EDUKATE SINGAPORE · GOLD STANDARD PRACTICAL CASE STUDY 226
A 99.9% uptime objective can be met while an important school service fails
A fictional enrolment portal suffers a 35-minute interruption during a busy application window. Across an otherwise available 30-day month, it still meets the defined 99.9% time-based availability goal. That valid arithmetic does not prove applicants could submit during the interruption or that the administrative outcome was fair.
All incidents, times, families and service goals are fictional. The header image is a contextual library terminal rather than a photograph of the invented outage.
The availability calculation
- Total reporting window: 30 × 24 × 60 = 43,200 minutes.
- Unavailability at a 99.9% objective: 0.1% × 43,200 = 43.2 minutes.
- Observed fictional outage: 35 minutes; uptime minutes: 43,165.
- Calculated time-based availability: 43,165/43,200 ≈ 99.919%.
- An example with 300 failed application attempts out of 2,000 gives 85% request success; it is a separate metric.
- Actual SLA compliance depends on the governing contract’s definitions and exclusions.
Chapter navigation
- 1. The percentage looks excellent until families try to enrol
- 2. Calculate the allowance in the declared thirty-day window
- 3. The same numerical percentage can describe different kinds of availability
- 4. A first page loading is not a completed application
- 5. An outage near a deadline has consequences that are not evenly distributed
- 6. The annual number can be even more misleading without its period
- 7. Monitoring, alerting and incident response have separate roles
- 8. The school must plan a fair alternative path
- 9. The correction should retain the uptime figure but qualify the assurance
- 10. Test the numbers using a calendar and a request ledger
- 11. A proportionate reliability improvement plan
- 12. The final finding distinguishes a measured target from an acceptable outcome
1. The percentage looks excellent until families try to enrol
A fictional school’s online enrolment portal advertises a monthly time-based availability objective of 99.9%. During a busy registration window, the service becomes unavailable for 35 minutes. The operator notes that, in a thirty-day month with no other outages, the portal would still exceed its numerical time-based objective. Several families, however, cannot complete a time-sensitive application during that window.
The school is tempted to conclude that a compliant monthly percentage proves the registration process worked well. The families’ experience demonstrates a different question. A service may be available for almost every minute yet fail at a time when access matters unusually much.
The operator does not necessarily have a dishonest monitoring system. A time-based metric can be calculated accurately and still be an incomplete measure of the user’s task. The school must distinguish calendar availability, successful application submissions, deadline fairness and continuity options.
All institutions, users, counts, incident causes and service-level statements in this case are fictional. The featured library-terminal photograph is contextual and does not document a real enrolment outage.
2. Calculate the allowance in the declared thirty-day window
A thirty-day month contains 30 × 24 × 60 = 43,200 minutes. Under a simplified time-based 99.9% availability target, the permitted unavailability share is 0.1%, or 0.001 of the total. Multiplying 43,200 by 0.001 gives 43.2 minutes.
The Google Site Reliability Engineering availability table gives the same 43.2-minute figure for a thirty-day month at 99.9%, assuming no planned downtime exclusions. The table also shows that a nominal 99.9% annual level corresponds to about 8.76 hours of allowed downtime in a 365-day year. The measurement period is indispensable.
With one 35-minute outage and no other counted downtime in the fictional month, available time is 43,165 minutes. The measured time-based percentage would be (43,165 ÷ 43,200) × 100, approximately 99.919%. Under that simplified metric it exceeds the 99.9% target.
This arithmetic is not a verdict about a contract. A real service-level agreement may specify a different measurement window, exclusions, monitoring methods or remedies. The fictional target is defined only so readers can see why a user-visible failure and a high measured uptime can coexist.
3. The same numerical percentage can describe different kinds of availability
Time-based availability divides time marked available by total eligible time. Request-based availability instead compares successful eligible operations with all eligible operations. Those methods need not give the same result, especially when demand changes dramatically over the day.
The Google SRE chapter on embracing risk discusses both approaches and why request success can be a useful user-centred measure. A portal may run during quiet overnight hours and be unavailable during the ten minutes of busiest submissions. Time-based success can then look better than the experience of applicants.
For another teaching scenario, suppose 2,000 eligible application submissions are attempted in the reporting period and 300 fail because of the outage. The fraction of successful submissions would be 1,700/2,000, or 85%, if there are no other failures and each attempt is counted under the stated rule. These invented numbers illustrate a request-based measure; they are not derived automatically from the 35-minute outage.
A good report labels whether it is describing available minutes, successful HTTP requests, fully completed applications or another meaningful outcome. Without the denominator and definition, a percentage is an attractive but ambiguous badge.
4. A first page loading is not a completed application
A registration portal can respond to a homepage request while a downstream identity check, file upload or payment verification fails. An availability monitor that checks only the homepage may mark the system healthy even though parents cannot submit valid applications. This is not evidence that any particular monitoring tool is faulty; it shows why monitoring scope matters.
The school’s true goal is that eligible families can complete applications fairly under the published process. A representative synthetic test might check whether a safe test application can move through the required steps, while real user monitoring can help identify errors without exposing private details.
Operationally useful success criteria can include form submission acceptance, authoritative confirmation numbers, data persistence and a retrievable receipt. A display reading ‘Submitted’ is not complete proof if the backend has not actually committed the application.
The Gold Standard aligns the metric with the outcome the family needs. Time-based uptime remains useful, but it should not replace an end-to-end submission-quality measure.
5. An outage near a deadline has consequences that are not evenly distributed
The fictional 35-minute interruption occurs during a widely advertised enrolment rush. It may affect a parent who planned to submit after work, a family sharing a device or someone needing accessibility assistance. The short incident can therefore carry consequences disproportionate to its share of total monthly minutes.
A service-level percentage cannot determine who missed a deadline or whether their application was eligible. Those questions require actual submission logs, communications, policy records and an equitable review process. The school should not assert that every family suffered equally, nor dismiss all affected families because a target was technically met.
A responsible incident record includes the outage start and end times, the relevant application window, the affected functions and the volume of requests where trustworthy data exist. Privacy and security controls must apply to those records.
The metric should help the school prioritise improvement, not excuse inaction. A formally strong percentage is of little comfort when the unavailable window coincides with an important user task.
6. The annual number can be even more misleading without its period
A claim such as ‘99.9% annual uptime’ means something different from ‘99.9% monthly uptime’. With a 365-day year, the time equivalent of 0.1% is 365 × 24 × 60 × 0.001 = 525.6 minutes, or 8.76 hours. That does not grant permission to schedule one uninterrupted 8.76-hour outage during a school-wide application deadline.
An annual target aggregates across very different service conditions. A particular ten-minute window with concentrated demand may merit stricter operational protection than a quiet period. Service objectives must reflect the nature of the service and the consequences of failure.
The fictional school should publish the actual measurement period and avoid swapping a monthly target for an annual one simply because one version makes a dashboard look more attractive. The relevant calculation should be reproducible from timestamps and the adopted rules.
Google’s time-based table assumes a specific availability calculation; it does not establish that a real school’s contract has identical exclusions or that an incident during critical operations would be acceptable to parents.
7. Monitoring, alerting and incident response have separate roles
Monitoring detects relevant service state and failure patterns. Alerting asks someone to act when the observed condition merits intervention. Incident response coordinates diagnosis, containment, recovery and communications. A system can have accurate uptime monitoring while its alert reaches an unavailable operator or its recovery process depends on an untested step.
The school should examine the actual incident timeline: when symptoms began, when detection occurred, who acknowledged the alert, how the failure was confirmed and when successful submissions resumed. A single green monthly score cannot explain these stages.
A useful drill can exercise an authorised test of the application path during a quiet period, check escalation and verify that continuity instructions are understandable to families. Such drills need careful scoping so they do not disrupt live applications or publish sensitive data.
The Gold Standard is a functioning service and response system. The dashboard is one instrument in that system, not the user-facing result.
8. The school must plan a fair alternative path
Time-sensitive educational applications need governance for disrupted submission channels. Depending on the actual programme’s rules and the authority of its administrators, options could include a clearly communicated authorised alternative submission path, documented support desk assistance or an approved adjustment of the deadline.
The technical team cannot unilaterally promise that late applications will count. The rules may come from another school authority or a binding process. A correct response separates what the incident technology establishes from what the authorised decision-makers are permitted to change.
Families should be given accessible instructions and a way to present evidence of an attempted submission without being asked to reveal passwords, confidential documents or private security tokens publicly. Communications should explain the known outage window and next step while avoiding unsupported blame.
An availability objective is not a substitute for governance when an operational failure affects rights or educational opportunities.
9. The correction should retain the uptime figure but qualify the assurance
A suitable public statement might say: ‘Our fictional portal experienced a 35-minute outage during the registration period. Assuming no other counted downtime, that represents approximately 99.919% time-based availability over a thirty-day month and remains above our 99.9% target. We recognise that affected families could not complete submissions during the interruption and are reviewing applications under the programme’s authorised procedures.’
This wording preserves a valid measurement and identifies the concrete failure. It does not claim the incident was harmless, that every request was processed or that deadline exceptions have already been approved.
The school should issue a dated correction if earlier materials equated 99.9% uptime with universal continuous access. It should keep the original outage report, monitored definitions, affected feature list and official administrative outcome. A reviewer can then reconstruct both technical performance and how families were supported.
Evidence-based service communication can be simultaneously precise and accountable. It should neither hide behind a percentage nor exaggerate an incident beyond the available facts.
10. Test the numbers using a calendar and a request ledger
In a classroom exercise, provide a thirty-day calendar with 43,200 eligible minutes. Ask students to calculate the allowed outage at 99.9%, the observed monthly availability after 35 minutes down, and whether the simplified objective was met. They should obtain 43.2 allowed minutes, about 99.919% availability and a numerically met target.
Then give a separate request ledger containing 2,000 attempts with 300 labelled unsuccessful during the incident. Students compute 85% request success under that record. Ask why both percentages can be correct. The answer is that the denominators differ: minutes in one calculation, operations in the other.
Advanced students can test how a page-health probe differs from a complete transaction check, and why service hours, planned maintenance and test traffic change the measurement specification. No exercise should require disrupting a real school portal.
The classroom skill is to read the denominator, timeframe and purpose before interpreting a success percentage.
11. A proportionate reliability improvement plan
The portal operator can improve observability of the full application workflow, investigate the specific outage, document alert and recovery steps and review dependencies likely to fail during concentrated demand. It should test changes before the next critical enrolment window and ensure staff know how to respond when the normal channel is unavailable.
The school should also decide what service-level indicator genuinely reflects family success. A single time-based availability objective may remain helpful for operational planning, while a completed-submission success indicator and a time-sensitive incident response standard may be closer to the educational task.
Governance should focus on the experience and fairness of real users, not merely accumulating more ‘nines’ in a marketing statement. A reliable alternative process can matter as much as a technically faster recovery when deadlines are involved.
The case does not prescribe a universal uptime target. The appropriate design follows risk, obligations, resources and the needs of the people using the system.
12. The final finding distinguishes a measured target from an acceptable outcome
On the stipulated fictional numbers, a 35-minute outage in an otherwise available thirty-day period yields about 99.919% time-based availability. That clears the defined 99.9% objective, but the outage coincides with a critical registration window and prevented some users from completing their intended task. Both are genuine findings.
The management decision is to retain the accurate calculated percentage, disclose the interruption, examine affected submissions and approve any permitted remedy through appropriate authority. The school should also improve its end-to-end service measures and continuity preparations.
No amount of percentage arithmetic proves what happened to each individual application without actual records. Conversely, one incident does not erase the numerical validity of a correctly scoped availability report.
The Gold Standard is a school service that is dependable when families need it, with clear evidence and fair processes when things go wrong—not a single green number on a dashboard.
Practical service reliability checklist
- Define eligible minutes, planned maintenance and the measurement window.
- Monitor completed application transactions, not only homepage reachability.
- Check time-based and request-based indicators separately.
- Keep an accurate incident record including actual user impact and privacy safeguards.
- Ensure a documented, authorised continuity route for deadline problems.
- Correct misleading public statements without erasing legitimate technical measurements.
Frequently asked questions
Can 35 minutes of downtime still meet 99.9% availability?
Yes, under a simplified 30-day time-based calculation with no other counted downtime: 35 minutes is less than the 43.2-minute allowance.
Does meeting uptime mean every parent could submit?
No. Service availability is distinct from completed application success, especially during a concentrated peak.
Why calculate a request-based success rate too?
Because the number of failed operations can reveal user impact that a time-averaged measure hides.
Does 99.9% guarantee a particular deadline will have no interruption?
No. It is an aggregate target over a specified measurement window, not a guarantee about every minute.
Can a school extend the deadline after an outage?
Only as permitted by its actual rules and authorised decision-making. A technical incident alone does not automatically establish such authority.
Primary technical references and connected cases
- Google SRE · Availability Table
- Google SRE · Embracing Risk
- Google SRE · Service Level Objectives
- The Gold Standard of IT Service Management
- The Gold Standard of Business Continuity
- Gold Standard Case 215 · Successful backups without usable restoration
- Gold Standard Case 223 · Mean reply time hiding a long wait
- Gold Standard · Core Requirements and Evidence
- Gold Standard · Casebook
Google SRE materials give general availability arithmetic and measurement principles. All specific figures and administrative policies in this case are hypothetical; the case is not a contract assessment of any real education provider.