Most engineering work is sold as a rate and a rough estimate, which puts the risk of the unknown on you. That is why it overruns. The cost is dominated by things nobody can see from outside: an abandoned dependency with no maintained successor, a query that degrades only past a certain table size, a test suite too slow and too flaky for anyone to trust, an application that will not boot on a fresh machine.
We sell it as two fixed prices, and the first one exists to make the second one honest. The assessment finds those things first. It is underwriting rather than a sales step, and it is why we can hold a fixed price afterwards.
Three problems, one method
Step one: the assessment
Two weeks of reading your application, your telemetry or your delivery history rather than talking about them, for one application or one team.
- A written report your senior management can act on, with every finding tied to the evidence behind it.
- Baselines acknowledged in writing before any work starts: production errors, flaky tests, response times or delivery flow, whichever the problem calls for. They are the difference between something the work broke and something that was always there, and they protect you as much as us.
- Fixed prices for each phase that follows, valid 90 days.
The guarantee. If the assessment concludes you should not do this work, or should not do it with us, you pay nothing. We would rather tell you that in week two than discover it in month four, and it means there is no incentive to manufacture scope.
Step two: the work, at a fixed price
Quoted from the assessment findings, phase by phase, before any of it starts.
- Small increments, each shipped and soaked before the next begins, so a problem traces to one change, and the work can pause between phases without leaving anything half done.
- Your team keeps shipping. Nothing we do asks you to freeze feature work.
- A review clause that stops the delivery clock when your reviewer is unavailable, rather than quietly consuming the schedule.
- Liability capped at fees paid, and the AI-tooling disclosure written into the statement of work rather than left for you to discover.
Where the work happens is your choice: our own managed machine, your Anthropic tenancy, or a virtual desktop you supply, on which your source code is never downloaded at all. Whichever applies is named in the statement of work rather than promised in a meeting, and each costs something. How we work with your code.
We do not publish project prices. The honest number depends on what the assessment turns up, and a published range would only invite everyone to expect the bottom of it.
What we do not do
We remediate software so that an auditor stops flagging it. We do not render compliance opinions, issue certifications, or sign anything an assessor relies on. Your assessor decides whether a control is met. Our job is to remove the condition that created the finding, and to hand you the documentation that proves when it was removed.
We do not promise performance or productivity numbers in production, and we do not evaluate individual developers. Production results depend on traffic, features and infrastructure that all move at once, so we agree a target and the environment it is measured in before the work starts, and report what production does. The Engineering Throughput Review looks at the workflow and the codebase, never at people.
We also do not staff a team onto your project. One named engineer does the work, from the assessment through to the last phase, which is the trade this practice makes deliberately.
How it starts
For software past its end of life, send us your Gemfile.lock. One file,
with no application source code in it, and shareable under a one-page NDA if you
need one. For a performance problem, tell us what is breaking and when your
next big day is. For AI tooling, tell us which tools your team has and what
you expected them to change.
From there: a video call to confirm scope and what is forcing the date, a short agreement, and two weeks.