#731 Display And Store Phone Numbers In One Format

[b-at-neu]

Closes #731

Summary

• Phone numbers now display consistently everywhere via a shared, tolerant formatter, and new phone_number answers are stored normalized. No backfill — existing rows render unchanged through the tolerant path.
• Hand-rolled digit-strip-and-mask, not libphonenumber-js — confirms the ticket's recommendation, since the validator is already deliberately loose and the masking rule is US-only by design.

Changes

• lib/constants.ts — added normalizePhoneNumber, formatPhoneNumber, and the two format-dispatching helpers normalizeShortAnswerValue / formatShortAnswerValue, placed beside matchesShortAnswerFormat.
• lib/types.ts — added format: ShortAnswerFormat | null to ApplicationReviewAnswer (AnswerQuestion already had it).
• prisma/data/applications.ts — selects format through the live question relations (including for orphaned/soft-deleted questions) and maps it into all four branches of normalizeApplicationAnswers.
• prisma/actions/profile.ts (updateGlobalAnswer) and prisma/actions/applications.ts (createOrUpdateApplicationAnswer) — the write path now runs each short-answer value through normalizeShortAnswerValue instead of a bare .trim(). Runs strictly after existing validation; the profile-copy branch for file_upload answers is untouched.
• components/features/answer-display.tsx and components/features/application-answers-list.tsx — the single-value and chip-list short-answer branches render through formatShortAnswerValue, covering both the profile view and the reviewer/applicant answer list.
• tests/unit/constants.test.ts — unit coverage for normalizePhoneNumber/formatPhoneNumber (10-digit, 11-digit leading 1, already-normalized, + international, 00 international, 7-digit local, a validator-passing value that fits no mask — each round-tripped normalize → format) and for the dispatchers' identity behavior on other formats.
• tests/db/phone-number-normalization.test.ts (new) — confirms updateGlobalAnswer and createOrUpdateApplicationAnswer both store a digits-only value for a punctuated input.
• docs/WORKFLOWS.md — one clause each in AP-2, AP-6, AP-11, PM-9 describing the storage/display rule.

Testing plan

• Admin: set a global question's format to Phone number. As an applicant, answer it (617) 555-0100 on /profile, reload — the read-only card shows (617) 555-0100.
• Confirm the stored value is 6175550100 (digits only) via the DB or the edit field.
• Answer a second phone question +44 20 7123 4567 — stored +442071234567, displayed unchanged as +442071234567.
• Manually insert/edit a legacy-shaped row (555-123-4567) directly in the DB — the profile card and the reviewer view both show (555) 123-4567, and the row itself is untouched.
• Apply to a position with a phone-formatted position question; save an answer, submit, then open the application as a manager at /manage/applications/<id> — the answer renders masked in the answers list.
• Delete that position question (soft delete), reload the reviewer view — the orphaned answer still renders masked, below the live questions.
• Enter abc in a phone field — still Enter a valid phone number inline, nothing saved (unchanged behavior).
• Check an email, url, and zip_code question and a long answer — rendering byte-identical to before this change.
• A multi-value legacy short_answer row (more than one value, format phone_number) — each chip formats independently.

Automated checks

• npm run prettier:check — pass
• npm run eslint:check — pass
• npm run tsc:check — pass
• npm run test:unit — pass (77/77 in tests/unit/constants.test.ts; full unit suite 423/423, one pre-existing unrelated failure in tests/unit/email-delivery-events.test.ts due to no local DATABASE_URL)
• tests/db/phone-number-normalization.test.ts added but not run locally — no Postgres available in this worktree (DATABASE_URL unset, and the local db:start port is already occupied by an unrelated container). Will run in CI's run-tests check.

Notes

• The edit input (AnswerField) deliberately keeps showing the raw stored value while the read-only card shows the mask — formatting a controlled autosave input would desync the stepper's dirty/diff tracking for a cosmetic gain. type="tel" and the validator are unchanged.
• findDivergingGlobalAnswers is string-exact, so re-saving an application phone answer can flag a purely cosmetic divergence against an untouched legacy profile value. Cosmetic only — "Revert to profile answers" still writes the correct profile value; not worth normalizing the copy path, which would be a backfill by the back door.
• The 00 → + promotion is the one rule the ticket left implicit — it follows from the validator treating both as the same prefix, and is covered by a round-trip unit test.

[vercel]

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

Project	Deployment	Actions	Updated
aplio	Ready	Preview	Sep 25, 2026 2:19pm UTC

[b-at-neu]

Code Review — Cycle 1 · needs revision

1 open — 1 🟡 Low

R1-L1 🟡 Low — AnswerMismatchNotice renders orphaned short_answer values raw (no formatShortAnswerValue), so a legacy multi-value phone_number row shows unmasked extra digits in the mismatch notice — the one render surface the ticket's "verified by grep" acceptance criterion missed. Fix: run values through formatShortAnswerValue/pass format down from AnswerDisplay.

aplio/components/features/answer-mismatch-notice.tsx

Lines 39 to 45 in 6b29bf0

	<ul className="mt-1 flex flex-col gap-0.5">
	{values.map((v, i) => (
	<li key={i} className="font-medium">
	{v}
	</li>
	))}
	</ul>

[b-at-neu]

Revision — Cycle 1

fixed R1-L1 · 2db1cd3

[b-at-neu]

Code Review — Cycle 2 · approved

1 open — 1 ⚪ Nit (see inline)

[b-at-neu]

Revision — rebase

rebase: docs/WORKFLOWS.md (merged independent insertions), tests/unit/constants.test.ts (merged independent import additions) · 8ff9988

[b-at-neu]

Revision — Cycle 3

fixed R2-N1 · b015b01

R3-L1 investigated: no code-level cause found. The diff has no schema/migration/route changes, no env-dependent code, and nothing edge-runtime-incompatible. The failed deployment (dpl_8L29hqQgchXrAciyJaKx4NCA45WB, commit 8ff9988) predates this branch's merge from dev; the very next deployment on the same PR (merge commit d77c17d, functionally identical app code) completed successfully, and the Vercel build log itself is human-only / vercel CLI is deny-listed for agents. Given the identical code redeployed clean, this reads as a transient build/infra hiccup (likely in vercel.json's preview-only prisma migrate deploy step against the Neon preview branch) rather than something this PR introduced.

[b-at-neu]

Code Review — Cycle 4 · approved

0 open — clean

[b-at-neu]

Revision — Cycle 4

rebased onto origin/dev · e3c1561 · rebase: docs/WORKFLOWS.md (union merge — both sides added distinct non-overlapping sentences to the same AP-11/PM-9 paragraphs; kept both)

[b-at-neu]

Code Review — Cycle 5 · approved

0 open — clean