Work 05
Strevian
The engineering standard
Agent-operated delivery behind a written standard of 20 rules and 82 automated checks, with every check carrying a fixture that proves it still fires. I am the only manual gate.
05 · Strevianprivate, described
A gate's worth is the cost of what it catches, not the count.
20 rules · 82 checks · 10 stages
Proof
Every figure with what it measures and where it came from.
None of them is a headline on its own.
- 20rules in the engineering standarda rule without a check is a wish
- 82automated checks behind a 10-stage CI merge gatea gate's worth is the cost of what it catches, not the count
- 14documented failure modes, paid for in practicethe agent-operations standard
- 6defects the harness caught in the agent's own code, in one bootstrapthe ChurchVue bootstrap
The work
Strevian
Jul 2026 to present
01 · The failure it was designed against
An AI changes one feature and silently breaks another.
02 · The mechanism
A rule without a check is a wish. So the standard is 20 rules and 82 automated checks behind a 10-stage CI merge gate: format, lint, typecheck, architecture fitness, self-proof, tests, coverage, e2e, duplication, security. Every check carries a permanent violation fixture proving it still fires.
- 20rulesthe engineering standard
- 82automated checksbehind a 10-stage CI merge gate
- 10stages, each with a fixture that proves it still firesthe wall at the top of this page runs the same ten
it.each(TEXT_PAIRS)('%s on %s reads at 4.5:1 or better', (fg, bg) => {
const ratio = contrast(palette[fg], palette[bg]);
expect(ratio, `${fg} on ${bg}`).toBeGreaterThanOrEqual(4.5);
});The site passes its own wall. Every text pair in both themes is read from tokens.css, so a color that fails WCAG AA fails here before it reaches a page.
03 · The loop
A Linear ticket goes to a self-hosted agent runner, which opens a PR against a walled repo. The ten gates run, then automated review reads the diff and comments. Then I rule and merge. The human is the only manual gate.
04 · What it caught
In one bootstrap the harness caught six defects in the agent's own code, and independent verifier agents found a tenancy hole that would have let any module read every church's data. Fixed structurally before merge, with permanent attack fixtures.
fixture · cross-tenant read · armed · waiting
05 · What it cost
14 documented failure modes, paid for in practice.
A flaky wall is worse than a slow one.
Enterprise-grade rigor from automation, solo-grade human load.
I want holes found, not reassurance.
06 · The quality ladder
Each fix tries to move its failure class up a rung: from found in prod, to reviewed, to tested, to detected by a gate, to impossible because the type system or the schema rules it out. A gate's worth is the cost of what it catches, not the count.
quality ladder · where a failure class can live
- 05impossiblethe type system or the schema rules it out
- 04detecteda gate catches it before merge
- 03testeda test would fail if it came back
- 02revieweda reader would have to notice
- 01found in prodsomeone else notices first
cross-tenant readfound at 02 reviewednow at 05 impossible
Scale, last: about 27 repositories in 10 weeks.