SERVICE-LEVEL PLANNING AND MEASUREMENT WORKSHEET
Original fictional examples and blank form. Checked 2026-10-28.
https://showmeyoursaas.com/articles/service-level-agreement/
References:
https://sre.google/sre-book/service-level-objectives/
https://www.atlassian.com/legal/sla
Educational planning material, not a ready-to-sign agreement.

DEFINE THE MEASURE
Covered operation:
Eligible service / product / audience:
Measurement window and timezone:
Indicator (SLI):
Target (SLO):
Consequences under actual governing terms (if any):
Authoritative evidence source:
Eligible numerator:
Eligible denominator:
Failure classification:
Exclusions and their source:
Missing-data treatment:
Owner and review process:
Do not add a remedy, exclusion, or claim deadline merely because another provider uses it.

ORIGINAL TIME-BASED EXAMPLE
Window: 30 days = 43,200 eligible minutes.
Unavailable time: 60 minutes.
Availability = (43,200 - 60) / 43,200 * 100 = approximately 99.861111%.
Target: 99.9%.
Numerical downtime allowance = 43,200 * (100 - 99.9) / 100 = 43.2 minutes.
Remaining numerical budget = 43.2 - 60 = -16.8 minutes.
This example misses its numerical target.
At exactly 43.2 unavailable minutes, it meets the target within the lab's tiny floating-point tolerance.
For a 31-day window, the same arithmetic allows 44.64 minutes.
These numbers do not establish actual contractual entitlement or acceptable harm.

ORIGINAL REQUEST-BASED EXAMPLE
Eligible requests: 1,000,000.
Classified failures: 500.
Successful requests: 999,500.
Availability = 999,500 / 1,000,000 * 100 = 99.95%.
This measure is not the same as time-based availability.
No eligible requests: percentage and target result remain unknown, not automatically 100%.

DATA PREPARATION QUESTIONS
Have overlapping unavailable intervals been merged before counting time?
Are time periods aligned?
Are requests duplicated by retries?
Does the target concern eventual success, latency, durability, or another property?
Which failures count under the actual measurement policy?
Could a zero count represent missing collection instead of a real zero?

REVIEW RECORD
Observed period and source:
Definition version:
Inputs:
Calculated result:
Missing or disputed observations:
Actual agreement question requiring qualified review:
Operational next step and owner:
Verified outcome:
Use saas-availability-lab.mjs to reproduce the arithmetic with explicit inputs.
The lab monitors no service and interprets no agreement.