Writing · 12 essays · 14,723 words
Notes from the work.
Written from real incidents, with the root cause and what changed. No hot takes; where the evidence is thin, the essay says so.
Honest systems
- A Check Is Only As Honest As Its Name
The most common defect I keep finding isn't a missing check. It's a check whose name promises more than its query delivers.
verification · testing
5 min read - Right Number, Wrong Story
A verified count answers how many. It never answers why — and the gap between the two is where I've done my most expensive damage.
verification · root-cause-analysis
5 min read - The Discipline of the Second Measurement
Almost every verification mistake I make has the same shape: one measurement comes back clean, and I stop.
verification · testing
6 min read
Working with models
- AI Evals Measure the Software, Not the Work
Most AI evaluation suites drift toward checking whether the system behaved correctly instead of whether the work it produced was any good — build the coverage matrix before you trust a single metric in one.
ai · evaluation
6 min read - Jev and the End of Prompt-and-Parse
TypeSafe's Jev returns a typed judgment and a calibrated number instead of a paragraph you then have to parse — worth testing against the heuristic it would replace, not worth trusting on the vendor's word.
ai · typesafe
7 min read - Physics, Not Prompts: Where Generative AI Doesn't Belong
A model that reasons in tokens will hallucinate on physical law — the honest answer for deep-tech R&D is surrogate models grounded in physics plus a knowledge graph of prior art, and knowing where not to reach for an LLM matters as much as knowing where to.
ai · physics-ai
6 min read
Practice
- PDF Accessibility Is Harder Than It Looks: Two Proposals That Didn't Convert
The hardest part of document accessibility isn't fixing the tags — it's diagnosing whether a file can be fixed at all, and an honest diagnosis sometimes costs you the deal.
accessibility · pdf
5 min read - Training Engineers for the Six Stages of AI-Assisted Development
Engineers don't skip to advanced AI-assisted workflows because they're told to; they climb a six-stage ladder one wall at a time, and training only works when it meets people at the wall they're actually stuck on.
training · ai
6 min read - What Building an Accessibility Practice Taught Me About Real Conformance
Automated conformance is an assertion; manual testing plus a code-level fix is evidence — and building a practice on the second one kept costing me deals the first would have won.
accessibility · standards
5 min read
Building in public
- The Compliance Trap: Why Measuring AI Adoption Is Harder Than Rolling It Out
Reporting that a practice was adopted is not the same claim as reporting that it worked, and collapsing the two in a leadership report is the standing trap of any measured rollout.
building · measurement
5 min read - The Log Is the System: Building Astram on One Append-Only Event Stream
Every dashboard in my personal operating system is a view of one log, nothing is ever hand-edited, and that one rule — not a feature list — is what makes the system trustworthy enough to run a multi-venture life on.
building · astram
5 min read - What Replate Taught Me About Building for Someone Else's Life
A venture with no external deadline doesn't lose a fair fight against client work — it loses by default, silently, and the fix isn't more discipline, it's reprioritizing by what can't be undone.
building · replate
5 min read