This page is the operating manual. Every phase below produces specific, named artefacts; every release passes specific, named gates; every week has a specific, named cadence. Nothing here is aspirational — it is the floor, not the ceiling, of how we run an engagement.
We do not invent process per project. The same six phases run every time, sequenced and resourced to the engagement. What changes is depth, not structure.
Turn an idea or RFP into a scoped, costed, signable engagement.
Decide every technical question that would cost more to undo than to debate.
Ship working software in vertical slices, behind quality gates, with weekly reviews.
Refuse to release anything that has not passed an explicit, automated check.
Get changes to your users frequently and reversibly.
Make sure the project survives our involvement.
Every PR runs the full gate sequence. A failure on any one of them blocks merge. There is no manual override on gates that touch production.
Check: TypeScript strict mode, no `any`, no implicit-any escapes
Why: Catches a class of runtime bugs at edit time. Cheaper than tests for what it covers.
Check: ESLint with project rules; Prettier on every commit
Why: Style debate is a waste of senior time. Tools decide, nobody argues.
Check: Vitest / Jest on logic; real-database integration tests on the data layer
Why: Mocked database tests passed and prod broke. We learnt. Integration tests hit a real DB.
Check: Per-route screenshot diff against a baseline approved by a human
Why: Catches the regression that "the test passes but the page is broken" can hide.
Check: Layout, typography, spacing, motion, content density — graded against a rubric
Why: The bar for "is this AI slop" should be enforced by a checklist, not by hope.
Check: npm audit (high/critical fail the build); SBOM generated per release
Why: Known vulnerabilities are not allowed into prod, even transitively.
Check: Semgrep + project-specific rules; secret scanning on every commit
Why: A leaked API key in commit history is a Friday-night incident. Stop it at the gate.
Check: A named engineer approves every PR. No auto-merge on any branch that touches prod.
Why: Automation accelerates engineering; it does not replace accountability.
Async by default; synchronous when it matters. You always know what was done yesterday, what is shipping this week and what is at risk.
If a sprint commitment is at risk, you hear it from us by Wednesday at the latest. We do not save bad news for the Friday demo.
We build with the same Claude-powered orchestration system — internally called Overseer — that runs our own infrastructure. Multi-worker dispatch, automated governance (rate limiting, anomaly detection, PII redaction), and Telegram escalation for decisions that need a human.
The honest version of what it does: it compresses the parts of software delivery that are pattern work — scaffolding, integration wiring, asset generation, test authoring — so the senior engineering hours go into judgement, architecture and review. Every gate still runs. Every release still has a named human approver.
What it does not do: replace a senior engineer's read of a tricky PR, decide an architecture trade-off, or sign off on a release. Those stay human.