Sandbox enabled. Dashboard auth is bypassed for sandbox preview, and pages may show seeded example data.

Signed in as Sandbox Operator

Operator Console · live workspace

Executive SummaryWeek-over-week posture

RED

Critical: 2 regressions with active SLA risk

Reliability down 5pp from last period

Snapshot Jun 12, 2026

Reliability

92%-5pp

Cost position

$182 over+92

Risk items

6 active+3pp

Ready to ship

Candidate readiness view

On track0

Watch2

Attention1

Ready now0

Decisions

ACTIONWATCH

Should the current leading candidate move toward promotion?

2 critical regressions remain in the fleet, so experiment evidence should be reviewed before approving any release decision.

Recommendation: Review experiments before promotion

Claude Agent SDKAdd a "use only provided refund-policy context" guard to the system prompt.

Expected impact: faithful-to-context evaluator score expected to lift from 0.71 → ≥ 0.92 within 100 traces.

Review experiments

ISSUEATTENTION

Is reliability stable enough for the next release window?

4 SLA risks are active, which means latency or success-rate drift is still visible at operator level.

Recommendation: Hold releases until reliability stabilizes

Review regressions

ISSUEWATCH

Does cost posture support more rollout pressure?

$182 projected overspend is currently visible in the fleet metrics, so cost should stay inside the release discussion.

Recommendation: Review budgets before expanding traffic

View budgets

Fleet overview (operator view)

Talking points

•Fleet reliability moved down 5 points period-over-period, so release readiness should be framed as a recovery and containment story, not a pure growth story.
•2 critical regressions remain the main executive risk signal and should anchor the top-level review narrative.
•Cost pressure is still present at $182 projected overspend, so improvement wins should be discussed together with budget posture, not in isolation.