spike: @adobe/data adoption evidence (chat-state container port)

[2witstudios]

⚠️ Evidence artifact — NOT for merge

This branch exists to answer the @adobe/data adoption spike with running code. Nothing here is imported by any production call site; apps/web/src/app/spike/adobe-data/ is a throwaway harness route. Do not merge. Close it once the migration epic is scoped, or cherry-pick src/state/chat/* into that epic's PR A.

Full verdict + migration epic proposal: spike page q9ysdsvdada5c6jp1fq0abpj.

Verdict: ADOPT — after E1, at the E1→E2 boundary

No blocker found. Bindings, SSR, hydration, StrictMode, undo and zustand coexistence all work under Next 15 + React 19, with zero any / as any / @ts-ignore and a clean monorepo typecheck. Not "adopt now" only because E1 PR5/PR6 are in flight on zustand and the rails forbid a mid-wave platform swap.

Versions: @adobe/data@0.9.83, @adobe/data-react@0.9.83.

The decisive proof

apps/web/src/state/chat/__tests__/conversationMessagesFacade.test.ts and pendingStreamsFacade.test.ts are mechanical copies of the two existing store suites with exactly one line changed each — the container construction in beforeEach. Every it() title, every arrangement, every assertion is byte-identical.

49/49 pass. The 133 pure applyX tests are untouched and green, because the transactions call those same functions.

bun vitest run src/state/chat src/stores
→ 55 files, 817 tests, 0 failures

What is in the port

File	Role
state/chat/chat-data-plugin.ts	components / archetypes / indexes — one entity per conversation, one per live stream
state/chat/chat-state-plugin.ts	applyX transitions as transactions, selectors as computed (db.derive), value-returning glue as actions
state/chat/conversationEntry.ts	the container swap in one function: read entity → run the untouched pure applyX → write columns
state/chat/facade/*	zustand-shaped facades, so existing getState() call sites are unchanged over the new container
app/spike/adobe-data/*	RSC → DatabaseProvider → useObservableValues harness

Extra evidence suites: aiActionUndo (5), computedObservables (9), zustandInterop (3), reactBinding (7 — SSR, StrictMode, propagation, unmount, hydration).

Measurements

• Shared First Load JS: 237 kB → 237 kB. No change to the shared baseline; the library is route-scoped and tree-shaken out of routes that don't import it.
• @adobe/data client chunk: 52.4 kB raw / 16.9 kB gzip. Ported plugin + facades + harness page: 17.3 kB raw / 5.0 kB gzip.
• Whole client build: +71.7 kB (+0.20 %) — and that includes the throwaway harness route.
• bun run typecheck (Turbo, monorepo): clean. eslint on all new files: clean.

Friction the migration epic must carry

1. computed and actions cannot share one Database.Plugin.create call (0.9.83) — declaring both silently collapses inferred actions to {}. Splitting across an extends boundary fixes it. Undocumented; dictates plugin layout for every adopter.
2. Database.Plugin.ToStore<P> drops declared indexes — use ToTransactionContext<P> / Database.Read<ToDatabase<P>>.
3. Transactions may only return void | Entity, so startLoad's "return the generation" contract becomes an action.
4. Undo/redo requires the plugin to declare its own applyOperations transaction — runtime throw, nothing in the types says so.
5. Optional fields have no ECS equivalent (null-encoded columns projected back to undefined).
6. Opaque JS payloads need mutable: true or ToType wraps them in DeepReadonly.

Verification gaps (stated honestly)

• React render/hydration tests did run in the .pu worktree — the epic's documented dual-React failure did not reproduce here (@adobe/data-react resolves React from apps/web/node_modules, the same copy vitest uses). Re-run on the main checkout before relying on it.
• /spike/adobe-data was not loaded through a live dev server: it sits behind the auth middleware, and a spike must not modify an auth boundary to make itself reachable. next build compiles it (route listed at 21.8 kB) and SSR/hydration is asserted in tests instead. A signed-in manual load remains the one open gap.

🤖 Generated with Claude Code

https://claude.ai/code/session_01B1feUaxUzUP3Z6oDLTtFDd

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 56fd8420-96c2-4e28-9a00-1ddac496ca1d

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch pu/adobe-data-spike

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}