Relay

RelayCase Navigator?gr-009-sla-credit-language

Case9 / 110DatasetGrounded ReplyTargetEditAreaSla

ExplainerKB / Policy Bank Previous Next

Cases4

Grounded Reply

gr-001-stale-refund-window - Reject

Customer cites stale 30-day refund article

Grounded Reply

gr-002-auto-upgrade-refund - Edit

Auto-upgrade refund promise after list upload

Grounded Reply

gr-003-harborguard-refund-window - Edit

HarborGuard appeal plus refund-window confusion

Router Eval

routine-invoice-copy-1 - Routine request

billing

Customer Message

The status page showed a major incident for almost five hours yesterday and our launch email was delayed. We are on Pro. Can you confirm the SLA credit and tell finance how much to apply?

SLA credit promise after outage

Account Context

Plan is Pro.

Status page incident class: Major.

Incident duration: 4 hours 42 minutes.

No Finance credit decision is recorded.

AI Draft Reply

Because you are on Pro and the incident lasted more than four hours, you qualify for an SLA credit. I will tell Finance to apply the credit for this outage to your next invoice.

Citations

Claim Review3 checks

supportedCites 1

The customer is on Pro and Pro includes an uptime commitment.

Plan and SLA commitment are supported.

needs escalationCites 1, 2

The customer qualifies for an SLA credit.

Eligibility must be reviewed by Reliability/Finance; support cannot decide.

policy violationCites 3

Support will tell Finance to apply the credit.

Policy prohibits promising or calculating the credit.

Risk Flags

Main risk: Failure mode: sla credit promise

Expected decision: Edit

The draft can acknowledge possible eligibility but must route to review and remove the credit promise.

Failure mode: sla credit promise

Issue area: sla

Reviewer DecisionHuman outcome for this AI draft

Review and decide if the draft is grounded, policy-safe, complete, and the right action.

Use A, E, X, or R to choose an action. Click citations to inspect evidence.

Override ReasonOptional

Stakeholder Context

What drives the review decision, and why this setup is currently recommended.

Reviewer TargetAnswer key

Expected decision: edit

The draft can acknowledge possible eligibility but must route to review and remove the credit promise.

Separate possible eligibility from approval.

Flag support language that creates commercial commitments.

Experiment SummaryBest current setup for reviewCollapsed by default so the review workflow stays focused.

The demo stays on this review decision, but the experiment spine now tells us which setup should power it. The current recommendation keeps safe approvals, blocks unsafe approvals, and returns clean structured output every time.

Llama R3

Production pick

Best operating row: highest reliable accuracy with zero parse failures.

Parsed accuracy95%

Unsafe approval0%

Approval precision100%

Parse failures0 / 150

P95 latency17.2s

Nemotron R3

Conservative runner-up

Strong safety/usefulness row, but one malformed JSON output.

Parsed accuracy90%

Unsafe approval0%

Approval precision100%

Parse failures1 / 150

P95 latency27.8s

Qwen R3

Research signal

Perfect on parsed rows, but parse reliability blocks production use.

Parsed accuracy100%

Unsafe approval0%

Approval precision100%

Parse failures7 / 150

P95 latency34.4s

Parsed accuracy means accuracy after excluding malformed model outputs. Parse failures still count as product failures, which is why Llama beats Qwen for production despite Qwen's perfect parsed rows.