INDEPENDENT SOFTWARE / DELIVERY & VERIFICATION
A small system.
Deliberate architecture.
This website demonstrates how I connect requirements, implementation choices and evidence. Its engineering experiments use original synthetic data.
HTML for reading. Workers for computation.
Astro renders the pages and writing. A small shared enhancement adds appearance preferences, search navigation and printing. The homepage switches between prerendered resilience states; it carries the selected scenario into the full Studio without loading the simulation worker.
The shared TypeScript model package keeps numerical behavior separate from presentation. Studio and Cooling Lab use native custom elements and dedicated browser workers. Control Room uses an isolated Solid interface for its controller studies. Model versions, bounded inputs and reproducible links keep calculations inspectable.
Content and experiments have different lifecycles.
Two Cloudflare Workers separate the main site and its CMS from Control Room. EmDash stores published writing in D1 and media in R2. Preview and production have separate data, media and session bindings. The private editing interface is separate from the public reading experience.
Local records, explicit sharing.
The optional Studio notebook stores validated inputs in this browser. It recomputes results, detects conflicting edits and merges reviewed imports transactionally. Public experiment links and result exports exclude private titles and observations; notebook backups can include them and remain the visitor's responsibility.
Verification follows the requirement.
Numerical checks cover energy balance, convergence and reproducibility. Runtime checks exercise publication, removal, discovery and API behavior. Browser checks cover keyboard interaction, mobile composition, unavailable storage and failed calculations. Build checks enforce separate budgets for initial pages and lazy experiment journeys.
Releases require successful quality checks for the selected commit. Preview smoke checks verify public routes, release identity and the deployed script graph, including workers and lazy imports. A production release uses successful preview evidence for that same commit.
Trade-offs worth making explicit.
- Editorial reading uses HTML and small shared controls instead of framework hydration.
- Advanced studies and exports load on demand to protect the first visit.
- The experimental Solid renderer is confined to Control Room.
- Published metadata powers search; the current bounded corpus does not need a separate search service.
- Code rollback and data recovery are separate operations. A successful upload alone does not establish application readiness.
Inspect the evidence.
The model excerpts come from the calculation source. The verification notebook records expected and computed results for its build. Neither is a claim of measured long-term service availability or a calibrated facility.
Open Resilience Studio →