PromptScan: pass
SLO Definition Workshop
Facilitates defining SLOs for a service: picks user-facing SLIs, proposes targets grounded in current performance, sizes the error budget, and specifies the alerting policy.
Add a trust badge to your README
[](https://waxmark.app/items/slo-definition-workshop/scan)
Run an SLO definition exercise for the service below. The output is a draft SLO document the owning team can argue about in a one-hour meeting. Propose specific numbers; a draft with "TBD" targets gives the team nothing to react to.
<service>
{{SERVICE_DESCRIPTION}}
</service>
<user_journeys>
{{USER_JOURNEYS}}
</user_journeys>
<current_metrics>
{{CURRENT_METRICS}}
</current_metrics>
<dependencies>
{{DEPENDENCIES}}
</dependencies>
<business_context>
{{BUSINESS_CONTEXT}}
</business_context>
<current_metrics> holds whatever the team has: availability over the last 30/90 days, latency percentiles, error rates, throughput. <dependencies> lists upstream services and their SLOs if known. <business_context> covers who the users are, what failure costs, contractual SLAs, and the team's capacity for reliability work. Where an input is empty, proceed and mark every dependent number `[assumed: reason]`.
Method:
1. For each journey in <user_journeys>, identify the one or two moments where the user would notice failure, and define an SLI for each as a ratio of good events to valid events, measured where the user experiences it (load balancer or client, not inside the service). Write the exact event definitions: what counts as valid, what counts as good, what is excluded and why.
2. Propose a target for each SLI from <current_metrics>: start at or slightly below current performance, then adjust for <business_context>. Explain each target in one sentence. A target far above current performance is a project, not an SLO; say so and propose an interim target.
3. Check targets against <dependencies>. An SLO tighter than a hard dependency's SLO is unachievable without redundancy; flag it and propose either a looser target or the mitigation it would require.
4. Choose the window (rolling 28 or 30 days is the default) and compute the error budget in concrete units: minutes of downtime, number of failed requests at current traffic. Numbers people can picture drive better conversations than percentages.
5. Define the alerting policy: a fast-burn alert (budget consumed at a rate that exhausts it in hours; page) and a slow-burn alert (consumes it over days; ticket). State burn-rate thresholds and lookback windows for each. Avoid alerts on raw thresholds that do not map to budget.
6. Write the error-budget policy: what the team does when the budget is exhausted (freeze risky deploys, prioritize reliability work) and when it is healthy (ship faster). Tie it to <business_context>.
7. List the questions the team must answer in the meeting, each with the two or three realistic answers and how each answer would change a number above.
Use the service's and users' vocabulary. Prefer fewer SLIs that matter over many that are easy to measure.
## Output format
**Service and users** - 2-3 sentences restating scope from the inputs.
**SLIs** - one block per SLI: journey, name, good/valid event definitions (exact), measurement point, exclusions.
**SLO table** - columns: SLI, proposed target, current performance (from metrics), window, error budget (minutes or request count at current traffic), confidence (`measured` / `[assumed: ...]`).
**Dependency check** - table: dependency, its SLO (or unknown), SLIs affected, finding.
**Alerting** - table: SLI, fast-burn (burn rate, lookback, action), slow-burn (burn rate, lookback, action).
**Error budget policy** - 4-8 lines: triggers and the team's response at each state.
**Decisions for the meeting** - numbered list; each question, realistic answers, and the number each answer changes.