I investigate software problems and build tools that keep accepted work, proposed changes, and test evidence traceable.
A failed edit should leave the last accepted document available. A software correction should come with a test and a record another person can inspect. Those are practical problems behind my work in Estra Logics.
See a demo preview · Inspect the portfolio · Contact me
Microsoft Semantic Kernel #14454 — merged October 27, 2026.
- Problem: constrained Pydantic fields could put constraint metadata where a text description belonged, preventing JSON serialization.
- Change: preserve field descriptions and compatible metadata fallbacks, with regression tests for both.
- Result: the correction was merged upstream. The pull request preserves the source, review, and contribution history.
| Start with | What you can inspect | Stage |
|---|---|---|
| CML Local Alpha | Validate a document, reject a broken edit, and retain its last accepted state. Download alpha.2. | Alpha; static document checks |
| Estra Logics project guide | Glasshouse preview, CML Trust, Pattern Atlas, and a route into each project. | Project directory |
| Engineering Work Record | Dated investigations, validation results, corrections, and handoffs. | Evidence archive |
I'm S.J., the independent creator behind Estra Logics, the platform. CML — Continuity Markup Language — is its underlying language and engine. It describes accepted state and proposed changes alongside the host software that performs the work.
Fermata is the working method: observe, check the evidence, pause when a condition is unresolved, and continue when the conditions are met.
My interests include mathematical patterns, harmonics, music and visual continuity, and historical NASA programs. Research observations stay separate from interpretations and verified software results.
I guide the questions and direction, with AI as a technical partner. My work spans software investigation, validation tooling, engineering production, and technical documentation. Discuss a project.
Human-led · AI-assisted · Evidence-bound


