Resources

Technical notes, published when they are verified.

No content calendar, no thought leadership. These are engineering write-ups about decisions we made and defects we found, written so that someone else can check them.

Technical notes

Published.

Failure analysis · 6 Nov 2026

When “no chest pain” becomes chest pain

A reproducible polarity inversion in code-switched clinical negation: the sentence, the extracted fact, the mechanism inside the pipeline, and why the validator could not catch it.

Open defect R1

Design rationale · 6 Nov 2026

Why no language model writes our clinical notes

What deterministic rendering buys, what it costs in fluency and template coverage, and the exact point at which the argument stops being honest.

Design decision

In progress

What we are writing next.

Listed so the gaps are visible, not to promise delivery dates.

  • Subject attribution: why “my father had a heart attack” is hard, and defect R4.
  • What fail-closed validation can and cannot catch — a taxonomy of errors it misses.
  • Audio retention design: what 24 hours buys and what it does not.
  • The 12-case development suite: what it is, and why it is not a benchmark.

Each of these will be published with the defect it concerns still open, or with a resolution date if it has been fixed.

No newsletter

There is no mailing list and no subscription form on this site. We are not collecting addresses to market to. If you want to follow the work, the honest route today is to ask for a demonstration and stay in contact by email.

Request a demonstration

Primary sources

Where the evidence actually lives.

Defect register
Eight open, reproducible extraction defects with reproduction strings — research page.
Architecture
The four-stage pipeline, the fact-graph schema and the operational envelope — platform page.
Known limitations
Security posture, retention behaviour and everything we do not yet have — security page.
Clinical policy
Safety principles, clinician responsibility and claim discipline — responsible AI.

Evaluate

Reading about defects is not the same as reproducing them.

Request a research demonstration Explore research collaboration