Web codebase split — apps/web → apps/web-snow + apps/web-summaryto

Web codebase split — deferred task

Status: Deferred to Phase 4 of the summary.to execution plan.

Trigger to execute: summary.to needs an independent release cadence — i.e. when shipping a summary.to change is blocked, slowed, or risk-coupled to SNOW releases (or vice versa). Until then, the single apps/web codebase with host-routed sub-apps is sufficient.

Current state

Surface	Brand	Service	Codebase
`platform.snowmed.health`, `app.snowmed.io`, `snowmed.health`	snowmed	snow_web	`apps/web`
`summarisenow.snowmed.health`	summarisenow	snow_web	`apps/web`
`*.summary.to`, `summary.to/Dr.X`	summary_to	snow_web	`apps/web`

SNOW is the engine (internal), not a customer-facing brand. The B2B brand is snowmed.

All three brands deploy out of one Vercel project. Host-routing branches at the React root into per-brand subtrees.

Target state (when triggered)

Surface	Brand	New service	New codebase
`platform.snowmed.health`, `app.snowmed.io`, `snowmed.health`, `summarisenow.snowmed.health`	snowmed, summarisenow	snow_web	`apps/web-snow`
`*.summary.to`	summary_to	summaryto_web	`apps/web-summaryto`

Two Vercel projects. Independent CI, bundles, cache, CSP/headers.

Why defer

Reason to do it now	Counter
Cleaner brand isolation	Already achieved via host-routed subtrees
Smaller bundles	Not yet a measured problem
Independent deploys	Not yet a release-cadence conflict
Easier carve-out for entity split	Carve-out optionality preserved via brand-tagged P&L; physical split not required yet

Reason to defer	Cost if done now
Duplicate auth wiring	~1 week eng
Shared component package needed	Refactor `apps/web/src/components` → `packages/ui-core`; ~1 week
×2 CI/CD pipelines	Ongoing maintenance + slower local DX
×2 Vercel cache, ×2 cold-start budget	Cost + perf surface area

Pre-work (do now, while still one codebase)

To make the future split a near-mechanical lift:

Branch on host at React root: <SnowApp /> / <SumNowApp /> / <SummaryToApp /> subtrees with their own routers and layouts.
Move shared primitives (forms, file upload, auth hooks, theme) to packages/ui-core even within the monorepo — no ../../web/src imports across subtrees.
Brand-tag every emitted event (brand axis, distinct from service axis). See brand-vs-service taxonomy.
Independent visual systems per brand: no shared CSS variables or design tokens leaking across subtrees.
Per-brand environment-variable convention: VITE_SNOW_*, VITE_SUMNOW_*, VITE_SUMMARYTO_*.

Execution plan (when triggered)

Create apps/web-summaryto from the <SummaryToApp /> subtree. Pull only the imports it actually uses.
Lift shared deps into packages/ui-core if not already.
New Vercel project snow-web-summaryto; bind summary.to apex + all *.summary.to subdomains.
Rename existing Vercel project’s domain bindings to drop *.summary.to.
Update API CORS allowlist to source-of-truth (no per-codebase config).
Cut over via DNS — both projects can serve in parallel for a soak period.
Drop the <SummaryToApp /> branch from apps/web once traffic confirms.

Definition of done

summary.to and *.summary.to resolve to apps/web-summaryto Vercel project
apps/web no longer contains summary.to-specific code paths
CI builds both projects independently
No regression in auth, upload, or share flows on either brand
One full release cycle each ships independently without coordination