Write amounts and dates one way, and re-render only what changes (app review batch H)

[EmotiveImpact]

Before: US dollar communes showed "US$32.10". Dates came in several styles: most screens said "9 Oct", but the expense window said "Due 2026-10-06", Settle and pledges "due 2026-10-06", notices and fund contributions "09/10/2026", the quick finder "Due 2026-11-04" and "Task · 2026-10-20", and reminder and receipt dates always carried the year. Each amount built its own number formatter, and money was formatted in five places (the email worker included). The whole app shared one context, so in the render-count test a minute check that found nothing new re-rendered 312 components, a letter typed in the Notes filter re-rendered the whole app (124 components), and every render rebuilt every commune's views from its records (about 280 ms for a busy year). A letter typed in the Expenses filter re-rendered every expense card (217 components for 40 expenses). packages/core was reached by relative paths, its tests ran nowhere, and two of its test files imported modules that do not exist.

After: amounts read "£12.50", "$32.10" and "€8.00", and every plain date reads "9 Oct", with the year when it is not this year ("6 Oct 2025"), the finder included. One cached formatter writes every amount and one writes every date, and reminder emails write amounts the same way. Typing in the Expenses or Notes filter re-renders only that screen (5 components); in Chromium with the CPU slowed four times, a keystroke in the Notes filter reaches the screen in 16 ms or less, against about 200 ms before (480 ms with a year of expenses loaded). A minute check that finds nothing new re-renders nothing, and a commune's views are built once per read of it: a busy year now takes about 11 ms, 25 times faster. packages/core is imported as @commune/core, and its tests run in CI with the web tests.

This is app review batch H, built on #33 (tidying), so review that first; the diff here shows only this part. Besides the money and date text listed above, the one visible change is that the Notes filter clears when you leave Notes, as the Expenses filter always has.

What changed, by finding

Finding	Change
50	formatMoney in packages/core/src/format.ts makes one formatter per currency and keeps it. The three currencies the app offers use their narrow symbol; any other keeps the code-prefixed form ("CA$", "A$") so it cannot be mistaken for one of them. Every amount in the web app goes through it: the live model, the expense window, Settle, Expenses, search results, funds, receipts, announcements, task history and the activity log. The reminder worker keeps a copy (apps/reminder-worker/src/money.ts, since it deploys on its own and imports nothing from outside its folder), and a test checks the two agree.
47	formatDay writes "9 Oct", with the year when it is not this year, and never moves a date across a time zone. The expense window, Settle, notices, pledges, fund contributions, receipts and reminder days use it. Migration 0040 writes search_commune_records out in full so each result also carries label (its words without the date) and date (YYYY-MM-DD or null); the finder writes "Due 4 Nov" and "Task · 20 Oct" from them, and shows the old text if the database has no 0040 yet. The old meta text stays for clients released before this one.
17	useApp() keeps its shape, so no screen changed how it reads state, but behind it is now a store: each component re-renders only when a field it read has changed. A minute check that finds nothing new keeps the objects it already had. Each commune's views are built once per read (readyPlace), and rebuilt when the date changes, so "Overdue" moves on at midnight. The Notes search text now lives in Notes (Expenses' already did), so like the Expenses filter it clears when you leave the screen or switch commune; Search keeps its own words and the app keeps a copy so browser Back finds the same search, and the expense and note cards are memoised. Commands keep their identity between renders and read the latest records when they run.
64	packages/core is imported as @commune/core through a path in apps/web/tsconfig.json and an alias in vite.config.ts and vitest.config.ts. It is not a workspace package, because that needs a lockfile change, which the network here blocks. Its tests now live in packages/core/tests and run in Node in the web test command, so CI runs them. The two test files that imported missing modules and an unused onboarding module are deleted, and the expense split tests moved into core and check that every split adds up. Batch 3 had already made readMoneyInput the one parser, so Add expense, corrections, funds, pledges, receipts, recurring bills and saved splits accept and refuse the same amounts; money.test.ts pins them.
65 (from #33)	equalShares, an unused copy of core's equal split, is deleted.

Not in this pull request: the minute check's download size (finding 18) and drawing every expense card (finding 19). Other date styles stay as they are, because they say something else: day headings in Activity ("Thursday 9 October"), the Hub week list, task and month headings, and times of day.

Releasing

Nothing hosted was touched. Migration 0040 follows 0039 and only adds two fields to search results, so the previous client keeps working. This client works with or without it: without 0040 the finder shows the old date text. Apply it and refresh the Data API schema cache. meta can be dropped by a later migration once no released client reads it.

Needs the owner

• Say when to drop meta from search results: once no released app version reads it.

How it was checked

On this branch:

• 642 web tests in 64 files, packages/core's among them; the web, local API and worker typechecks; and a production build.
• 28 database tests through run-local.sh (2 more need a Neon branch), one of them new for 0040, then dump-schema.sh --check.
• 22 local API suites on a disposable database and port, one of them new (search dates), and the reminder worker's 7 unit tests and 4 database suites.
• The secret scan over the files and every commit.

In Chromium, production builds of #33 and of this branch (pointed at the local API) ran against one disposable database: the three sample communes, in pounds, euros and dollars, and a household with 800 expenses over a year.

• The text listed above changed and nothing else did: "$32.10" on Studio East's Expenses, Settle and Home (was "US$32.10"); "Due 29 Sept" in the expense window; "due 6 Oct" on Settle and in the fund window; "9 Oct" on notices and contributions (was "09/10/2026"); "Due 11 Oct" and "Event · 18 Oct" in the finder. September reads "Sept", as the app's existing dates already write it. Pound and euro amounts read as before.
• The Expenses filter and its chip, the Notes filter and full search's words survive opening and closing a window, and browser Back returns to the same search.
• Typing, with the CPU slowed four times: three alternating runs of each, timed from key press to the next frame painted.

Typing in	Loaded	Before: typical, slowest	After: typical, slowest
Notes filter	on opening	208 ms, 352 ms	16 ms or less, 104 ms
Notes filter	a whole year	480 ms, 1,120 ms	16 ms or less, 80 ms
Expenses filter	on opening (84 cards)	24 ms, 312 ms	24 ms, 320 ms
Expenses filter	a whole year (800 cards)	56 ms, 2,152 ms	44 ms, 2,216 ms

The Expenses filter was already local to its screen, so it gains least: with a year loaded, the slowest tenth of keystrokes fell from 360 ms to 136 ms. Its slowest keystroke is clearing the filter with a year loaded, which draws all 800 cards again; that is finding 19, left for a later pull request.

How

The store holds the state the provider computes on each render. useApp() returns a view of it that notes which fields a component reads, and React's useSyncExternalStore re-renders that component only when one of those fields has changed. Derived values keep their objects while what they are built from is unchanged, so a field changes only when what it shows does. infra/neon/schema.sql was regenerated with 0040; its diff shows the whole change to search_commune_records.

🤖 Generated with Claude Code

https://claude.ai/code/session_01A6P7aTcMBccooYYr4xYnVS

---

Generated by Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
commune-web	Error		Oct 9, 2026 9:11pm UTC

[EmotiveImpact]

CI did not run on this commit. All three jobs (Web app; Database, local API and worker; Secret scan) were marked failed two seconds after they were queued, before a runner was assigned: no step ran and there are no logs. The workflow file is the same as on #33, where the same jobs passed at 20:10 UTC. GitHub fails jobs this way when the account cannot start Actions, for example when the month's included minutes or the spending limit are used up; the push runs on main in April failed the same way. It could also have been a passing GitHub fault. I can't re-run jobs from here (the API returns 403), so this needs the owner: press "Re-run all jobs" on the run, and if the jobs fail again within seconds, check Actions usage under Settings, Billing and plans.

On this commit, everything CI runs passes locally: 642 web tests, the web, local API and worker typechecks, the production build, 28 database tests and the schema check, 22 local API suites, the worker's tests, and the secret scan over the files and the whole history.

The Vercel preview error is the account issue already noted on #24 to #33, not this change.

---

Generated by Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-12T02:02:22.034338Z	dc69a71	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.