For service owners and the people they answer to
Over what window is availability measured. What counts as unavailable. What is excluded. Who has to notice and claim. A supplier who chooses those four afterwards can satisfy almost any headline figure, which is why this page hands over the definitions instead of a number we have not independently measured for your operation.
The morning after a bad day, somebody senior asks what the agreement says. The service owner reads it properly for the first time, and discovers that the interruption fell inside a maintenance window, or lasted less than the minimum duration, or affected a subset of functions the definition does not cover, or would have needed a claim within a period that has passed.
Nothing was breached. The agreement worked exactly as drafted. It simply was not describing the event that happened, and the gap between the two was set by definitions that were skimmed during procurement because the headline figure looked reassuring.
The measurement window is the quietest of the levers. Measured monthly, a bad afternoon is visible. Measured quarterly, the same afternoon is diluted into insignificance, and the supplier reports a healthy figure while the operation remembers the day everything stopped.
Then the definition of unavailable, which is usually written as total inaccessibility. An operation degraded to the point of being unusable — timing out, failing intermittently, working for some users — is frequently not unavailable under a definition drafted by the party who would owe something. Everyone who runs a service knows that partial degradation is the common case and total outage is the rare one.
And the remedy, when it exists, is a credit against future fees. The organisation lost a day of work, missed commitments to its own customers, and receives a discount on next month’s invoice from the supplier who caused it. Nobody involved thinks that is compensation, and it is worth being honest that it is not designed to be.
A supplier controls the measurement window, the definition of unavailable and the exclusion list — so a headline availability figure carries no information until those three are fixed, and quoting one before they are agreed is a decoration rather than a commitment.
This page therefore does not quote an availability percentage, and the reason is worth stating plainly because it will look like evasion. We have not independently measured a figure for your operation, and the architecture this whole page set belongs to refuses to quote numbers nobody measured. A figure produced for a page is a figure chosen to look good against definitions the reader has not seen.
What is offered instead is the four definitions in writing before commercial terms. The measurement window is monthly rather than quarterly, because a quarterly window is the single most effective way to make a bad day disappear. Unavailable includes degradation severe enough that the work cannot be performed, rather than being limited to total inaccessibility. Maintenance is excluded only where it was notified in advance and taken inside an agreed window; unnotified maintenance is an outage. And the claim procedure does not require you to notice — the measurement is ours to report and a shortfall is reported to you whether or not anybody on your side raised it.
That last one matters more than it sounds. A remedy conditional on the customer noticing, computing and claiming within a deadline is a remedy that mostly goes unclaimed, and every supplier knows the claim rate is a fraction of the entitlement. Reporting the shortfall ourselves removes the mechanism that quietly makes the commitment cheaper than it appears.
The credit is described honestly as what it is: a proportionate reduction against fees, not compensation for your losses. It is not designed to make an organisation whole and no service credit from any supplier ever has been. What it does is create a cost that makes the supplier’s incentives point the same way as yours, and it should be evaluated on that basis rather than as insurance.
Severity and response are separated from availability because they answer a different question. A severity scale is defined by the effect on your operation rather than by our judgement of the component, response is a commitment to a first human response rather than to a resolution time nobody can honestly promise, and the escalation path is a named route rather than a queue.
Whether the event met the definition of unavailable — measured by the agreed definition, which includes degradation preventing the work.
Whether the window hides it — measured by a monthly measurement window written into the agreement rather than a quarterly one.
Whether the maintenance exclusion applies — measured by whether advance notice reached your nominated contact before the window.
Whether anybody has to claim — measured by the measurement being reported to you unprompted, with the credit applied without a claim.
When a human responds — measured by the committed first-response time per severity, measured from your report.
Who to escalate to — measured by a named escalation route agreed at contracting rather than a queue discovered during an event.
no availability percentage is quoted here, because none has been independently measured for your operation and a figure without agreed definitions carries no information. No resolution time is committed — only a first human response, because nobody can honestly promise when an unknown fault will be understood. A service credit is a proportionate fee reduction and is not compensation for your losses. An independent SOC 2 Type II attestation is in progress and no report exists yet.
A supplier reporting on its own availability is the weakest form of the measurement, and that is true here as much as anywhere. Events land in your log stream, which means your own monitoring can form an independent view rather than relying on ours — and a disagreement between the two is a conversation worth having rather than a problem to avoid.
Your own health checking should run against the functions your operation actually depends on rather than against a general endpoint, because the difference between those two is precisely the degradation case that most definitions exclude. Suppliers know that a general endpoint stays green during a partial fault.
And your incident tooling receives our notice rather than being replaced by a portal you have to remember to check. A status page is a publishing surface, not a notification, and treating one as the other is how an organisation finds out about an event from its own users.
Because the figure is the part a supplier controls least honestly and the definitions are the part that decides everything. Any supplier can present a reassuring number and then choose a quarterly window, a total-inaccessibility definition and a generous exclusion list, and satisfy it comfortably in a year the customer would describe as bad.
It is also consistent with how every other page here works: no number is quoted that has not been measured for the operation it is quoted at. Producing an availability figure for a marketing page would be exactly the unmeasured claim this architecture refuses everywhere else, and making an exception for the number most likely to be relied on would be the wrong place to start.
What is committed is the four definitions, the unprompted measurement, and the credit applied without a claim. Each of those is checkable in the agreement before signature, and none of them depends on believing a figure.
On independent assurance the position is unchanged: a SOC 2 Type II attestation is in progress and no report exists yet, availability would fall within the scope of the one under way rather than being a separate exercise, and it is an attestation with a defined period rather than a certification.
Ask for the measurement window first. Monthly or quarterly is the single largest lever on what any figure means, and a supplier offering a quarterly window with a high number has offered you less than one offering a monthly window with a lower one.
Ask what unavailable means, and specifically whether severe degradation counts. Total inaccessibility is the rare case; the common case is a service that responds slowly enough or intermittently enough that the work cannot be done, and most definitions are written to exclude exactly that.
Ask who has to claim. A remedy conditional on the customer noticing, calculating and claiming inside a deadline goes largely unclaimed and every supplier knows the rate. A commitment to report the shortfall unprompted is worth more than a higher headline figure with a claim procedure attached.
And be clear internally that a service credit is not compensation. It aligns incentives and it does not make anybody whole. If the operation genuinely cannot absorb a bad day, the answer is in the continuity design rather than in the commercial terms, and no supplier’s agreement changes that.
A written agreement on the measurement window, the definition of unavailable, the exclusion list and the claim procedure — settled before commercial terms are discussed at all.
Settling the definitions first is the whole recommendation of this page, and it applies to every supplier rather than only to us. Once they are fixed, a number becomes meaningful and can be argued about sensibly. Before they are fixed, arguing about the number is arguing about nothing.
Set up your own health checking against the functions your operation depends on at the same time, so that the independent view exists from the first week rather than being assembled during the first incident.
Then agree the escalation route by role, and use it once during the trial deliberately. An escalation path first exercised during a real event is one nobody has tested, and the test costs a phone call.
Deliberately, and it will look like evasion until you have read the four definitions. Any supplier can present a reassuring figure and then choose a quarterly window, define unavailable as total inaccessibility, and exclude enough that a year you would call bad satisfies it comfortably. We have also not independently measured a figure for your operation, and quoting one we had not measured would be exactly the unmeasured claim refused on every other page here. Fix the definitions and a number becomes meaningful; before that it carries no information.
They are for nearly everyone, and most definitions of unavailable are written to exclude exactly that. Total inaccessibility is the rare case; the common one is a service responding slowly enough or intermittently enough that the work cannot be performed while a general health endpoint stays green. The definition offered here includes degradation severe enough to prevent the work. Ask every supplier this question specifically — it separates an agreement describing your actual risk from one describing a rarer event.
They do not, here or anywhere, and this page will not pretend otherwise. A credit is a proportionate reduction in fees and it is not compensation; no supplier’s credit has ever made an organisation whole. What it does is put a cost on our side of a bad month so the incentives point the same direction as yours, and it should be evaluated on that basis alone. If your operation genuinely cannot absorb a bad day, the answer lives in your continuity design rather than in anybody’s commercial terms.
We measure and report it to you unprompted, and a credit is applied without you claiming. That is deliberately different from the usual arrangement, where the remedy depends on the customer noticing, calculating and claiming within a deadline — a mechanism every supplier knows produces a claim rate far below the entitlement. You should still form an independent view from your own log stream and your own health checks, and a disagreement between the two measurements is a conversation worth having rather than a problem to avoid.
There is none, and a supplier offering one is committing to something nobody can honestly promise — the time to understand a fault nobody has seen before is not knowable in advance. What is committed is a first human response within a stated time per severity, measured from your report, with severity assessed by the effect on your operation rather than by our view of which component is involved. A resolution commitment in an agreement is either padded until it is meaningless or it is a promise that will be broken on the day it matters.