InnerWeb
Skip to main content

SLA and update requirements in a system contract: a guide

A contract for a safety system should describe not only the price and scope, but also how the vendor maintains and updates the software. This guide points to SLA, version and update window clauses, and to the ways in which the plant can check that the vendor meets them.

3-6 mo.
pilot before full rollout
2-2.5 h
full end-to-end test run

What the contract should describe besides the price

For purchasing and IT, the key is to separate four areas: the subscription with SLA, updates, custom development and responsibility for hardware. The system is offered on a subscription basis with a one-time implementation fee, and the pilot runs for 3 to 6 months before full rollout, so the contract should clearly state when the period covered by the SLA starts and how it treats the pilot period.

The documentation lists the maintenance options: server and licences with SLA, maintenance of transmitters including battery replacement as part of the service, and rental of mobile devices, all billed as a monthly subscription. It is worth treating each of these options as a separate clause of the contract, so the plant knows exactly what it pays for and what it can require.

SLA clauses: what to pay attention to

We do not give specific response times here, because they depend on individual pricing and agreements with the plant. We do name the elements that should always appear in the contract, regardless of the figures.

  • Definition of a report and the channel for receiving reports.
  • Priority classes: separately a failure of a safety alarm, separately an error in an auxiliary function.
  • Response time and resolution time for each class, stated in numbers.
  • Service hours and the way time outside them is counted.
  • The way SLA is measured: data source and reporting frequency.
  • Consequences of missed deadlines.

How to tell a failure from a non-failure in the clause

Exceptions that describe invalid input, such as missing access, a missing page or a bad request, are excluded from reporting as non-failures. It is worth reflecting this in the contract: events caused by wrong user actions do not count towards the failure SLA, but can still be reported as questions.

Safety alarms as a separate class

Work safety alarms are not silenced by a cache failure: if a failure occurs, the alarm still rings and the log records the problem. The plant can write into the contract a requirement that a failure of an auxiliary component may not silence an alarm, and ask the vendor for evidence.

Contract items to describe

Contract items to describeHorizontal bar chart: 3 maintenance options, 6 SLA clauses and 4 clauses on versions and update windows.0246number of itemsMaintenance options in thedocumentationMaintenance options in the documentation: 3 number of items3SLA clausesSLA clauses: 6 number of items6Clauses on versions andwindowsClauses on versions and windows: 4 number of items4
Source: and the summaries in this guide

Clauses on versions and update windows

Contracts most often lack a description of how an update is introduced and how the plant accepts it. It is worth writing in: advance notice of an update, a time window outside the plant's working hours, the way a change can be rolled back, and an obligation to give the version number after every change.

The version number is the basis of identification. In error reports the release version is the short hash of the commit of the deployed image, the same as the image tag in the registry, so every error can be tied to a specific change. The contract may require the vendor to give this short hash in its release notice.

An update and the plant's hardware

InnerWeb is the producer of software that integrates with the client's existing systems, so an update on the InnerWeb side does not replace the client's readers or cards. The contract should record this as an explicit limit of responsibility: a software update does not create an obligation to replace access control hardware.

How to check that the vendor meets the clauses

A clause alone is not enough if the plant has no way to verify it. The best evidence is what the vendor produces in its normal work: test results, the error report and the change history.

Every change in the repository goes through static analysis, style checks and unit tests, and merging is blocked until the result is green. Running browser end-to-end tests after a fix catches user-visible errors that unit tests miss. The full suite takes about 2 to 2.5 hours, and the contract may require the result of the last run before every major update.

The plant may also require an annual review: a list of releases, a list of reports with response times and a list of known, unfixed bugs. For the last list, a good model is the practice of marking a known bug in a test as an expected failure with a ticket number, never by skipping the test.

Video

Frequently asked questions

Should the contract state numerical response times?

Yes, in numbers and separately for each priority class. The values are set in the individual pricing.

Can a software update require replacing readers?

No. An update on the InnerWeb side does not replace the client's readers or cards, because the system integrates with the existing access control.

How can I check that a fix does not break other functions?

Ask for the result of the latest E2E run and for the list of known bugs marked in the tests as expected failures.

Prepare the contract clauses

We will go through the SLA and update clauses for your plant with you. Book a conversation with InnerWeb.

Let's talk