits 1 only when one is POISONED.
A real round on master is recorded in the PR with its counts and time.
Hand-written foreign lines do not grow (own-language gate green).
Given a completed frontier round N with reused verdicts, when running t27c frontier --audit N, then the command reruns approximately 1 in 16 REUSE specs selected by audit_pick, prints "agrees" or "POISONED" per spec, and exits 1 if any verdict disagrees with its rerun.
Requirements
FR-001: t27c frontier --audit <round> MUST rerun REUSE specs selected by audit_pick(spec hash word, round) at approximately 1 in 16 sampling rate
FR-002: The command MUST print one line per audited spec showing "agrees" or "POISONED"
FR-003: The command MUST exit with code 1 when any audited verdict disagrees with its rerun (audit_result), otherwise exit 0
FR-004: The seal lookup used by decide MUST be refactored into a shared seal_of function
FR-005: The dead script scripts/gen_w367.py MUST be deleted
FR-006: The t27b ledger row for specs/compiler/zig_shadow_initializer.t27 MUST be added to docs/reports/t27b_expectations.json
Success Criteria
t27c frontier --audit 1 prints one line per audited spec and exits 0 when all agree
t27c frontier --audit 1 exits 1 when any spec shows "POISONED"
The command completes within 30 seconds for a typical round
Hand-written foreign lines do not increase (own-language gate passes)
A real round on master is recorded with counts and time
Activity
frontier --audit: rerun a sample of reused verdicts (Closes #8556) (#…