FE-1839: Shorten and de-jargon Voice and Brunch panel copy

[kostandinang]

🌟 What is the purpose of this PR?

Make the Voice and Brunch panel text short and plain. The voice consent card was a dense paragraph of internal terms, and Voice notices said things like "Check canonical history; no automatic retry was made." This continues #9886's activity wording, which stopped naming Brunch in turn labels, across the rest of the panel.

🔗 Related links

• FE-1839 (internal)
• FE-1793 (internal)

🚫 Blocked by

• #9886, branch ka/fe-1793-merged-stack-fixes.

Stack: main → #9883 → #9886 → this PR.

🔍 What does this change?

Copy only; no behaviour changes.

• Voice consent card: the same disclosures in plainer words. "OpenAI processes your microphone audio and the conversation text. Your transcribed words, the request sent to the assistant and its answer are saved. Audio isn't saved, and captions stay in this browser."
• Voice issues popover: the popover shows each message's first sentence as its title, so every notice now leads with a short, complete sentence, such as Your earlier words weren't sent. or Couldn't confirm your message was sent. Notices no longer say "admission", "canonical history", "utterance" or "Live", and say "the assistant" instead of "Brunch".
• Voice status bar: the blocked-audio status reads Audio blocked. Select Play to listen.
• Connection errors: the ten Live errors under Technical details keep their technical first sentence and drop the repeated "No automatic retry was made."
• Audio settings, Labs settings, Ledger notices, Activity tool rows, sweep results and the experiment draft card: shorter wording without "settled", "canonical read" or "Ledger revision". For example, Activity reads Reading confirmed passages and Checking for Ledger updates.
• Naming: inside the assistant panel, copy says "the AI assistant" instead of "Brunch". Labs settings keep Use Brunch, where the name tells the two assistants apart.
• Unchanged: instructions sent to the Live model, server API errors and prompt chips.

Every change, before and after

Where	Before	After
Consent card, Live	OpenAI processes microphone audio for voice and transcription, and text to prepare briefs and summaries. Brunch saves your finalized spoken words, the brief and its answer. Spoken captions stay in this browser; Petrinaut does not save audio.	OpenAI processes your microphone audio and the conversation text. Your transcribed words, the request sent to the assistant and its answer are saved. Audio isn't saved, and captions stay in this browser.
Consent card, Realtime	OpenAI processes live audio and speaks the interviewer's words. Petrinaut saves finalized answers—not audio.	OpenAI processes your audio and speaks the interviewer's questions. Your transcribed answers are saved; audio isn't.
Test microphone failure	Microphone access was not available. Check your browser permissions and try again.	Couldn't access the microphone. Check your browser permissions.
Words cancelled by new speech	Your earlier utterance was not sent because you started speaking again. Use the composer to send it.	Your earlier words weren't sent. You started speaking again, so they're in the composer if you still want them.
Words cancelled by a conversation switch	Your earlier utterance was not sent because the conversation changed. Use the composer to send it.	Your earlier words weren't sent. The conversation changed, so they're in the composer if you still want them.
Words dropped while busy	That utterance was not retained. Wait for the pending input, then use the composer to send it.	Those words weren't sent. They're in the composer to send when the assistant is ready.
Answer not confirmed	Your message was admitted, but its response could not be confirmed. Check canonical history; no automatic retry was made.	Couldn't confirm the answer. Your message was sent; check the conversation before sending it again.
Send not confirmed	Voice admission could not be confirmed. Check canonical history before sending again; no automatic retry was made.	Couldn't confirm your message was sent. Check the conversation before sending it again.
Assistant didn't finish	Brunch did not complete this turn. Check the conversation; no result was offered to Live.	The assistant didn't finish. Nothing will be spoken; check the conversation.
No reply to speak	Brunch settled without a spoken answer. Check the conversation.	The assistant finished without a reply to speak. Check the conversation.
Summary failed	Brunch finished, but its spoken summary could not be prepared. Read the written answer; no automatic retry was made.	Couldn't prepare a spoken summary. The written answer is in the conversation.
Spoken answer not delivered	The answer could not be sent to Live locally / was rejected by Live. Check the conversation; no automatic retry or replay was made. Acceptance does not confirm playback.	Couldn't speak the answer. The written answer is in the conversation.
Instruction to Voice not delivered	The continuation instruction could not be sent to Live locally / was rejected by Live. (Same ending.)	Voice may be out of sync. Check the conversation before relying on what it says.
Blocked audio status	Audio playback is blocked. Select Play voice audio to hear Live.	Audio blocked. Select Play to listen.
Live connection errors (10)	e.g. Transcription failed. No fallback or automatic retry was made.	e.g. Transcription failed.
Realtime: answer pending	Previous answer waiting. Please try again after it is sent.	Still sending your last answer. Try again in a moment.
Realtime: not heard / too long	We didn't catch that. Please try again. / That answer is too long. Please try a shorter response.	We didn't catch that. Try again. / That answer is too long. Try a shorter one.
Microphone permission / device	Allow microphone access in your browser settings, then reconnect voice input. / No usable microphone was found. Connect or select one, then reconnect voice input.	Microphone blocked. Allow access in your browser settings, then reconnect. / No microphone found. Connect or select one, then reconnect.
Voice service errors	…give the diagnostic reference to an operator. / …preview is unavailable or disabled. Continue with the text composer.	…share the diagnostic reference with support. / …is unavailable. You can keep going in chat.
Audio settings	Applies next session. Mute your mic while the agent is idle to preview. / Audio controls are unavailable until Voice is connected.	Applies next session. To preview, mute your mic while the agent is idle. / Connect Voice to use audio controls.
Labs settings	Brunch is unavailable because this site has no Brunch endpoint configured. / Make Voice mode available for Brunch conversations. / Switch to an alternative voice engine if you're having trouble with Live.	Brunch isn't set up on this site. / Talk to Brunch by voice. / Use the alternative voice engine if Voice isn't working well.
Ledger notices	A newer Ledger revision exists. Ask Brunch to refresh this view before relying on it. / The recorded explanation predates a newer Ledger revision. Ask why again to assess the latest account.	This view is out of date. Ask the AI assistant to refresh it. / This explanation is out of date. Ask why again for the latest.
Activity tool rows	Reading settled passages (and conversation sources) / Checking Ledger revision	Reading confirmed passages (and conversation sources) / Checking for Ledger updates
Sweep with nothing to do	No settled range to sweep / The conversation has no settled user entries.	Nothing confirmed to sweep yet / None of your messages in this conversation are confirmed yet.
Draft card	The model changed since the canonical read… / This draft was prepared before this editor was loaded… / Run is blocked by an unsupported restriction… a reporting-only exploration needs your explicit acceptance. / …or ask Brunch to draft it again.	The model changed since the AI assistant read it… / This draft is from an earlier session… / Can't run: a restriction isn't supported. Ask the AI assistant to revise the proposal, or tell it you accept a run that only reports on it. / …or ask the AI assistant to draft it again.
Draft preparation errors	Draft requires the latest settled canonical net read after changes. / The latest canonical read has no document revision.	The AI assistant needs to read the latest model before drafting. / The AI assistant's latest model read has no document revision.
Experiment follow-up tooltip	The experiment finished, but its results could not be sent to Brunch.	The experiment finished, but its results could not be sent to the AI assistant.
Activity: connection check	Checking Brunch connection / Could not reach Brunch	Checking connection to the AI assistant / Could not reach the AI assistant
Ledger region label (screen readers)	Brunch Ledger	Ledger

Pre-Merge Checklist 🚀

🚢 Has this modified a publishable library?

• Adds a patch changeset for @hashintel/petrinaut (audio settings copy).

📜 Does this require a change to the docs?

• No. The docs paraphrase the consent card, and that paraphrase is still accurate.

🕸️ Does this require a change to the Turbo Graph?

• No.

⚠️ Known issues

• The consent acknowledgement key (petrinaut:live-voice-interview-disclosure:v3) is not bumped, because the card makes the same claims, so people who already agreed won't be asked again. The new text says OpenAI processes "the conversation text", which is broader than the old "text to prepare briefs and summaries". Bump to v4 if product or legal want everyone to agree again.
• "Confirmed" stands in for the internal "settled" (passages agreed in the Ledger). Happy to match whatever word the Ledger uses.

🐾 Next steps

None planned.

🛡 What tests cover this?

• Updated: tests and one story that assert the old wording now assert the new wording. The Live warning test now checks the per-kind message.
• The website voice and Brunch suites (827 tests) and the Petrinaut assistant panel suite (317 tests) pass, apart from the known local localStorage failures that also happen on main.

❓ How to test this?

1. Run turbo run lint:tsc lint:eslint test:unit --filter @hashintel/petrinaut --filter @apps/petrinaut-website.
2. Start yarn dev:brunch, open the assistant and select Start voice mode. Check the consent card shows the new text, and that Test microphone reports in the new wording.
3. In Voice, start speaking again before a turn is sent and check that the Voice issues popover shows Your earlier words weren't sent. as the title.

[vercel]

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

Project	Deployment	Actions	Updated
hash	Ready	Preview	Oct 6, 2026 5:20pm UTC
petrinaut	Ready	Preview	Oct 6, 2026 5:20pm UTC
petrinaut-docs	Ready	Preview	Oct 6, 2026 5:20pm UTC

1 Skipped Deployment

Project	Deployment	Actions	Updated
hashdotdesign-tokens	Ignored	Preview	Oct 6, 2026 5:20pm UTC

[cursor]

PR Summary

Low Risk
User-visible strings and test assertions only; no logic, APIs, or auth/data paths changed.

Overview
This PR replaces jargon-heavy Voice and Brunch assistant copy with shorter, plain language. Behavior is unchanged; strings, tests, and Storybook expectations were updated together.

Voice consent, status, warnings, connection errors, and audio settings now say things like “the AI assistant” instead of “Brunch,” drop terms such as “canonical history,” “admission,” “utterance,” and “Live,” and remove repeated “No automatic retry was made” tails from Live technical errors. Audio help is tightened (e.g. preview and “Connect Voice to use audio controls”).

Brunch panel surfaces get the same treatment: Labs settings, Ledger stale notices, Activity tool row titles (“confirmed passages,” “Checking for Ledger updates”), sweep empty state, experiment draft cards, and voice bridge notices all use simpler wording.

A patch changeset for @hashintel/petrinaut documents the audio settings copy change.

^{Reviewed by Cursor Bugbot for commit 74b433b. Bugbot is set up for automated code reviews on this repo. Configure here.}