principle_01
Measure before you optimise
Every performance engagement starts with a reproducible load test against the current system. A profile beats an opinion, and it turns "it feels slow" into a number we can hold ourselves to.
//architecture
What we optimise for, what we refuse to do, and the stack we reach for. Published because a prospective client should be able to disagree with our approach before they pay for it, not after.
principle_01
Every performance engagement starts with a reproducible load test against the current system. A profile beats an opinion, and it turns "it feels slow" into a number we can hold ourselves to.
principle_02
PostgreSQL until it genuinely will not do. A queue before a service mesh. New technology has to earn its place in a written trade-off, because you inherit the operational cost of every clever decision we make.
principle_03
Timeouts, retries with backoff, idempotency keys, circuit breakers and dead-letter handling are designed in at the start. Retrofitting them into a system that assumed the network always works is most of what a rescue project costs.
principle_04
Repositories, pipelines and infrastructure state live in your accounts from day one. Architecture decision records explain why each structural choice was made, so the next engineer does not have to guess.
//stack
Chosen per project against the constraints, not by preference. Anything on this list we have run in production and been on call for.
//delivery
01
3–5 days
We read the code, the schema and the incident history, then interview the people who operate the system. Output is a written constraint map: current load, real bottlenecks, coupling, and the changes the business actually needs.
02
1–2 weeks
Target architecture with decision records, a migration path that keeps production running, capacity model with headroom, and a delivery plan sequenced by risk rather than by convenience.
03
Project-dependent
Two-week increments, trunk-based, every change behind CI with tests and review. You get a working environment from week one and a demo at the end of each increment — no six-week silences.
04
1–3 weeks
Load tests to the agreed target, failure drills including a restore from backup, security review against OWASP ASVS, then runbooks and a working session with your team. We leave when they can run it without us.
//constraints
Stated plainly, because discovering it three weeks in wastes your money as well as our time.
If we cannot instrument it, we cannot support the claim that it works. Metrics, logs and traces are part of the deliverable, not a phase two.
Rewrites are the most expensive way to be right about architecture. We will propose one only when incremental migration has been costed and is genuinely worse.
No proprietary wrappers, no undocumented deployment folklore, no hostage architecture. Handover includes a working session with your team.
Every estimate breaks down into named work at a published day rate. If the scope is too vague to cost, we say so and propose a discovery phase instead.