Skip to content

usertest: typed answers when the mic doesn't work (#199) - #202

Merged
michielbdejong merged 1 commit into
mainfrom
claude/usertest-typed-answers
Sep 29, 2026
Merged

michielbdejong merged 1 commit into
mainfrom
claude/usertest-typed-answers

Conversation

@michielbdejong

Copy link
Copy Markdown
Contributor

Closes #199.

  • A typed-answer box under the controls: Enter sends, Shift+Enter starts a new line. Voice stays the default.
  • Falls back to typing when there's no microphone (missing, broken or refused), no speech recognition, or the mic is lost mid-session. The screen is then recorded without sound. Before this, the session couldn't start in those cases.
  • Start order: the page now asks for screen and mic before it creates the session, so a refused screen share no longer leaves an empty session folder.
  • The moderator gets typed turns marked [Tester, typed], and script.md tells it never to ask a typing tester to speak up.
  • Recording: transcript.jsonl records input: typed|voice per turn; meta.json records input and typedTurns. analyze.mjs labels typed turns TESTER (typed).

Tested locally:

  • the moderator with a dummy key and a local stand-in for the Claude API (no real key);
  • the page in Claude's built-in browser with faked screen sharing: a typed-only start without a mic.

Not verified: real speech recognition, real Chrome screen sharing, and deployment to the droplet (not included).

The Dutch support (#198) is stacked on this branch and follows as its own PR.

🤖 Generated with Claude Code

Voice stays the default. Under the controls the session page now has a
box for typed answers (Enter sends, Shift+Enter a new line). A typed turn
goes to the moderator like transcribed speech, marked "[Tester, typed]",
and is recorded with input: "typed" in transcript.jsonl; meta.json gets
`input` (how the session started) and `typedTurns`.

Without a microphone (none, broken or refused) or without speech
recognition, the session now starts anyway, typed only, recording the
screen without sound; a recognizer that loses the microphone mid-session
switches to typing too. Media is requested before the session is created,
so a refused screen share no longer leaves an empty session behind.

script.md tells the moderator not to ask a typing tester to speak up;
analyze.mjs labels typed turns for the analysis.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@michielbdejong
michielbdejong merged commit 6d39e10 into main Sep 29, 2026
18 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

in the user testing app, allow the user to type in case their microphone doesn't work

1 participant