ScatterSpoke v3 — September 2026 update

ScatterSpoke v3 launched at the start of August 2026. This page covers what shipped in the five weeks after — through the release of 6 September — and where each change is documented. It builds on ScatterSpoke v3 (Aug. 2026), which remains the inventory of the platform as a whole.

The shape of this update: the first week went to fixes around permissions, sign-in, and seat counting. The rest was three pieces of new product: retrospectives gained a real Group phase with group voting and a guided discussion; the MCP server grew from “create a survey” into the whole survey loop; and Ask learned to answer questions about the product itself from the live documentation.

Retrospectives

The Group phase is back

The v3 notes said the manual Group phase was gone and grouping was done by AI clustering. We heard clearly that teams want to group by hand, so it’s back as a proper phase. Guided boards now run Feedback → Group → Vote → Review, and Group is the only phase in which the structure of the board changes. See Group related feedback.

• Drag one card onto another to make a group. The drag follows the pointer, drop targets light up, and the drop says what it will do before you let go. A card keeps its origin column through a cross-column drop, and every card wears a column badge during Group so nothing loses its context when the lanes lose their headings.
• Groups are first-class. A group has a name (where retro AI is available, a suggested name arrives shortly after you make one, and stays editable), its own comment thread, and menu operations — rename, pin, detach a card, disband. Drag a group by its header to re-anchor it in another column.
• AI grouping creates groups instead of replacing themes. Group with AI is offered from a strip across the top of the wall, states how many cards are still on their own, and promises that groups people already made are left alone. Manual groups are pinned and protected. Every run records an outcome — merged into N groups, found no duplicates, could not finish — and shows it to everyone, so a run that wrote nothing no longer ends in silence.
• Phase moves are confirmed. The stepper and the dock’s Next open a one-line confirmation before moving the room. Moving into Group reveals a still-private board in the same step, so no one sees the wall with cards hidden.

Boards that were mid-session when this shipped simply gained the next step; boards already in Vote or Review were untouched. Migrated boards had their existing themes carried across as groups with their votes intact.

Voting on groups

The Vote phase was rebuilt around groups.

• Tap to vote. The whole card body is the tap surface, with a budget bar showing what you have left. Votes land on a group as one topic, so several versions of the same issue no longer compete.
• Blind or open voting. The facilitator chooses in board settings. Blind is the default on a new board: totals stay hidden until Review so early votes don’t steer the room. This is independent of anonymity, which controls whether author names show on feedback.
• I’m done voting. Participants can mark themselves finished, and the facilitator sees a roster of who has. The participation nudge understands groups, so it won’t prompt about cards that are already covered by a vote on their group.

A guided discussion in Review

The Review phase is now a two-pane discussion: a queue of groups and standalone cards with their vote counts on the left, and the current topic on the stage. The facilitator drives it — pick from the queue, or Next and Next card — and everyone follows the same stage. A manual queue order sticks. Review is reorder-only; regrouping belongs to Group, and the board says so.

Facilitator controls

• Rename a retro after it’s created. The board’s title was previously fixed at creation. Facilitators, team admins, and workspace admins can now change it.
• Edit columns on an active board. The creator or current facilitator can rename, add, or remove columns. The last column can’t be removed, nor can one that still holds cards or an anchored group. See Cards, columns, votes, and comments.
• One sharing ladder. Scope and access mode were two separate settings; they are now four ordered rungs — Only one team, Anyone in the workspace, Link, email required, Link, no sign-in — each admitting a strict superset of the one above, identical in the create modal and the Share panel. Scope can be changed after creation. See Inviting participants and guest links.
• Workspace default access mode. Admins can set the sharing rung new boards start on, under workspace settings. An SSO-enforced workspace can’t store a guest rung as its default. Changes to a board’s access mode or scope are written to the audit log.
• Refusals look like refusals. An off-team member opening a team-restricted board sees an access-limit state that names the team, with a way back — not “Something went wrong, try again.”
• The lock icon shows the board’s state, open or locked, instead of the action it would take.

Reopening a closed retro

Reopen was limited to workspace admins, and since every migrated retro arrived closed, that meant a support request just to reopen your own board. Team admins can now reopen any retro in their team, and a retro’s creator can reopen their own board as a team member. Watchers never qualify. See Closing a retro.

Guests

Guests joining by link have parity with members on the grouped board — they can group, vote, and comment under the same rules — and they now appear in the presence roster. Guest boards also update live rather than refreshing every few seconds.

Fixes

• On a busy board a new comment could appear twice, and deleting either copy removed both. Votes could double-count the same way. Both fixed.
• Certain drags could crash the board.
• Keyboard focus returns to where you were after confirming a deletion, and the grouping controls had an accessibility and wording pass.
• The create-retro modal and share popover stay usable at any viewport height.

Surveys

• The live survey page shows its questions. The builder locks on publish, which left no way to read what respondents are being asked. A read-only Questions section now appears on the live page. See Manage surveys.
• Static-survey responses render as a form, not as an interview transcript — answered over asked, skipped questions named. See Responses and transcripts.
• The Results Brief on a static survey is metered, not plan-gated. On Free, the first analysis of a static survey spends one Agent Survey credit and re-runs are free; on paid tiers it spends nothing. Previously it was only available on plans that include Agent Surveys. See Usage credits.
• Invitation emails only promise anonymity when the link is anonymous. A verified invitation used to claim it too. Invitations and reminders now arrive from the same notifications address as the rest of ScatterSpoke’s mail.
• The share link is created when you open Share or copy it, not merely because you opened the survey page — so Distributions no longer shows a pending share-link row nobody asked for.

Fixes

• Agent Surveys could repeatedly fail to save.
• The survey builder could refuse to save after an edit, or after a turn of the agent’s Design interview. Both fixed.
• The participant date picker opened on today’s month regardless of the selected value, and date-only answers drifted a day across time zones.
• Participation counts on migrated surveys are accurate.

Ask

Ask now answers questions about the product, not only about your feedback.

• Product help from the live docs. Ask searches this documentation site before answering a how-to question and cites the page it used as a titled link — Closing a retro, Migrating legacy boards — rather than a raw URL. Where the docs are silent, Ask says so rather than guessing.
• Guarded actions. Ask can carry out a request — creating a plain static survey, for instance — with a confirmation step before anything is written.
• Links and threads. Markdown links and URLs in answers are clickable. Every thread has a shareable address (?thread=…), and the message list follows a streaming answer without fighting a reader who has scrolled up. See Threads and pinning.

The MCP server

In August the MCP server could create a survey and read results. It is now the whole loop: create, review, send, manage recipients, answer, pull results, analyse — 33 tools in all.

• Create either kind of survey. create_survey takes mode: STATIC | AGENT. An AI-interview survey created over MCP is subject to the same plan and credit checks as the web builder, and never silently downgrades to a questionnaire. Static questions accept options, scale bounds and labels, multi-select, and required — a choice question with no options is refused with the fix spelled out, rather than created and then hidden from participants.
• Nothing is emailed that nobody previewed. preview_email renders exactly what a recipient will read — envelope, body, and the questionnaire as a plain-text form — and issues a confirmation token bound to the survey, its questions, the recipients, and the privacy mode. email_survey requires that token and refuses a stale one.
• Lifecycle and recipients. get_survey, update_survey, close_survey, archive_survey, list_recipients with per-link delivery state, revoke_link, and get_share_link for a link anyone can open. Sends can be scheduled, carry a reminder cadence, and target a team.
• Read what people said. list_responses (with a table format for bulk reads) and get_response return answers joined to their question text and, where the distribution collects identity, the participant. get_survey_results gives per-question aggregates plus the signal summary and an honest coverage note when only part of the responses have been analysed; generate_results_brief runs the same analysis as the web Generate summary button. list_retro_cards reads a board as column → group → cards with votes and comments.
• Answer in one call. On a fixed questionnaire, start_response lists every question up front and answer_question accepts the whole set at once. AI interviews still go one turn at a time and say so. A reusable share link can be answered over MCP too.
• Tool names settled. send_survey became email_survey, share_survey became get_share_link, analyze_survey became generate_results_brief, ask became ask_across_feedback, and the stats and claim tools were renamed to match. The /mcp page’s catalogue mirrors the live server.

Workspace administration

Domains

A workspace’s email domains now have a place of their own: Settings → Domains. It lists every domain that governs you — SSO routing domains with whether routing is active, and verified domains on Business and above — and explains in plain language what each one causes: who can be invited, whether guests are admitted, whether magic links apply. It flags a configuration that needs attention, such as SSO being enforced with no routing domain active.

• Restrict invites to your domains is a workspace toggle, locked on when SSO is enforced.
• Domain capture at signup (Business and above). When a domain is claimed and capture is on, a new signup with that email domain can’t create a stray workspace — they’re added to yours as a Participant and routed there. Participants never consume a seat. Existing workspaces and invited users are untouched.
• The old “Approved email domains — request to join” card has been removed; the Domains page replaces it.
• Adding or removing a domain on a multi-domain SSO provider could drop the others. Fixed.

Users, teams, and roles

• The Users screen shows Participants alongside Seat Holders, with a segmented filter, badges, search, and remove. Seat counts now count seat holders, not everyone who has ever answered a survey. See Invite users and seats.
• Team admins can administer their team — roster, roles, add and remove members, and Settings → Teams — not only reopen its retros. Any seat holder can create a team. See Teams and Roles and permissions.
• The audit log is linked from the settings navigation. See Audit log.
• Admins can change any user’s role except the workspace owner’s.
• Admins no longer lose their user-management controls when they aren’t on page one of the member list, and the Settings navigation follows what your role can actually do.
• A seat-capacity refusal now reaches the person who hit it as a real message, instead of a generic server error.
• New users were being given the display name “there”.
• Some Free users were routed into a fresh workspace on login instead of their own. Creating a second workspace is now an explicit choice.

Signing in

• The Use SSO button works again. A pre-flight check that could never succeed was dead-ending Use SSO and /login?sso=true links even when the identity provider was healthy. The check is gone and both paths fall back to native SSO.
• Every reviewed domain of a multi-domain provider can sign in, and configured SAML name fields are preserved.
• An SSO-enforced workspace shows the two guest-link rungs disabled, with the reason, when sharing a retro — instead of offering a choice the server would refuse. See SSO overview.

Under the hood

Not features, but you’ll feel them.

• Retro boards load faster, and a busy board stays consistent for everyone looking at it.
• A source of intermittent 502 errors on long-lived connections was fixed.
• Seat-request emails reach every admin they’re meant to, even in a large workspace.
• Survey and settings controls work properly on phones, and accessibility fixes landed across the app.
• When something is refused, you see a plain-language message rather than raw error output.

Note

For the platform as a whole — plans, retro formats, Signals, billing, branding — see ScatterSpoke v3 (Aug. 2026). Anything not mentioned here is unchanged from that page.