Threshold discontinuity
Changing n can suddenly add or remove an allowed miss when ceil(n x target) crosses an integer. More checks do not change the rule smoothly.
Probability
System Availability Confidence Calculator
Estimate the prospective probability that a finite set of independent availability checks will meet an SLA success ratio under an assumed availability rate.
PROSPECTIVE SLA SAMPLE PLAN
Ask whether a verification sample is likely to pass under an assumed uptime rate
This is a design-stage operating-characteristic calculation. It turns an SLA percentage into an exact whole-number pass rule, then sums the binomial outcomes that satisfy it.
Prospective SLA-pass confidence -
Sample miss risk -
Required available checks -
Allowed misses -
Expected available checks -
Expected misses -
LIVE DECISION RECORD
SLA pass-tail ledger
Rows around and above the whole-number pass threshold expose the exact probability assigned to each sample outcome.
| Available checks | Missed checks | Exact probability | Cumulative probability | SLA classification |
|---|
CURRENT CALCULATION PROCESS
Formula, current substitution, intermediate values, and reconciliation
X ~ Binomial(n,p); r = ceil(nt); confidence = P(X >= r) = sum from k=r to n C(n,k)p^k(1-p)^(n-k)
| Symbol | Meaning and unit | Current value |
|---|---|---|
| assumedAvailabilityPct | Assumed availability per check (%) - Probability that one independently sampled check is available. | 99.5 |
| checks | Independent observation checks - Count of checks in the proposed verification plan. | 120 |
| slaTargetPct | SLA pass target (%) - Minimum available-check ratio used to classify the sample. | 99 |
Waiting for valid inputs.
WHO THIS MODEL SERVES
A scoped decision aid, not a universal forecast
Primary audience: SLA owners, reliability analysts, auditors, and test designers choosing a finite availability-check plan.
Decision boundary: Use before sampling to compare pass likelihoods under a stated availability assumption; it does not estimate availability from completed observations.
HOW TO DESIGN THE SAMPLE
Five steps from service claim to pass probability
- Define one check: endpoint, region, user path, and available/unavailable rule.
- Enter the availability rate the plan is intended to discriminate or verify.
- Choose a count that reflects genuinely independent evidence, not raw polling volume.
- Enter the SLA ratio and inspect the rounded-up required count and allowed misses.
- Compare pass confidence with the governance threshold and archive the live tail ledger.
SAMPLING FUNDAMENTALS
Five ideas behind the pass tail
- Bernoulli check
- Each sampled check has two classified outcomes: available or unavailable.
- Prospective confidence
- The chance the rule passes if the assumed per-check rate is correct.
- Integer gate
- The percentage target becomes ceil(n x target), because partial successes do not exist.
- Tail probability
- Confidence sums every mutually exclusive outcome at or above the pass count.
- Effective sample size
- Dependence can make many close checks contain less information than their raw count suggests.
FORMULA AND DEFAULT SUBSTITUTION
Make the pass rule visible before reading confidence
r=ceil(nt); P(pass)=sum from k=r to n of C(n,k)p^k(1-p)^(n-k)
For 120 checks and a 99% target, r=ceil(118.8)=119, so only one miss is allowed. With p=0.995, the live result sums P(X=119)+P(X=120). No normal approximation is used.
DEEPER CONFIDENCE ANALYSIS
Three levers that change the operating characteristic
Assumption sensitivity
A small gap between assumed availability and target can cause a large confidence swing because probability mass sits near the boundary.
Clustered outages
If one incident creates many adjacent failed polls, the independent-binomial model understates tail variability; incident-level blocks may be required.
WORKED PLAN CASES
Two sample plans that answer different questions
Monthly API verification
At 120 checks, a 99% SLA needs 119 available checks. The plan asks how often a 99.5%-available endpoint clears a one-miss gate; it does not claim the endpoint is 99.5% available.
Zero-miss boundary
With a 100% SLA target, every check must be available and confidence is p^n. The boundary makes the compounding effect of a zero-miss rule explicit.
TERMS FOR SAMPLE GOVERNANCE
Six terms used in the result
- Check
- One independently classified service observation.
- Assumed availability
- The per-check success probability under the scenario.
- SLA target
- The minimum sample success ratio required to pass.
- Allowed misses
- n minus the rounded-up required-success count.
- Pass confidence
- Probability mass in all passing sample outcomes.
- Operating characteristic
- How often a fixed rule passes at a given true success probability.
EVIDENCE RETENTION
Preserve the protocol, not just the percentage
Store monitor definitions, timestamp cadence, location mix, maintenance exclusions, incident-correlation treatment, the assumed-rate source, and the pass threshold. A count without its sampling protocol is not reproducible evidence.
LIMITS AND EXCLUSIONS
Conditions required by the binomial plan
- Checks are assumed independent with one stable availability probability.
- The calculator is prospective; it does not produce an interval from observed data.
- Detection quality, false positives, and false negatives are outside the model.
- Maintenance-window exclusions must be resolved before counting checks.
- Passing a sample rule is not proof of contractual compliance or future performance.
RELIABLE SOURCES
Primary references for the probability model
SLA SAMPLE-PLAN FAQ
Questions about prospective pass confidence
Is this a confidence interval for observed uptime?
No. It is a prospective pass-probability calculation for a proposed sample plan. Use a proportion interval after observing successes and failures.
Why round required checks upward?
An SLA threshold is a minimum. A fractional required count cannot pass, so the model uses the smallest whole count that reaches the entered percentage.
Are minute-by-minute checks independent?
Often they are not. Autocorrelation during one incident can make clustered checks behave like fewer independent observations.
Can this prove the true availability rate?
No. It conditions on the assumed rate and reports how frequently the sampling rule would pass under that assumption.
Why can high assumed availability still yield modest pass confidence?
A strict target and small allowed-miss count can leave substantial probability mass on the failing side.
What sampling evidence should be retained?
Keep check timestamps, cadence, outage deduplication, exclusions, the assumed-rate basis, and the exact pass rule.
IMPORTANT SAMPLING NOTE
Independence is an evidence claim
Do not treat polling frequency as effective sample size without testing dependence. Contractual interpretation requires the actual SLA definition and an agreed observation protocol.