Summary

  • Okta (see the directory entry) publishes four availability statements for one identity service; three of them carry a number, and none of the three is accompanied by a printed measurement rule for that service.
  • The only okta.com document that prints a divisor, an exclusion list, a credit ladder and a claim deadline is the Auth0 Platform service level agreement, and it governs a different product.

Four papers, one service, and the question a buyer asks in the last week of a negotiation: if this is down, what was it supposed to be up? The answers differ by document, by vocabulary and by two orders of tolerance.

The first paper is the one a new customer is likeliest to sign. The version posted on the vendor's domain is headed OKTA, INC. MASTER SUBSCRIPTION AGREEMENT, carries the marker MSAQ1FY26, and is dated 2025-02 in its filename with no effective date printed in the passage I could read (Master Subscription Agreement). Section 2.1(a) is the whole availability undertaking, and it says:

Okta shall use commercially reasonable efforts to make the Service available to Customer 24 x 7 during the Term (except for any unavailability caused by Non-Okta Products or a Force Majeure Event).

That is a duty, and a real one. It is also a duty with no meter attached: around the clock, at commercially reasonable efforts, minus two named carve-outs. No percentage, no divisor, no observer, no credit, no deadline. A buyer who signs on this text has been promised continuous service and given nothing with which to check the promise.

The second paper prints a number. A separately posted Subscription License and Professional Services Agreement carries, in Section 3.2:

Okta shall (i) provide Support Services to Customer during the Term; and (ii) provide Customer with at least 99.9% availability of the Service in accordance with Okta’s then current Service Level Agreement.

Ninety-nine point nine percent, in a contract (Subscription License and Professional Services Agreement). And then the sentence hands the operative terms to a document it does not reproduce. Incorporation by reference is how contracts work; it becomes a problem for the reader when the referenced document is not at the address the company's own paperwork points to.

That address is the agreements index. It is a public page, not a login wall, and it lists a Master Subscription Agreement, prior versions, support terms, a Product Schedule and a Product Subscription Reference Guide. It does not list a standalone Service Level Agreement or Service Level Policy. It also contains no sentence saying which of the two agreements above governs a new customer. The instrument that would make 99.9% computable is named in a contract and absent from the index that contract points at.

The guide the index does list is harder to reach than it looks. Its link resolves to a support-hosted page that returned a loading shell and a CSS error to every retrieval (Product Subscription Reference Guide). No text, no version marker, no effective date. I do not know what is in it and I will not guess. This is the largest open question in the file, because that guide is the document the current master agreement defines by reference and the one place a product-specific commitment would plausibly live.

The third paper has the biggest number and the most careful sentence in the corpus. The public trust page displays "12-Month Availability: 99.99%", and then, in the same breath:

Because Okta's architecture is highly multi-tenant and includes a large set of use cases, products and geographical regions, the availability displayed here is an average of Okta service incidents that affected at least 10% of customers and is not intended to be representative of each customer's specific service availability.

Read that as an engineer and it is doing a great deal of work (Okta Trust). The figure averages incidents that touched at least ten percent of customers, so an incident that hit nine percent is not in it. It is an average across products and regions, and it says outright that it is not your availability. No as-of date and no start or end for the twelve months appears on the page as I read it. The operator published a headline and then published, in the same paragraph, the reasons not to treat the headline as the promise. That is more candour than most vendors volunteer, and it is also the cleanest available statement that the headline is not the contract.

The second trust surface behaves like a filing cabinet with the drawers locked. The security trust center lists "Status Monitoring of SLA" as a category, shows no commitment, no formula, no credit tier and no claim window, and puts much of its library behind a "Get access" control. Whether an Okta-service service level document exists behind that door is something I did not establish.

Now the document that has everything. The Auth0 Platform service level agreement, header Auth0-Platform-SLA-Rev-4.5.2 02-01-22, publishes the complete regime in two pages: at least 99.99% average monthly availability of Core Services, measured and monitored from Okta's facilities, under a stated division, Availability Service Level % = (total minutes per month – Downtime (in minutes)) / (total minutes per month). It lists its exclusions — emergency maintenance, force majeure, customer load and penetration testing, and alpha, beta or non-generally-available features. It prints a credit ladder of 5, 10, 20 and 50 percent of monthly subscription fees, capped at 50 percent, and requires the customer to claim in writing within five business days of the event giving rise to the credit.

Every clause of that is what a buyer wants, and none of it is attached to the service this article is about. The denominator, the observer, the exclusions and the deadline are printed for a different product. Anyone who lifts that formula into a discussion of Okta's identity service has made the easiest available mistake in this file.

The fourth statement is the oldest and the loosest. A whitepaper last updated in March 2023 says Okta "guarantees 99.99% uptime and zero planned downtime", and then reports its own results: "We achieved a 99.9958% uptime in 2021 and 99.9969% uptime in 2022" (whitepaper). Two annual figures, in marketing material, with no published measurement basis. Note the vocabulary: uptime, not availability; guarantees, in a whitepaper rather than in an agreement. Whether those annual numbers rest on the trust page's ten-percent threshold is not stated anywhere I read.

What the percentages permit, in minutes. Everything in this section is my arithmetic, by division from printed percentages. None of it is Okta's measurement of anything.

At 99.99%, a 30-day month of 43,200 minutes permits 4.32 minutes of unavailability, and a 31-day month permits 4.464. At 99.9%, the same months permit 43.2 and 44.64. Over a 365-day year of 525,600 minutes, 99.99% permits 52.56 minutes and 99.9% permits 525.6. The gap between the professional-services figure and the Auth0 standard is therefore 38.88 minutes in a 30-day month and 473.04 minutes in a year, just under eight hours. These are not neighbouring standards; one is ten times more forgiving than the other.

The whitepaper converts the same way: 525,600 × 0.000042 gives about 22.1 minutes of reported unavailability across 2021, and 525,600 × 0.000031 gives about 16.3 minutes across 2022. Both sit inside the 52.56 minutes that 99.99% would allow in a year. That comparison can be made against a tolerance. It cannot be made against a method, because no method is published.

Where the reconstruction stops. Four shapes of one promise. A duty with no quantity in the current master agreement. A quantity with a missing instrument in the professional-services agreement. A headline that disclaims itself on the trust page. A complete, auditable regime for another product. Across the eight addresses cited here, no page states which agreement governs a new customer, and no page prints a counting rule for this service's service-level commitment.

What a buyer can still do is worth saying plainly. Put the question in writing before signature and name the document that is the service level agreement, with its version. Ask who measures, from where, and how one cell's month is computed. Ask which maintenance counts against the commitment, since the one Okta instrument I could read excludes emergency maintenance and, by its silence, leaves ordinary maintenance standing. And keep your own availability series: a record you hold is the only input to this argument that nobody else supplies for you.

Every PDF finding above is scoped to the passages I could retrieve through page-text extraction, not to a rendered read of each whole document. The 99.9% is that document's figure, on its own page; the current master agreement prints no figure at all. A promise without a meter is still a promise. That distinction is the whole point, and this file will not let you collapse it.