The Unjournal · Pivotal Questions Initiative · Internal

CM Workshop Facilitation Guide

Zoom setup, tool options, and session-by-session notes — May 8, 2026

Live annotation monitor: Open hypothesis-monitor.html → in a browser tab. Polls every 30s; flashes and beeps on any annotation containing #zoom; one-click copy to Zoom chat.

Two distinct problems

It helps to separate what we're trying to do into two distinct goals — they have different solutions and don't need to be integrated into a single tool.

Problem 1 · Real-time threaded discussion during the workshop

Can participants have multiple simultaneous sub-conversations — following different threads, flagging questions, reacting asynchronously — without fragmenting the room's attention or creating chaos in Zoom chat?

This is genuinely hard and may not have a good solution at this scale. See the options analysis below.

Problem 2 · Structured belief elicitation, feedback, and synthesis

Can we capture participants' quantitative beliefs, elicit structured reasoning, track belief updates pre/post workshop, and synthesize findings — with appropriate privacy controls and shareable outputs?

This is largely solved: the beliefs form handles input; AI synthesis handles output. The plan is informal/async via ChatGPT (shared account) on the exported transcript + chat + Hypothesis annotations, not a dedicated live tool.

Lessons from the wellbeing workshop (March 2026)

Google Docs failed in practice — but the lesson is nuanced.
  • Participants' focal point stayed in Zoom. The Doc competed for a tab people didn't reliably switch to. Conversation split across 2+ places.
  • Simultaneous editing required adaptation: when someone edits text above you, your text moves. Page and section breaks were tried, but that made things feel more disaggregated rather than less — a feature of simultaneous editing, not an unusual problem.
  • Zoom chat was chaotic: linear, unthreadable, multiple conversations colliding — but people used it anyway because it was right there. Turning it off isn't the answer — you need Zoom chat to corral people in and out of breaks and for logistics.

The nuanced lesson: The Doc could have worked if we explicitly redirected participants to it as the focus point — e.g., "Let's now have this discussion in the Google Doc." But that's also limiting: it asks people to shift their primary attention away from the video, which fights the natural instinct of a video call. Have a Google Doc standing by but don't offer it unless participants ask for it.

Problem 1 — Options for real-time discussion

No tool solves the core tension cleanly: anything that runs outside Zoom window requires a tab switch; anything inside Zoom is limited. Here are the options:

Substantive discussion — primary channel

Hypothes.is on the agenda page Primary for content

Participants annotate the agenda page or the beliefs form inline — questions, reactions, substantive pushback — which persist beyond the session and create a threaded, durable record.

  • Preferred for anything that needs a considered response — technical questions, disagreements, follow-up points, questions for a specific discussant
  • Both pages are already open in participants' browsers; each agenda topic has a "Discussion space" fold with labelled items
  • We already monitor annotations in real time (Claude Code polling) and can surface them verbally: "Someone's annotated a question for Aleksandra on the media fold..."
  • Sort sidebar to "Newest" (dropdown at top of Hypothes.is panel) to track incoming comments live
  • Works because it's explicitly directed — announce it clearly: "If your question isn't urgent, annotate it on the agenda page; we'll surface the best ones"
Immediate and procedural — secondary channel

Zoom chat Procedural only

Reserve Zoom chat for things that need immediate attention during the session — logistics, audio/video issues, quick reactions. Don't try to make it the substantive discussion channel.

  • Use for: "I can't hear you", "can you slow down?", "your mic is cutting out", time signals
  • Fine for quick +1s or brief reactions during a speaker's point
  • Co-host monitors and surfaces anything important verbally ("Someone in chat can't hear — can you check your mic?")
  • Do not try to fight it — people will use it. Channel it toward procedural use by example and prompting
  • Export chat log at the end — paste alongside AI Companion summary when running Claude synthesis (AI Companion does NOT capture chat)
Backup — only if Zoom chat gets cluttered

Zoom breakout channels After Break 1 only if needed

If Zoom chat becomes chaotic with multiple conversation threads colliding, create 2–3 labelled breakout rooms (or separate channels) for specific topic threads — but only after Break 1, and only if participants seem frustrated by the clutter.

  • Don't create them proactively — the added complexity isn't worth it unless chat is actually a problem
  • If you do open them: announce clearly what each is for; keep the main session as the primary space
  • Decision point: end of Break 1 — if S1 chat was unmanageable, set up before S2
Not using

Zoom Whiteboard Not for May 8

Not using. Adds a tab-switch burden without clear payoff at this group size. Hypothes.is discussion spaces serve the same "structured written input" function. May revisit for future workshops if there's a specific structured-brainstorm activity that needs it.

Not using

Slido Not for May 8

Not using. Would compete with the beliefs form for quantitative input and add another tool participants need to learn. The Hypothes.is sidebar covers the Q&A use case adequately.

Experimental — probably too short notice

AI mediators Interesting for future workshops

Purpose-built systems where AI agents moderate parallel discussion threads, surface recurring themes, and flag divergent views in real time.

  • Very interesting for future workshops — worth exploring properly for the next one
  • Likely too much to pull together for May 8 on short notice
  • The programmatic version (Claude Code on the transcript/form data asynchronously) is feasible and is the plan — see Problem 2 below

Problem 2 — Belief elicitation and synthesis

This is largely handled. The plan for May 8:

Primary instrument

Beliefs form Core tool

At uj-cm-workshop.netlify.app/beliefs — structured cost estimates, probabilities, credible intervals, expert distribution mode. Orient participants to it during break orientations and at the S3 live update moment. Not for real-time discussion.

Live transcription + notes

Zoom AI Companion On throughout

Generates a meeting summary with chapters and action items. Keep it running throughout S1 and the public part of S3.

AI Companion stays on throughout — including S2 and off-record S3. Disabling it doesn't meaningfully protect privacy: anyone can run Granola, a local recorder, or a phone. We pause cloud recording for off-record sections for our own operational clarity — to keep our shareable record cleanly chunked. Be clear with participants: "No cloud recording — but AI Companion is for our notes, not external distribution." Don't oversell it as a privacy guarantee.
Key caveat: The Zoom AI Companion captures audio and transcript but does not see the Zoom chat. If you want the AI synthesis to include chat content, you must manually copy the chat log and paste it alongside the AI Companion summary when running the Claude synthesis. The two are separate sources.

Active use during sessions: You or the co-host can query the AI Companion in real time — e.g., "What are the key questions that have been raised in the last 20 minutes?" — to get a live summary and surface things to raise verbally. Do this between topics and at session end as a bridging tool. Remember it won't include chat-based questions.

AI synthesis

Claude — async, ideally programmatic Plan for May 8

Asynchronous rather than a live tool — can be done in breaks or same-evening. Two levels of ambition:

  • Informal (minimum): export Zoom transcript + chat log + recent Hypothesis annotations (sorted by Newest) → paste to ChatGPT (shared account) → "Summarise the main points of agreement and disagreement from S1. What cruxes were unresolved?" Share key findings aloud at the start of the next session.
  • Programmatic (ideal): use Claude Code to pull Netlify form submissions as JSON, compare pre- and post-workshop belief distributions, and generate a structured synthesis. Automates the pre/post comparison and scales to future workshops.

Note: the beliefs form is designed for ongoing updates — participants can return after the workshop and revise their answers. The post-workshop follow-up email should invite them to do so, and the programmatic synthesis can pull submissions at multiple time points.

S3 public note: When recording resumes for S3 public (14:35), participants may naturally reference things discussed in the off-record S2. The AI Companion will capture this. Before releasing any S3 recording or transcript publicly, audit it for content that originated in S2 and redact accordingly. Tell participants clearly: "We'll review this section before release and redact anything that was originally raised off-record."

Async follow-up

GitHub Discussions

At github.com/unjournal/cm_pq_modeling/discussions — for before/after the workshop, not during. Three threads: Workshop, Substantive, PQ Framing. Share in the follow-up email.

Operator & Assistant Quick Reference

Compact action lists for David (facilitator/operator) and co-host (assistant). Expand the session script below for fuller notes.

Before the workshop

WhenDavidCo-host / assistantStatus
By May 6 Create Zoom meeting (auto-record on). Send participant email: Zoom link + agenda + beliefs form. Confirm co-host. Receive briefing. Test Zoom setup in a practice call (joining, co-host promotion, AI Companion). Done · Anthony confirmed
May 7 Send final reminder email. Test AI Companion in solo Zoom call. Submit test beliefs form response. Prepare ChatGPT synthesis prompt in a tab. Confirm they know: (1) open hypothesis-monitor.html as their primary annotation monitor — test the alert sound and #zoom flow; (2) synthesis task at Break 1 (use "Copy all" from monitor + Zoom transcript → ChatGPT); (3) internal-notes-only role in S2. Done
May 8 · 10:45am Open Zoom. Verify recording auto-starts. Pin agenda + beliefs form links in chat. Enable AI Companion. Join at 10:45. Receive co-host promotion. Open hypothesis-monitor.html in a dedicated tab — click anywhere to unlock audio, press "Test" to verify the beep. Open ChatGPT tab with synthesis prompt ready. Day-of

During the workshop — roles at a glance

MomentDavidCo-host
S1 open (11:00) Announce recording. Orient to agenda, beliefs form, Zoom chat, Hypothesis sidebar. Watch hypothesis-monitor.html — flashes and beeps on #zoom annotations; click the annotation link to see context; one-click copy to paste into Zoom chat. Also keep Hypothesis sidebar open (sorted Newest) for fuller view. Surface questions verbally: "Someone's annotated a question on this topic…"
S1 close (12:05) Wrap session. Hand off to Break 1. @AnthonyRowett: run the S1 synthesis — share 3-point summary in Zoom chat by 12:20.
Full synthesis protocol →

Setup (ideally done May 7): Open a fresh ChatGPT Pro conversation, no extended thinking. Paste these URLs and say "Read all of these before I ask you anything": uj-cm-workshop.netlify.app/agenda · /beliefs · /resources · unjournal.github.io/cm_pq_modeling/. Keep this tab open. Also open a second window with the frontier non-Pro model (o3 / o1) as a faster backup.

At break — three sources to collect:

  1. Hypothesis annotations: hypothesis-monitor → "Show all" → Copy all
  2. Full Zoom transcript: in the Zoom meeting window → click the AI Companion icon (speech bubble) → Transcript tab → select all text → copy. This is the full running transcript, not the auto-chapter summary — it captures everything said, including things the chapter summary condenses or omits.
  3. Zoom chat log: Zoom → Chat panel → click at top right → Save Chat → open the saved file and copy the text. (Or: click in the chat pane, Cmd+A, Cmd+C.) AI Companion does NOT capture chat — this step is essential.

Paste all three into the pre-loaded ChatGPT tab with this prompt:

Here are Hypothesis annotations, the full Zoom transcript (AI Companion), and the Zoom chat log from Session 1. Based on this and the workshop materials you read, give a concise structured response: 1. TOP 3 UNRESOLVED QUESTIONS FOR S2 — most worth pressing industry practitioners on in the off-record session 2. S1 OVERVIEW FOR S2 — broad topics discussed that bear on real costs, CDMO economics, industry ground-truth 3. AGREEMENT & DISAGREEMENT — where do participants converge/diverge? Along what lines (academic vs. industry, optimist vs. skeptic, specific parameters)? 4. KEY TECHNICAL CLAIMS — about costs, cell densities, media, growth factors, or viability that seem important or contestable 5. DATA & RESEARCH CITED — specific papers, datasets, or company data mentioned or alluded to 6. RESEARCH GAPS — what was flagged as missing evidence or needed research 7. KEY UNCERTAINTIES — the central unknowns participants seemed most uncertain about 8. NOTABLE OR SURPRISING CLAIMS — strongest or most unexpected things said, worth probing in S2 9. S2 DISCUSSION PROMPTS — 3 structured provocations to open S2 productively Be concise — readable in under 3 minutes.

Run the same prompt in the backup window simultaneously — share whichever finishes first.

Break 1 (12:10) Manage room, answer DMs. PAUSE recording. Brief beliefs form screen-share orientation.
S2 open (12:25) Announce off-record. Confirm no recording. Facilitate. Monitor Zoom chat. Notes optional — don't stress about it.
Break 2 (13:35) Answer participant questions about the beliefs form. Recording stays paused. Run the S3 framing synthesis in ChatGPT Pro (no extended thinking — context should already be loaded from earlier). Use only public sources:
S3 synthesis protocol →

Sources to include (do NOT paste S2 transcript — it's off-record):

  • hypothesis-monitor → "Show all" → Copy all — this includes public Hypothes.is annotations made during S2, which are already publicly visible on the agenda page
  • S1 Zoom AI Companion transcript excerpt
  • Zoom chat log from S1 (not S2 — chat during S2 is internal)

Paste all of the above into the pre-loaded ChatGPT Pro tab (which already has the agenda loaded) with this prompt:

Here are the public Hypothesis annotations (from both S1 and S2) and the S1 transcript/chat. You already have the workshop agenda loaded. Based on all of this: What unresolved questions from S1, and from the public Hypothesis comments made during S2, should S3 address? Consider the proposed S3 agenda topics and provide synthesis and suggestions for each. Cover: 1. CONSENSUS & DISAGREEMENTS — Where do participants agree and disagree? What findings seem most important to share publicly? Along what lines do groups tend to differ (academic vs. industry, optimists vs. sceptics)? 2. CRUXES — What seem to be the key cruxes — the specific points of disagreement whose resolution would most change conclusions about CM viability and funding? 3. RESEARCH & EVALUATION PRIORITIES — What specific papers, questions, or data would most reduce remaining uncertainty? What would The Unjournal most usefully commission? 4. ANIMAL WELFARE FUNDING IMPLICATIONS — Given what has emerged, what are the key considerations for AW funders? Should CM receive more or less funding relative to proven interventions? What would change the answer? Use the S3 agenda structure as your frame. Output is for internal use to shape how David opens S3 — concise, actionable, 3 minutes to read.

Share key framing points with David privately (Zoom DM or chat) before S3 opens. Do not paste output into the main Zoom chat.

Also: second beliefs form orientation for participants. Confirm recording still paused.
S3 off-record (13:50) Announce still off-record. Facilitate open discussion. Signal public transition at 14:30. Internal notes only. Monitor chat. No external sharing.
S3 public (14:35) RESUME recording. Announce on-record. Synthesis discussion + beliefs update moment. Confirm recording has resumed — if David hasn't already done it, do it yourself or send him a direct Zoom DM: "Resume recording now." Monitor chat. Flag S1/S2 questions worth raising for the record.
Workshop end (15:00) Thank participants. Share follow-up timeline in chat. Close Zoom. Save internal notes. Export chat log.

After the workshop

TaskNotes
RecordingDownload Zoom cloud recording. Review S1 + S3 public portion before sharing. Trim if needed.
AI synthesisFull Zoom transcript + chat log + Hypothesis annotations → ChatGPT → workshop synthesis draft. Use as basis for follow-up email.
Beliefs analysisPull Netlify form submissions (JSON). Run Claude Code on pre/post responses → distribution comparison. Ideally automated.
Follow-up emailWithin 2–3 days: synthesis, recording link (public sessions only), beliefs form link (invite post-workshop update), GitHub Discussions.
Beliefs update windowLeave form open. Participants can revise answers after digesting the workshop. Pull a second snapshot ~1 week later for the final pre/post comparison.

Monitoring Hypothes.is annotations during the session

Each topic on the agenda now has a Discussion space fold with 4 labelled items — including a "Questions for [discussant]" line. Participants can annotate any of these before, during, or after the session. The co-host's job is to surface relevant ones verbally.

Primary tool — hypothesis-monitor.html

Open uj-cm-workshop.netlify.app/hypothesis-monitor.html → in a dedicated browser tab. No setup required — works in any browser, no terminal needed.

FeatureHow it works
#zoom tagIf a participant includes #zoom anywhere in their annotation, the monitor flashes green, plays an audio beep, and shows a banner. The co-host clicks "Copy to clipboard" and pastes into Zoom chat with one keystroke.
All annotationsToggle "Show all" to see every annotation across all 5 workshop pages (Agenda, Beliefs, About, Resources, Critique) — not just #zoom ones.
Direct linkEach card shows the page name as a clickable link (e.g. "Agenda ↗") that opens the exact annotation in context on the page.
Copy allExports all visible annotations as formatted text for pasting into ChatGPT at break time for synthesis.
Undo clearIf the feed is accidentally cleared, "Undo clear" restores it for up to 60 seconds.
Always-onRuns entirely in the browser — no Claude Code or terminal needed. Keep the tab open; it polls every 30 seconds automatically.

Tell participants: "If you have a question or comment that needs to be raised in the Zoom session right now — not just for later — include #zoom anywhere in your Hypothes.is annotation. The co-host will see it immediately."

Backup — Hypothes.is sidebar directly

Keep the agenda page open in a browser tab with the Hypothes.is sidebar visible (sort by Newest). Covers the same ground but only shows the current page and requires manually watching the sidebar. Use as a complement to the monitor, not a replacement.

Session-by-Session Script

European morning drop-in (9:00–10:00 AM ET)

Informal · Not recorded · Small group
  • Open Zoom early for EU/CET participants who cannot stay for the full afternoon.
  • Informal: introductions, early framing questions, quick walkthrough of model/agenda if helpful.
  • Input is carried forward to S1 — note anything worth raising verbally later.
  • Recording off. AI Companion on (for your notes). No announcements needed.

Pre-workshop (10:45am ET — 15 min before start)

Internal
  • Open Zoom at 10:45. Verify cloud recording will auto-start.
  • AI Companion is on by default — no toggle needed.
  • Promote co-host immediately when they join.
  • Co-host: open hypothesis-monitor.html in a dedicated browser tab, click once anywhere to unlock audio, press "Test" to confirm the beep works. Switch to "Show all" mode initially to catch everything.
  • Pin agenda link in Zoom chat: uj-cm-workshop.netlify.app/agenda
  • Pin beliefs form link in chat: uj-cm-workshop.netlify.app/beliefs

S1 · Technical Foundations

11:00–12:10 ET · Public · Recorded
  • Opening (11:00): Announce recording. Point to the beliefs form and agenda in chat. Explain the two-channel plan: "Zoom chat is for immediate/procedural needs — audio issues, quick reactions. For substantive questions and comments, please annotate the agenda page via Hypothes.is — we're watching the sidebar and will surface the best ones verbally." Ask everyone to use Zoom hand-raise if they want to speak.
  • Opening script: "This session is being recorded. Two things in chat: the agenda page (you can annotate it inline to leave a question — we'll surface them) and the beliefs form. Use Zoom chat for quick reactions or if something needs immediate attention. For a considered question or comment, annotate the agenda — it's more permanent and we'll pick them up. Hand-raise if you want to speak so we can call on you."
  • Screen sharing tip (for presenters): Ask Aleksandra (and anyone else sharing slides) to use Zoom Layout → "As background" and drag their self-view panel to minimize — this keeps their face visible alongside the slides. You can also ask them to shrink their participant thumbnail by clicking and dragging the edge of the camera strip at the top of the Zoom window.
  • During discussion: Co-host watches hypothesis-monitor.html for #zoom alerts; also keeps Hypothesis sidebar open (sorted Newest) for fuller view. Surfaces good questions verbally. Also tracks timing: message David in Zoom chat ~5 min before each sub-topic ends (e.g. "~5 min left on cell lines"). If a sub-topic is seriously running over, unmute briefly and say so — better to cut than to crowd out later topics.
  • At 12:05: Wrap S1. Signal co-host to start the Break 1 synthesis task: open hypothesis-monitor "Copy all", paste with Zoom transcript + chat log into ChatGPT.

Break 1

12:10–12:25 ET
  • Co-host: paste full Zoom transcript + Zoom chat log + recent Hypothesis annotations (sort by Newest) into ChatGPT — "What were the key questions and unresolved points from S1? What should we ask industry practitioners in S2?" — then share a 2-sentence synthesis in chat by 12:20.
  • Co-host: 5-min beliefs form orientation. Show form briefly on screen share. Encourage people to open it during the break.
  • PAUSE cloud recording before introducing S2.

S2 · Scale-up & Industry Realities

12:25–13:35 ET · Off-record · NOT RECORDED
  • Verify recording is paused.
  • Announce: "We are now off the record. No recording is running. AI Companion remains on — it's for our operational notes, not a privacy guarantee, and participants can self-record regardless."
  • Co-host monitors Zoom chat. No shared note-taking surface for S2 — any notes stay in a private doc or are not taken at all.
  • Zoom chat is fine — it isn't a recording. But remind participants that chat is visible to all participants in the room.
No Google Doc, no shared notes for S2. Internal notes only. Keep them separate from anything that will be circulated externally.

Break 2

13:35–13:50 ET
  • Co-host: 5-min beliefs form orientation (second pass — for anyone who hasn't opened it yet).
  • Co-host: paste co-host S2 notes + any Hypothesis annotations from S1/S2 into ChatGPT — "What unresolved questions from today should S3 address?" Use the output internally to frame S3 opening; share verbally (not in chat, given S2 off-record content).
  • Recording stays paused — S3 opens off-record. AI Companion remains on throughout. Do not resume recording yet.

S3 · Off-record (first part)

13:50–14:35 ET · Off-record · NOT RECORDED
  • Verify recording is still paused.
  • Announce: "S3 opens off-record — same rules as S2. We'll switch to the public, recorded part at 14:35."
  • Open, candid discussion: what didn't get fully aired in S1/S2? Business/financial environment. Regulatory specifics. What would actually need to be true for CM to reach viability?
  • Co-host takes internal notes. Monitors Zoom chat for anything worth surfacing verbally. Does NOT share notes externally.
  • At 14:30: Signal that the public recorded part begins in 5 minutes. Invite participants to note anything they specifically want on the record in the next section.
No recording, no shared notes for this section. Internal notes only. (AI Companion stays on — it is not a recording for distribution.)

S3 · Public synthesis (second part)

14:35–15:00 ET · Public · Recorded
  • At 14:35: RESUME cloud recording.
  • Announce: "We're now on the record. This part will be shared publicly — transcript and recording. If you'd like a specific contribution kept off-record, say so and we'll pause."
  • Bleed-in note: participants may naturally reference things from S2 or the off-record S3 part. We will audit this section before public release and redact any content that originated off-record. Tell participants: "We'll review this before release."
  • Synthesis discussion: consensus and disagreements; does researcher/industry divide map to specific cost components? Most important remaining uncertainties. Research and evaluation priorities for The Unjournal. AW funding implications.
  • Brief beliefs update moment: "Open the beliefs form — you can revise your CM_01 estimate in light of today. We'll compare the pre- and post-workshop distributions." Give 2–3 minutes.
  • At 15:00: Thank participants. Share follow-up timeline in chat. Close Zoom.

Post-Workshop (same day or next morning)

Zoom Settings Checklist

Account-level (zoom.us/account/setting)

Per-meeting (create the May 8 Zoom meeting)

SettingValueNotes
Auto-recordCloudVerify this is on before the meeting starts
Waiting roomOffToo much friction for a 20-person expert workshop
Mute on entryOnPrevents join noise
Host video defaultOnSet in account settings
Zoom AI CompanionOn throughoutStays on for all sessions — AI Companion is for operational notes, not a recording for distribution. Only cloud recording is paused for off-record sections.
PasscodeOn (embedded in link)Don't share passcode separately

Co-host setup

Cannot be added as alternative host (requires same Zoom org). Promote live at 11:00am:

Participants panel → hover co-host name → More → Make Co-Host. Takes 5 seconds. Do this before announcing the opening.

Pre-Workshop Checklist

By May 6 (2 days before)

May 7 (day before)

May 8 (day of, 10:45am)

What to Avoid

For future workshops: The "AI mediators" angle is worth exploring properly — systems where AI agents moderate parallel sub-threads, surface consensus/divergence, and feed back into the discussion in real time. Didn't pull together in time for May 8 but could be a significant improvement for the next one. The beliefs form + async synthesis approach works but it keeps elicitation and discussion separate rather than integrating them.

Whereby (whereby.com) was recommended as a Zoom alternative for future events — reportedly easier for non-technical participants to join (no app required, browser-based), and highly rated for reliability in clinical/patient settings. Worth testing for the next workshop. Also consider Microsoft Teams for participants who work in enterprise environments.

Internal facilitation guide · CM Workshop May 8, 2026 · Not for participant distribution