Mail Interaction Parity
Use this as the acceptance workflow for Mail behavior. The target is observable
interaction parity: the same intent, focus transition, visible state, keyboard
result, optimistic feedback, and recovery path. Do not claim product parity from
static inspection alone.
Before testing
- Read
mail-backends, inbox-reads-and-triage, email-drafts, and
context-awareness before touching connected mail or application state.
- Run the app with deterministic synthetic mail when Gmail credentials are not
available. Synthetic mail is not a live Gmail mailbox.
- Open Mail and Superhuman side by side at the same viewport size. Record the
exact Superhuman build/date and the Mail commit under test.
- Seed or choose disposable messages that cover: one message, multi-message
thread, unread, starred, labelled, draft, scheduled, attachment, HTML body,
failed mutation, and partial account coverage.
- Never send during the default pass. If a live round trip is explicitly
approved, use only addresses the current user explicitly allowlisted for
that run; use an exact user-approved subject/body and verify both Sent and
Inbox afterward. Do not persist private addresses in fixtures or docs.
Side-by-side protocol
For every case in references/interaction-matrix.md:
- Reset both products to the same starting state.
- Perform the numbered Superhuman steps and record the resulting URL, focused
element, visible rows, toast/banner, network state, and elapsed feel.
- Repeat the exact intent in Mail, including the same alternate input paths
(mouse, keyboard, touch, drag, paste, blur, reopen, refresh, and failure).
- Mark
match, partial, or gap. A partial requires a concrete next
action; a gap must name the missing product capability or state.
- Re-run the case after each fix, then run its neighboring cases. A fix is not
complete until success, cancellation, loading, empty, error, rollback, and
refresh/deep-link states remain distinguishable.
Use stable Mail landmarks where present: [data-mail-email-row],
[data-thread-key], #mail-search, #mail-search-suggestions,
[data-search-item], [data-mail-compose], and the application-state keys
described in context-awareness.
Safety and truthfulness
- Draft, queue, and inspect by default. A click, shortcut, or agent action that
sends is an external side effect and needs explicit approval.
- Do not use random recipients, generated addresses, or third-party inboxes for
round-trip tests. The user's per-run allowlist is hard, not a suggestion.
- Treat a local fallback as local. Treat a failed or incomplete account read as
an error/partial result, never as an empty or successful mailbox.
- Do not report “bug free” or “1:1” globally. Report the tested case IDs,
evidence, and remaining gaps.
Verification
Run focused unit/action tests first, then a browser pass at 1440×900, 1024×768,
768×1024, and 390×844 when the browser harness is available. Check mouse,
keyboard, touch, focus, loading, empty, error, undo, deep link, and refresh.
Capture console errors and failed requests. For action changes, call the action
and read the result back. For UI changes, exercise the exact broken path in a
real browser or clearly record why the browser was unavailable.
The exhaustive cases are in
references/interaction-matrix.md; official comparison anchors are linked
there so the behavior baseline can be revisited when Superhuman changes.
1---2name: mail-interaction-parity3description: >-4 Exhaustively test Mail against Superhuman Mail interaction-by-interaction.5 Use when changing Mail UX, validating keyboard/gesture/autocomplete behavior,6 or running a side-by-side parity pass.7metadata:8 source: https://skillmd.com/skills/builderio/mail-interaction-parity9---1011# Mail Interaction Parity1213Use this as the acceptance workflow for Mail behavior. The target is observable14interaction parity: the same intent, focus transition, visible state, keyboard15result, optimistic feedback, and recovery path. Do not claim product parity from16static inspection alone.1718## Before testing19201. Read `mail-backends`, `inbox-reads-and-triage`, `email-drafts`, and21 `context-awareness` before touching connected mail or application state.222. Run the app with deterministic synthetic mail when Gmail credentials are not23 available. Synthetic mail is not a live Gmail mailbox.243. Open Mail and Superhuman side by side at the same viewport size. Record the25 exact Superhuman build/date and the Mail commit under test.264. Seed or choose disposable messages that cover: one message, multi-message27 thread, unread, starred, labelled, draft, scheduled, attachment, HTML body,28 failed mutation, and partial account coverage.295. Never send during the default pass. If a live round trip is explicitly30 approved, use only addresses the current user explicitly allowlisted for31 that run; use an exact user-approved subject/body and verify both Sent and32 Inbox afterward. Do not persist private addresses in fixtures or docs.3334## Side-by-side protocol3536For every case in `references/interaction-matrix.md`:37381. Reset both products to the same starting state.392. Perform the numbered Superhuman steps and record the resulting URL, focused40 element, visible rows, toast/banner, network state, and elapsed feel.413. Repeat the exact intent in Mail, including the same alternate input paths42 (mouse, keyboard, touch, drag, paste, blur, reopen, refresh, and failure).434. Mark `match`, `partial`, or `gap`. A `partial` requires a concrete next44 action; a `gap` must name the missing product capability or state.455. Re-run the case after each fix, then run its neighboring cases. A fix is not46 complete until success, cancellation, loading, empty, error, rollback, and47 refresh/deep-link states remain distinguishable.4849Use stable Mail landmarks where present: `[data-mail-email-row]`,50`[data-thread-key]`, `#mail-search`, `#mail-search-suggestions`,51`[data-search-item]`, `[data-mail-compose]`, and the application-state keys52described in `context-awareness`.5354## Safety and truthfulness5556- Draft, queue, and inspect by default. A click, shortcut, or agent action that57 sends is an external side effect and needs explicit approval.58- Do not use random recipients, generated addresses, or third-party inboxes for59 round-trip tests. The user's per-run allowlist is hard, not a suggestion.60- Treat a local fallback as local. Treat a failed or incomplete account read as61 an error/partial result, never as an empty or successful mailbox.62- Do not report “bug free” or “1:1” globally. Report the tested case IDs,63 evidence, and remaining gaps.6465## Verification6667Run focused unit/action tests first, then a browser pass at 1440×900, 1024×768,68768×1024, and 390×844 when the browser harness is available. Check mouse,69keyboard, touch, focus, loading, empty, error, undo, deep link, and refresh.70Capture console errors and failed requests. For action changes, call the action71and read the result back. For UI changes, exercise the exact broken path in a72real browser or clearly record why the browser was unavailable.7374The exhaustive cases are in75`references/interaction-matrix.md`; official comparison anchors are linked76there so the behavior baseline can be revisited when Superhuman changes.