Working Observations
Notes from the intersection of personal training, AI governance, and evidence-grounded practice.
These are working notes — observations from building things and noticing what connects. We log every step of our process the same way a good personal trainer logs every session: because the log is the program. Some patterns only become visible in the record.
What follows are the ones worth writing up.
The gym outgrows its trainer’s two eyes, and PISTON’s big lift gets cleared while the trainer is tangled in a cable machine across the room. The annotated version: the rule the episode had to break, the name research gives the move that replaced it, the one source quoted verbatim and linked, and a transcript of all thirteen slides.
Read →A new machine walks in having read every manual, asks for every machine at once, and gets one question back. The annotated version: the house protocol written out in full, every source quoted verbatim and linked, a transcript of all eleven slides, and what broke while the sources were being checked.
Read →Somebody turns up with forty-one sources stapled to the back of a training program and gets asked for one of them. The annotated version: every source quoted verbatim and linked, a full transcript of all eleven slides, and the two questions that do the actual work on a citation list.
Read →A comic about three AIs, one trainer, and the unglamorous habit that separates the ones who get better from the ones who get hurt. The annotated version: every source quoted verbatim and linked, plus a free four-question governance starter you can steal.
Read →I built nutrition tools for my clients, then found a site in Australia had gone after the same problem. A field note on the human/AI line: AI for the trustworthy-data and availability layer, the human for relationship, judgment, and the did-it-actually-work check — with the numbers that say where it sits.
Read →Where the 2026 work on agent reasoning meets ALCOA+ data integrity: an autonomous agent's reasoning record can pass the Contemporaneous test and still not be the true account — a new failure class, with the receipts to check it yourself.
Read →A builder-to-builder field note: why automated accessibility checks throw false positives on modern JavaScript / Tailwind sites, how to sort them by cause, where declarative tests and a bounded local AI model actually help — and what nobody has measured yet.
Read →A case study of our own process — how a 530-word LinkedIn post was built on 6,000+ verified research artifacts, three draft variants, and the same verification discipline we use for federal grant evaluations.
Read →When your instincts as a trainer turn out to have names in the negotiation research — and the research hands you a better move for next time.
Read →