Journal
Notes from the workshop.
Build notes, proof essays, and research teardowns in one institutional voice. Nothing here is asserted on faith: every claim is backed by the repository or by real delivery data.
Pieces
-
What the intake Pilot has to save to pay for itself
Illustrative ROI arithmetic a buyer can redo with their own numbers: what an inbound-intake workflow costs a year at a coordinator-salary anchor, against a one-time $6,000 Pilot. Not a case study — no client, no result claimed.
-
What happens when the AI is wrong
The question a serious buyer asks first, answered as an architecture: no autonomous action, deterministic routing over model judgment, and a record that cannot leak — traced file by file through the studio's own two systems.
-
The gate that checks our own marketing site
A stdlib-only deterministic gate runs the studio's own marketing site through the same rigor as its software: forbidden trade dress, palette drift, pinned prices, and a no-tracker JavaScript policy, enforced by code on every push.
-
Why Assay is stdlib-only
Why the studio's proof CLI imports nothing outside the Python standard library — and why that constraint is what makes the zero-dependency posture binding rather than stylistic.
-
A committed ledger with no field for PII
The intake system commits its ledger to git so the authenticity claim is auditable — and builds the record so it cannot carry inquiry content: the shape, the allowlist, and the test that pins them together.
-
An automatic credit, no dispute required
Why a self-executing milestone-miss credit — owed the moment a delay is attributable, with no claim required — changes the risk calculus for a buyer burned by fixed-bid overruns. Argued from drafted mechanics, not client outcomes.
Start here
Start with a Pilot.
One workflow, fixed scope, 30 days — at a published price. Tell us what the work is and what "done" looks like; we reply within one business day.
Start with a Pilot