Tech stacks
Each form is implemented across several stacks. The links go to the per-stack agent docs in the monorepo.
Front-end with HTML / Lily Design System headless
Each form ships a single, consolidated front-end-with-html/ — one directory whose index.html is the single-page, step-by-step wizard and dashboard.html is the vetting
dashboard, sharing one css/ and js/. The scoring engine (js/types.js → js/rules.js → js/grader.js → js/flags.js, sometimes under domain-specific
filenames) is pure JavaScript, loaded as native ES modules. The Lily Design System HTML headless
library defines the markup/class contract every form's index.html/dashboard.html satisfy —
theme, locale, text-size, and share controls in the page header, and a vendored multi-theme CSS
catalogue under css/themes/.
Front-end with SvelteKit / Lily Design System Svelte headless
Each form also ships a single, consolidated front-end-with-svelte/ — one SvelteKit app that
serves both the wizard and the dashboard from RESTful, resource-oriented routes nested under src/routes/<slug>/, sharing one scoring engine (src/lib/engine/) and UI component set
(src/lib/components/ui/). Grading is a pure TypeScript pipeline (types.ts → *-rules.ts → *-grader.ts → flagged-issues.ts) with Vitest unit tests. The Lily
Design System Svelte headless library — five vendored helper components (ThemePicker, LocalePicker,
TextSizePicker, SharePicker, DateTimePicker) plus a 45-theme CSS catalogue under static/themes/ —
defines the component contract.
Read the SvelteKit stack doc →
Back-end with Rust / axum / axa
Each form has a Rust back-end JSON API built on Axum and axa — no HTML rendering layer at all: no Tera templates, no HTMX, no Alpine.js, no Lily Design System, no CSS, no static assets. The schema is relational and per-table: one SQL migration, one sqlx model, and one RESTful scaffold controller per SQL table (never a single table with a JSONB blob), running against PostgreSQL.
The Rust scoring types mirror the front-ends' TypeScript/JavaScript types via serde(rename_all = "camelCase"), so every implementation stays aligned at its JSON boundary.
SQL migrations
Each form has PostgreSQL migrations in Liquibase SQL format under sql/ — the source of truth
every generated representation derives from. Filenames follow NN_create_table_<name>.sql,
with COMMENT ON TABLE and COMMENT ON COLUMN statements for every column. Primary keys are
UUIDv4 via gen_random_uuid(), and every table has created_at, updated_at, and deleted_at timestamps.
XML representations
An XML + DTD pair is generated per SQL table entity. The XML schema matches the SQL schema one-to-one, providing an alternative wire format for systems that prefer XML over JSON.
Read the XML representations doc →
FHIR HL7 R5 representations
Each SQL entity is also represented as a FHIR HL7 R5 JSON resource, validated against the official
HL7 validator_cli.jar in CI. This is the interoperability format expected by EHR systems and
health-information exchanges that have standardized on FHIR R5.
Protocol Buffers representations
A Protocol Buffers .proto schema is generated per SQL table entity, so a form's data can be
serialized, transmitted, or code-generated into any language the protobuf compiler supports.
Read the Protocol Buffers doc →
OpenAPI 3.1 representations
An OpenAPI 3.1 .yaml specification is generated per SQL table entity (plus one combined openapi.yaml per form), so a form's resources can be described, mocked, validated, and
code-generated by any OpenAPI-compatible tooling.