Changes to Engage

Changelog

Read recent changes to Engage, from new features to fixes and improvements. The source repository holds the full development history.

17 November 2026

6 commits
  1. Featuresense-box
    Publish priorities, discuss, link proposals and inline amendments (#150)

    Adds review-then-publish Sense Box priorities with public discussion and contributor status links; one priority can inform multiple proposals. Adds block-anchored inline amendments with fenced, recoverable application. Installs @ui-sdk/react and @ui-sdk/core.

    c44e0d5 ↗
  2. Featurecollaboration
    Public story and photograph contributions (#143) (#145)

    * feat(collaboration): model public story matters and guest intake Register a collaboration Matter type with no governance capability and no lifecycle stage, so a public invitation can never be deliberated, balloted, amended, ratified or published as policy. Add guest submission with a versioned consent scope, private media storage, metadata stripping for JPEG and PNG, reviewer-only permissions, and a durable rate limit plus Turnstile action for the public form. Refs #143 * feat(collaboration): publish the story invitation and contribution form Add /collaborate/ and /collaborate/<slug>/ where anyone can read the invitation and send story answers, photographs, or both without an account or party membership. The form asks for publication channels, credit, edit and AI-edit permission and an optional reply address before sending, keeps every answer when a submission fails, and never claims publication has happened. Refs #143 * feat(admin): add the story review queue and invitation editor Reviewers see each contribution with its answers, private photographs and consent summary, then record a decision, a note and where an approved item was used. Editors own the invitation wording, questions and open state without a deployment; question keys stay stable so earlier answers remain attached. Refs #143 * test(collaboration): cover guest intake, consent and private media Prove the guest path needs no account, consent cannot be widened from the client, private-only contributions cannot be approved, an anonymous name is never stored, concurrency cannot overwrite a decision, and image bytes stay private and metadata-free. Real-MongoDB cases cover unique references, version writes and the BSON media round trip. Refs #143 * docs(collaboration): record the matter model and launch obligations Describe the collaboration Matter, its consent and data handling, what editors can change without code, and the legal and image-policy gaps that must close before the invitation is promoted publicly. Refs #143 * docs(agents): require new features to be configurable by default Add a standing rule: editor-facing wording, questions and options belong in stored configuration, while permissions, consent vocabulary and security bounds stay in reviewed code. Refs #143 * chore(changelog): record public collaboration matters Refs #143 * fix(collaboration): make optional story answers explicit Refs #143 * chore(changelog): record the contribution form clarification Refs #143 * docs(collaboration): separate accounts, membership and contributing Refs #143 * chore(changelog): record the collaboration documentation note Refs #143 * feat(collaboration): delete material on withdrawal and by retention sweep Withdrawing a story now removes its text, name, reply address and photographs immediately, keeping only the review record and its date. Add a monthly sweep for declined material after 90 days, legacy withdrawals and approved reply addresses after 12 months. The sweep is dry-run by default and records what it would clear. Tell contributors the photo and third-party rules at intake, and warn reviewers that withdrawal is irreversible for the material. Refs #143 * test(collaboration): cover withdrawal clearing and retention windows Refs #143 * docs(collaboration): add pre-launch safeguards and privacy text Write the operating rules for photographer credit, other identifiable people, minors, AI-assisted edits, publishing and mistaken publication, and mark what still needs an owner decision. Extend the privacy policy (version 1.3) for the contribution purpose, its reviewer access and its retention schedule, while it remains owner-authorised with legal review outstanding. Refs #143 * fix(db): allow a local MongoDB selection timeout override A slow local Docker MongoDB took longer than 400 ms to answer its first selection, so development silently fell back to the in-memory preview and pages looked empty. The override applies locally; production keeps its 3 s default. Refs #143 * chore(changelog): record contribution withdrawal, retention and safeguards Refs #143 * chore(changelog): refresh after merging main * feat(collaboration): manage multiple invitations and typed profiles Replace the single seeded invitation with an authorised invitation directory: create as a draft, open, pause, close, archive and restore, with contribution counts per invitation. Archiving and drafting have no public page. Each invitation is created with one reviewed contribution profile. A written responses invitation never asks for photographs, a name or edit permission, and the server rejects photos sent to it, so a consultation does not inherit the storytelling form. Refs #143 * fix(collaboration): keep review history and finish the guest journey - Load existing notes and the contributor message, and only overwrite them when edited, so a status change cannot erase the record of a photographer's permission. Write the decision and its audit event in one transaction. - Make withdrawal and retention clearing terminal for the material, and offer 'Accepted as consultation input' as a separate outcome from publication. - Snapshot the invitation title, purpose and question labels with each contribution, and refuse an answer to a removed question rather than dropping it silently. - Allow several placements in the same channel and show the recorded history. - Keep the Save action available when 'withdrawn' is chosen, require a declared upload size, add photo previews, and allow blob: only for local previews. - Give contributors a reference, explain what 'no name' means for a photo, and require a reply address before AI-assisted edits. Offer an optional account next step after contributing. Refs #143 * test(collaboration): cover the reviewed journeys and remove a flaky event read Cover invitation lifecycle, profile isolation, notes preservation, terminal states, accepted private input, question snapshots, stale answers, repeated placements and reference search. Select the settings event by content rather than position, because several can share a millisecond. Refs #143 * docs(collaboration): record invitation management, profiles and journey evidence Refs #143 * chore(changelog): record collaboration invitations, profiles and fixes Refs #143 * feat(collaboration): publish an approved story gallery Show an approved story on /collaborate/<slug>/stories/ only while it is approved, the contributor allowed website use, it is a story-profile contribution, its material is intact and a reviewer has not hidden it. Credit follows the contributor's choice and no name is stored for an anonymous story. Photographs are served from a public route that re-checks publication on every request, so hiding or withdrawing removes access immediately. Reviewers can hide a story from the gallery, or restore it, without changing its approval. The action is versioned and audited. In-memory array filters now honour $in, which the gallery query needs and the preview previously ignored. Refs #143 * test(collaboration): cover gallery consent, hiding and withdrawal Refs #143 * docs(collaboration): document the approved gallery and its limits Refs #143 * chore(changelog): record the approved story gallery Refs #143 * chore(changelog): refresh after merging main * feat(collaboration): publish explicitly and measure interest Separate approval from publication: an approved story appears in the gallery only after a reviewer presses Publish, and hiding or showing it later keeps the first publication date. Approval alone never makes material public. Adopt the adult-only pilot rule with a simple 18+ attestation, no date of birth, and no publication of identifiable children. Record the approved removal commitments in the safeguards. Add aggregate interest counts for arrivals, first interaction, accepted contributions and the optional account step. A row holds only invitation, UTC day, stage, count and expiry. There is no cookie, session, address, identifier or typed content; the public endpoint is rate limited and the rows expire. Refs #143 * docs(collaboration): record publication, minors, removal and funnel rules Correct the gallery claims about a missing public image route and about what withdrawal can erase. Add dated screenshots of the gallery states, the explicit publication control and the interest panel so the review can see them. Refs #143 * chore(changelog): record explicit gallery publication and interest counts Refs #143 * feat(collaboration): let contributors correct or withdraw with a private link Each contribution receives a management token derived from its request reference and a deployment secret, so a retry returns the same link. Only a hash is stored and the raw token lives in the URL fragment, never in a request URL, log or referrer. The link shows the contribution and status, allows a correction that returns it to review and off the gallery, removes a published story from the gallery, or withdraws everything at once. The action is audited as the contributor's and the link stops working afterwards. Retention corrections from review: reply-address expiry runs from a stable approval date, and each sweep clear is conditional on the record still being in the expected state, with the media delete in the same transaction. The show action now requires a prior publication. Interest ratios say they are event ratios, not visitor rates. Refs #143 * docs(collaboration): record the management link, retention dates and pilot owners Fix the outdated guardian exception, point the empty gallery at the separate publication step, add management screenshots, and add an operations page that names the four unassigned responsibilities: inbox owner, backup, escalation and retention sweep operator. Refs #143 * chore(changelog): record contributor management and retention corrections Refs #143 * chore(changelog): refresh after merging main * feat(collaboration): optionally link a contribution to a signed-in account A member who holds a contribution's private link can link it to their Engage account after ticking explicit consent. The record stores the member identifier, time and consent version, and an audit event marks both linking and unlinking. Claiming is optional, changes neither publication nor approval, creates no permission, and leaves the private link as the only way to correct or withdraw the contribution. A second account is refused. Refs #143 * docs(collaboration): describe optional account linking and its limits Record that linked contributions do not yet appear on the member profile, bump the privacy policy to 1.6, and add a screenshot of a linked contribution. Refs #143 * chore(changelog): record optional contribution account linking Refs #143 * feat(collaboration): add an authenticated Horizon story feed GET /api/v1/horizon/stories serves published stories to Horizon with a bearer credential. It fails closed when unconfigured and rejects a missing or wrong token. Scope matches the public gallery: approved, explicitly published, website consent, story profile, material intact, not hidden. Consultation responses never appear. The payload carries answers, chosen credit, caption, media URLs with alt text, channels and consent flags, and excludes reply addresses, notes, member ids and tokens. Supports since/nextSince incremental polling and If-None-Match 304. Refs #143 * docs(collaboration): document the Horizon feed and its credential Refs #143 * chore(changelog): record the Horizon story feed Refs #143 * feat(profile): show linked published stories on the member profile A member who linked a contribution sees it in a Published stories section only while it is public in the gallery. The entry carries the initiative title, an excerpt of the first answer, the publication date, the photo count and a link to the story's gallery anchor. Unpublished, unapproved, hidden, withdrawn, cleared or consultation-linked material never appears, and the profile exposes no email, notes or management token. Hiding or withdrawing removes it from the profile at the same time. Refs #143 * docs(collaboration): note profile display and add profile evidence Refs #143 * chore(changelog): record linked stories on the member profile Refs #143 * test(horizon): add a simulated consumer and contract smoke poller The consumer contract checks the envelope, credit, website consent, media URLs and alt text, duplicate ids, dates and the absence of private keys. The smoke poller exercises 401, an authorised poll, one image download, entity-tag revalidation and an incremental read, and can run against staging or production. Refs #143 * docs(horizon): document the consumer smoke check and its evidence Refs #143 * chore(changelog): record the simulated Horizon consumer Refs #143 * refactor(collaboration): remove the Horizon-specific feed from the pilot The feed was a single-consumer prototype: a Horizon-named route, types and service, one deployment credential, no revocation, and no propagation of corrections, hiding or withdrawal to a consumer. That does not meet the external-sync design, so it moves to a focused follow-up (#157) with the manual export as the interchange in the meantime. Refs #143 #157 * feat(collaboration): assign operational owners in the admin panel Primary owner, backup, retention sweep owner, correction address and escalation details are now one audited administrator setting at /admin/collaboration/operations/. Personnel changes need no deployment. Saves require collaboration.manage, reject unknown or disabled members and a backup that duplicates the primary, resolve names for display, and record before/after values in an audit event. Closes the documentation-only ownership gap for the pilot. The roster now reports itself as unassigned until all three people and a contact are set. Refs #143 * docs(collaboration): record in-app owners and remove Horizon feed claims Point the operations runbook at the admin setting, remove the Horizon feed documentation and evidence, correct the privacy-version drift to the current policy, and add the owner-setting screenshot. Refs #143 #157 * chore(changelog): record operational owner settings and feed removal Refs #143 #157 * fix(collaboration): make operational coverage and correction addresses real One address model: the operations setting is the default, an invitation may override it, and an empty invitation field inherits. Public pages resolve that address instead of reading only the invitation. Readiness is revalidated on every read. An owner is ready only when the account is active, the membership verified and the member holds collaboration.review. The screen now distinguishes assigned from ready and names each reason; assigning someone no longer implies they can act. The settings write is versioned and transactional with its audit event, with a unique index on the global record, so concurrent saves cannot overwrite each other or leave a decision without its history. Refs #143 * docs(collaboration): record the address model and readiness checks Refs #143 * chore(changelog): record operational readiness and address inheritance Refs #143 * fix(collaboration): fail closed on signing secrets and tighten identity checks Production refuses to derive management tokens without a configured signing secret, instead of silently using the shared development value. A regression test covers production with and without a secret. The submission fingerprint now includes the invitation, so reusing a reference for another invitation is refused rather than returning the original record. The gallery publication API explicitly rejects written consultation responses. Refs #143 * docs(collaboration): restore the product model and correct readiness claims The product page had been overwritten with the operations runbook. Restore the model, consent, publication, claiming and retention material, then add the current ownership, address inheritance and readiness details. Correct the runbook's claim that a partial roster can be saved. Refs #143 * chore(changelog): record secret, fingerprint and documentation fixes Refs #143 * fix(collaboration): separate public story identity from management authority Public gallery and profile links now use a random storyId. The submission reference is never exposed publicly, so a visitor cannot replay a published story into a management token. A response is issued only when the request creates the contribution. Replaying an existing submission returns no token even with identical content; matching content is not proof of ownership. The original token still corrects and withdraws, and idempotent retries still avoid duplicates. Adds a unique partial index for storyId and a regression test that uses only information available from the public gallery. Refs #143 * fix(collaboration): treat a racing first owner save as a conflict Two simultaneous first saves hit the unique settings key. Resolve the duplicate key to a 409 version conflict instead of a 503, with real-MongoDB coverage for the race and for a stale save after success. Refs #143 * docs(collaboration): correct retention command and override monitoring Refs #143 * chore(changelog): record the management-token and owner-save fixes Refs #143 * fix(collaboration): never issue management authority to a lost insert race The duplicate-key recovery path returned the derived token even though that request did not create the contribution. It now returns no token, matching the sequential replay guarantee. Adds concurrent identical-submission regression tests on the in-memory preview and real MongoDB: one contribution, one token. Both fail if the recovery path returns the token. Refs #143 * chore(changelog): record the lost-insert-race token fix Refs #143

    2594e23 ↗
  3. Featuresettings
    Make public site settings stored and audited only (#128) (#144)

    * feat(settings): make public site settings stored and audited only (#128) Remove the environment fallbacks for search indexing, the header action, sharing image, social channels, branch directory, repository link and the public pulse poll. Stored values under Platform settings win; absent values use the documented code defaults. APP_MODE still supplies the production indexing default. The repository link is now editable in the Operations section instead of read-only, with URL validation, the operations audit section and no raw keys. Invalid saved values fall back to defaults and the settings page names the saved settings that could not be used. Align the stored-value sanitiser with the API validators so an invalid saved value cannot partly apply. Tests cover every removed variable having no effect, stored values winning, defaults, invalid fallback, API validation and audit. * chore(changelog): record public site settings migration (#128)

    90bc0c1 ↗
  4. Featuresettings
    CRM tuning and job gates as audited admin settings (#128) (#142)

    * feat(settings): move CRM tuning and job gates into audited admin settings (#128) Store CRM verification enablement, batch size, reconcile interval, outage alert threshold and request timeout, plus the scheduled digest gate, in the operational_settings collection. Each document is validated, defaulted and audited, and the admin settings page gains CRM verification and Scheduled jobs sections. The cron route, digest sender, verification cycle and health report now read enablement and tuning from the database. Missing settings mean off, matching the former unset-variable default. Invalid stored values fail clearly and are never clamped. The six environment variables and their PlatformEnv fields are removed. * docs(config): document CRM and job settings as administrator configuration (#128) Remove the migrated variables from .env.example, the environment table and the CRM and email guides. Document defaults, bounds, the admin locations and the deployment step required when upgrading from the environment gates. * chore(changelog): record CRM and job settings migration (#128)

    4e13843 ↗
  5. Featuresettings
    Operational health thresholds as audited admin settings (#128) (#141)

    * feat(settings): make operational health thresholds audited admin settings Move the eight operational health thresholds from environment variables to a single validated, audited setting: - scheduler warn/fail, digest warn/fail, verification warn/fail - ballot closure grace and email delivery reporting window The setting is one document in `operational_settings` (`_id: "thresholds"`) with documented defaults and whole-hour bounds. Warn must stay below fail. Validation rejects partial, fractional or out-of-range input field by field, and nothing is stored until the whole set is valid. Each change appends an event to `operational_settings_events` with the actor, time and only the changed fields, before and after. There are no secrets. `ops:health` reads the saved values and names each threshold's source as `settings` or `default`. A saved value that is missing, not an integer or out of range is reported as `unknown` with a plain-language action; the affected threshold is labelled as the default it fell back to. Warn/fail ordering is checked at evaluation too. The eight environment variables and their `PlatformEnv` fields are removed. `/admin/settings/` gains an Operational thresholds form built from the admin Card, Field, Input, Button and Alert primitives, with grouped pairs, defaults, ranges, inline errors, a success confirmation and the last audited change. The new `POST /api/admin/operational-settings` requires `settings.manage`, checks same-origin writes and returns field errors without internals. Tests cover validation, defaults, merging, audit shape, permission denial, same-origin rejection, API errors, health sources, invalid stored values, read failure, environment variables having no effect, and replica-set persistence. * docs(operations): point operational thresholds at the admin setting Remove the eight threshold variables from `.env.example` and the environment table. Document the administrator setting instead: defaults, whole-hour bounds, warn-below-fail ordering, and the audited change record. Record what remains on #128: CRM batch sizing, reconcile cadence, outage attempts, request timeout, the digest/CRM enablement gates and the site setting environment fallbacks. * chore(changelog): record operational threshold settings

    1ebab4e ↗
  6. Featurediscord
    Post the weekly digest as an embed (#131)

    * feat(discord): post the weekly digest as an embed (#129) Replace the plain-text weekly post with one Discord embed: a Horizon accent rule, an intro line, a linked field per featured proposal showing its stage and activity, and a short footer. The embed title links to Engage; the test post is labelled in the embed title and footer. Proposal titles remain escaped and mention-neutralised inside the embed, and allowed_mentions still permits only the configured role. The admin preview now renders the embed wording and links instead of the raw payload. * chore(changelog): record the Discord digest embed

    e27a109 ↗

16 November 2026

3 commits
  1. Featureprofiles
    Add profile visibility and CRM identifier seam (#121)

    * feat(profiles): add profile visibility and CRM identifier seam Let owners hide their public profile; other viewers and visitors receive 404 and author links are suppressed, while the owner keeps a preview. Extend the profile settings route to update biography and visibility together. Add an opt-in CRM identifier seam: with ENGAGE_PROFILE_CRM_IDENTIFIERS=on, an active CRM identity link becomes the canonical profile path and legacy member-id paths redirect with 301; the lookup is local and off by default. Add coverage for both behaviours and verify the visibility control in the browser harness. Refs #113 * docs(profiles): document profile visibility and CRM redirect Record the owner visibility switch, hidden-profile 404 and link suppression, the opt-in CRM identifier seam and its 301 behaviour, initials-only avatars, the absent self-service account deletion, and the display-only comment score. Refs #113 * chore(changelog): record profile visibility and CRM redirect Refresh the public changelog for the owner visibility switch and the opt-in CRM identifier seam, preserving older entries outside the generator's 80-commit window. Refs #113 * chore(changelog): record profile visibility after merging main Regenerate the snapshot on the merged branch and preserve older entries outside the generator's window. Refs #113 * fix(profiles): harden partial updates and CRM id collisions Write only the profile fields supplied by a partial update and return the persisted values, so a biography-only save cannot restore an older visibility choice. Refuse to publish a CRM person id that collides with another member id, and return 404 for an ambiguous path instead of rendering the wrong person. When profile settings fail to load, disable the form and show an error rather than defaulting visibility to public. Add coverage for partial updates and the 24-hex collision case. Refs #113 * chore(changelog): record profile hardening fixes Refresh the snapshot for the partial-update and CRM-collision fixes. Refs #113

    2767fcc ↗
  2. Featureprofiles,proposals
    Public member profiles and action-led policy changes (#117)

    * feat(profiles): add public member profiles and editable biography Build a viewer-relative profile read model over existing accounts, comments, proposals, amendments and branch memberships. Profile pages show an initials avatar, optional biography, geographic affiliation, contribution statistics and display-only comment karma, with paginated Comments, Proposals and Amendments tabs and deep links to the exact discussion anchor. Deleted comments, withdrawn amendments and proposals the viewer cannot open are excluded from history, counts and karma. Owners edit a plain-text biography through an origin-checked member route; no separate profile or contribution collection is introduced. Adds profile lookup indexes for comments and amendments. Refs #113 * feat(discussion): link authors to public member profiles Comment names and avatars, the proposal author and amendment authors now link to the public profile when a member id establishes authorship. Disabled accounts render as plain text and truncated historic records without a displayable name are not linked, so the interface never invents an author or offers a dead profile link. Top support and concern takeaways use the same link, and the proposal activity timeline targets the exact amendment anchor. Refs #113 * test(profiles): cover visibility, privacy, pagination and deep links Add in-memory read-model tests for private, branch and archived visibility, deleted comments, withdrawn amendments, legacy policyId comments, karma, deterministic pagination, disabled accounts, historic authors and owner-only biography writes. Add route checks for session and origin enforcement. Add a Chromium harness that follows an author link from a discussion, checks profile tabs and statistics, follows a comment deep link, verifies empty tabs and the 404 page on desktop and mobile, and wire it into CI after the production build. Refs #113 * docs(profiles): document public profiles, visibility and display-only karma Record the profile read model, owner biography field, viewer-relative filtering, profile indexes, biography route, browser harness, capability rows and the remaining privacy limits. Explicitly state that karma never affects eligibility, permissions or proposal authority. Refs #113 * test(profiles): cover the MongoDB karma aggregation Exercise the profile read model against MongoDB so the database-side karma sum is verified, not only the in-memory fallback. The suite skips when VOTING_TEST_MONGODB_URI is not configured and uses a unique database name. Refs #113 * refactor(profiles): rename comment karma to comment score Use the accurate product term for the sum of net scores on visible comments, and remove the always-visible explanatory note. Rename the internal field and function so the old term cannot return through future UI work. Refs #113 * feat(proposals): make policy changes follow the chosen action The policy-change form now treats amend, replace and repeal as distinct member tasks. The selected action drives the page heading, default title, primary question, effect timing and visible fields. Repeal asks for a rationale and hides commitments; an amendment can name the part to change; a replacement can carry proposed wording. Policy changes inherit the published policy's topic and governance context, so members no longer re-enter a topic or scope. Open change proposals for the same policy are shown before the form so members can join the discussion. The action-specific fields are stored with the proposal, included in the revision snapshot and locked once voting begins. Fixes the stale Amend title when Repeal is selected. Refs #114, #97 * test(proposals): guard action-led policy change form wiring Assert that the action selector announces the choice, the page updates the title and question from the chosen action, repeal hides commitments, and the form inherits the policy topic while showing existing change proposals. Refs #114 * chore(changelog): record public profiles and policy change UX Refresh the public snapshot for public member profiles and the action-led policy-change form. Preserve entries outside the generator's 80-commit window so the public history is not shortened. Refs #113, #114

    72c1cd4 ↗
  3. Featurediscussion
    Thread comments with Reddit-style voting and Best ranking (#110)

    * feat(discussion): render threaded replies with confidence-weighted voting Proposal discussion now stores parentCommentId and depth, renders one recursive comment component instead of a flat list, and keeps replies inside their parent's subtree when sorting. Deleting a comment that still has replies clears the body and keeps a placeholder so the thread does not lose context. Comments take upvotes and downvotes with one active vote per member. Responses return aggregate score, counts and percentage plus the caller's own choice; voter identities are never serialised. The default Best sort applies a Wilson score lower bound to every sibling group, with Top score, Newest and Oldest as alternatives and deterministic tie-breaking. Refs #57. * chore(ci): verify threaded discussion in a real browser Runs the isolated Playwright harness after the perimeter check. It proves parent-first nesting, sibling sorting, deep mobile layout, reply/delete interactions and that the comment API returns aggregate signal only. Refs #57. * docs(discussion): document threading, comment votes and Best ranking Updates the canonical model, API contract, capability matrix and known gaps. Records that comment votes are display ranking only, that voter identities are never returned, and the residual risks around rate limiting, audit and exact aggregate downvote counts. Refs #57. * chore(changelog): record threaded discussion and comment voting Refreshes the public snapshot after the discussion feature, CI verification and documentation commits. The generator's existing 80-commit window still drops older entries (#96); no history was edited by hand. Refs #57. * fix(discussion): make comment vote writes atomic under concurrency Adds a voteVersion compare-and-swap to comment vote updates. A concurrent vote bumps the version, so a stale write is retried against the latest state rather than overwriting it. Covers 25 simultaneous voters and same-member double clicks against both the in-memory database and a real MongoDB replica set. Refs #57. * chore(changelog): record atomic comment vote updates Refs #57. * fix(activity): show the comment's net score, not upvotes alone Member activity previously labelled a contribution with its upvote count, which is misleading now that comments can be downvoted. Show the net score as points, consistent with the discussion thread, and keep vote identities out of the feed. Refs #57. * chore(changelog): record the activity score fix Refs #57. * chore(changelog): regenerate after merging main * fix(discussion): guard tombstoned comments, empty menus, and vote actions

    9235057 ↗

15 November 2026

8 commits
  1. Featureadmin
    Administrator responsibility queue (#54) (#106)

    * feat(admin): add the administrator responsibility queue (#54) Group ballot outcomes, closing ballots, unbound wording, overdue advancement, amendment decisions, member verification and publication into one queue. Every item names the authority that makes it actionable and is filtered server-side by permission and branch scope. A failed source degrades to one short message for its group. * docs(status): record the administrator responsibility queue (#54) The queue is built and tested; multi-amendment derivation, collision review and role legibility remain outstanding. Note that publication failures have no durable failed state to queue. * chore(changelog): refresh the snapshot for the responsibility queue * fix(admin): never imply a ballot close date that was not recorded A legacy closed ballot with no close time now says only that it closed, instead of presenting the current date as the close date. * chore(changelog): refresh the snapshot for the queue close-date fix

    b2fd65a ↗
  2. Featureproposals
    Member-facing decision record (#54) (#105)

    * feat(proposals): add member-facing decision record Compose one matter's immutable voted wording, revision history, amendment outcomes, certified ballot result and publication state in a permission-aware read model. Derive adopted wording under D15 (zero applied amendments) and D16 (one applied amendment), with explicit review states for two or more applied amendments (D17-D18), approved-but-unapplied amendments and mismatched result wording. No individual votes, voter or author identifiers cross the read-model boundary. Add /proposals/[slug]/record/ and link it from the proposal page. * docs(governance): record the decision-record read model Document the member-facing decision record, the D15/D16 derivation, the explicit D17-D18 collision-review seam and the outstanding administrator queue. Mark S2/#54 partial and note the reporting limit on inter-amendment gaps. * chore(changelog): refresh the snapshot for the decision record

    be5f523 ↗
  3. Featureci
    Run the perimeter browser harness on every pull request (#38) (#104)

    * fix(security): bound the perimeter browser harness and always clean up The harness now has a hard 180-second deadline (ENGAGE_PERIMETER_TIMEOUT_MS) that stops the run and fails instead of hanging, signals the preview server before any await, escalates to SIGKILL, bounds Chromium shutdown, and kills the child from a process exit handler as a last resort. A page error, failed request or CSP violation that appears after a page's own check also fails the run. Adds 'perimeter:browser' for local use; a build and ENGAGE_PLAYWRIGHT_MODULE are still required. * feat(ci): run the perimeter browser harness on every pull request Runs scripts/security-perimeter-browser.mjs after the build and after the existing Playwright 1.57.0 install, reusing the same ENGAGE_PLAYWRIGHT_MODULE path rather than installing a second copy. Extends src/lib/ci-workflow.test.ts to compare step positions, so it fails if the step disappears, moves before the build or the Playwright install, installs its own Playwright, or if 'perimeter:browser' and the harness file drift apart. * docs(security): record that the perimeter harness runs in CI States what the CI step proves, how it is made safe to run unattended, and the remaining gap: the production build against the in-memory seed, not Cloudflare/Coolify, a real MongoDB or the real providers. * chore(changelog): refresh the snapshot for the perimeter CI step

    9478783 ↗
  4. Featurenotifications
    Durable outbox for ballot opened and closing soon (#33) (#102)

    * fix(notifications): write the delivery dedupe key the unique index requires The notification_deliveries unique index is on dedupeKey, but in-app and push delivery documents only set a composite _id. Every insert after the first therefore collided on a missing dedupeKey and was swallowed as a duplicate, so production could create at most one inbox item in total. Write the composite value to dedupeKey (and a push equivalent), keeping the existing _id dedupe contract and index shape unchanged. * feat(notifications): wire ballot opened and closing soon through a durable outbox Add notification_outbox, written immediately after a committed lifecycle transition and drained by npm run notifications:drain (dry-run by default, --apply to send). Two events are wired: a ballot opened by the lifecycle scheduler (new_ballots) and an open ballot inside the approved seven-day window (closing_deadlines). The audience is the voting electorate. Candidates are confirmed with getProposalEligibility and canViewProposal, the same predicate and visibility gate used when casting a vote, and eligibility is checked again at drain time so a revoked membership is suppressed rather than notified. Rows carry one unique dedupeKey per member, ballot and event; the drain claims with a conditional lease, retries at 1/2/4/8 minutes and stops at five attempts. Terminal rows carry a two-year retention TTL. Enqueue and delivery can never fail, block or alter a governance transition, and the scheduler report is unchanged. governance_outcomes and contribution_responses stay unwired: the result route is being refactored and the comment-outcome event needs a product decision. * chore(changelog): refresh snapshot for the notification outbox

    78c07c0 ↗
  5. Featuresecurity
    Response headers, hash-based CSP and cross-origin enforcement (#38) (#100)

    * feat(security): add perimeter headers, CSP and cross-site request rule Every response now carries nosniff, a strict referrer policy, frame denial, a minimal Permissions-Policy and a Content Security Policy. HSTS is sent only in production when HTTPS is provable. The policy keeps Astro's build-time script and style hashes and allows only the external hosts the app uses: Cloudflare Turnstile (script, frame, connect) and the first-party Plausible host (script, connect). Inline style attributes are allowed through style-src-attr only, because tally and progress bars use them; script-src has no unsafe-inline and no directive permits unsafe-eval. The service-worker registration was is:inline, which Astro does not hash, so it is now a processed script. Astro's form-post origin check is kept, and allowedDomains trusts the production host over HTTPS so its comparison uses the public origin behind the proxy. The middleware adds what Astro's check deliberately exempts: every state-changing method, JSON bodies, a cross-site Sec-Fetch-Site, an Origin that differs from the trusted public origin, and an absent Origin when a session or dev cookie is present. Bearer cron calls and webhooks carry no cookie and stay allowed. Refs #38 * test(security): verify the CSP and origin rule in a real browser The production build is served with a disposable in-memory database. Chromium checks the home, proposal, sign-in and Sense Box pages for CSP violations, console errors and failed requests; proves the Turnstile script and frame and the Plausible script and event endpoint reach the network layer; asserts the document and JSON header sets; and shows same-origin fetches and native form posts succeed while browser, JSON and cookie-without-Origin state-changing requests are refused. Cloudflare and Plausible are intercepted and answered locally; the script never contacts production. Refs #38 * docs(security): record the perimeter header set, CSP and origin rule Adds docs/security/perimeter.md as the operating contract for headers, CSP sources and the cross-site rule, with the exact values, every external host and its reason, the scoped style-src-attr allowance, verification commands and residual risk. Corrects the pages that said there is no middleware and that a CSP, HSTS, frame or content-type headers are absent, updates the Turnstile and API/authorisation contracts, and records the #38 and I10 state in the programme tracker and known gaps. Remaining work is explicit: proxy headers for static assets and Astro's early 403, a reporting endpoint, CSRF tokens, and production header observation. Refs #38 * chore(changelog): refresh snapshot for the perimeter security slice Refs #38 * docs(security): clarify origin-check exemptions and deployment requirements States the form-like and safe-method exemptions precisely, scopes includeSubDomains to the sending host rather than fusionparty.org.au, and records that production must configure PUBLIC_ORIGIN or BETTER_AUTH_URL or browser writes are refused against the proxy's localhost origin. Refs #38

    0aea7b5 ↗
  6. Featureops
    Durable job records and operational health checks (#39) (#99)

    * feat(ops): record authorised scheduled runs in a durable job ledger Add a job_runs collection with a 90-day TTL on startedAt and a withJobRun helper that records success or failure after the job finishes. A failed ledger write or summariser is logged and never changes the job result or error, so observability cannot break the thing it observes. Wire the governance scheduler (apply and dry-run), the ballot digest endpoint and the membership verification endpoint. Only authorised runs are recorded: 401 and 409 responses are not scheduled runs. Status codes, bodies, secrets, scheduling and governance writes are unchanged. Refs #39 * feat(ops): add exit-coded operational health checks and ops:health Evaluate governance scheduler recency and errors, ballot deadlines that passed without closure, digest cadence and held deliveries, verification recency and due backlog, recent failed or held email, and Turnstile configuration. Every threshold is named in the report and settable by environment variable; checks are count-only and never expose secrets, provider responses, member identifiers or ballot choices. /api/health now publishes coarse governanceScheduler and jobs states. A database failure stays 503, while a degraded job stays 200 so a container restart loop cannot hide a missing external scheduler. npm run ops:health prints JSON (or --human) and exits 0 healthy, 2 degraded or unknown, 1 failing or unreachable, requiring MONGODB_URI. Refs #39 * docs(ops): record job cadence, health command and runbook Document the job_runs ledger, the documented cadence and stale thresholds for each scheduled job, the ops:health command and exit codes, what each degraded state means, the maintainer escalation placeholder and the runbook actions. Record explicitly that production cron wiring and production verification are not done, so #39 remains partial. Update the tracker I10/S8 rows and known gap 21. Refs #39 * chore(changelog): refresh snapshot for the operational observability slice Refs #39 * fix(ops): warn before a scheduled job reaches its fail threshold A 15-minute job judged healthy for 24 hours leaves a full day blind to a complete outage. Each time-based check now has two tiers: degraded past warn (governance scheduler 2h, digest 1h, verification 2h) and failing past fail (26h, 24h, 24h). Within warn stays ok; a missing run stays failing; a failed last run stays failing. Warn and fail are configurable with the existing fail variables plus GOVERNANCE_SCHEDULER_WARN_HOURS, DIGEST_WARN_HOURS and MEMBERSHIP_VERIFICATION_WARN_HOURS. A warn value that is not below fail reports unknown with the conflicting values instead of trusting either. The report's thresholds block names both values and their source. Check ids, report shape, the ledger and route behaviour are unchanged; ops:health still exits 2 degraded and 1 failing. Refs #39 * docs(ops): document two-tier warn and fail job thresholds Show cadence, warn and fail for each scheduled job, explain that warn is a small multiple of cadence (not one missed tick) and why a 24-hour-only alert hid 96 missed 15-minute ticks, and record the invalid-pair behaviour, per-state meanings and escalation timing. Refs #39 * chore(changelog): refresh snapshot for the two-tier thresholds Refs #39 * fix(ops): report a failing job as failing on the health probe * chore(changelog): refresh after rewriting the synthetic token fixture

    23f88d8 ↗
  7. Featureballots
    Validate recorded outcomes against the bound tally (#28) (#86)

    * feat(ballots): validate recorded outcomes against the bound tally Record a formal ballot outcome only after comparing the officer's result with the outcome computed from the votes bound to that ballot and immutable revision (#28). The comparison is deliberately limited to what the decision register settles: Yes greater than No approves and a tie does not (D9); abstentions are recorded and excluded rather than given a denominator (D10, O1); no quorum, turnout or minimum-participation threshold applies (D8). The basis is versioned as `simple-majority-d9-v1`. A contradicting result is refused unless the officer supplies a written reason, and is then stored as an explicit override. The result keeps the tally at certification, the rule basis and the computed result; the `decision.recorded` event records the actor, reason, computed result and recorded result alongside the existing atomic result and ballot certification. Identical retries remain idempotent, conflicting replacements are refused, and historical results are returned unchanged and never rewritten. Both admin result forms now show the computed outcome and its tally before submission and require the override reason when the chosen result contradicts it, using the shared Field, Textarea and Badge primitives. * test(ballots): cover computed outcomes, overrides and stored evidence Add rule-level cases for approval, rejection, a tie, zero votes and abstention-heavy ballots, plus the two admin form behaviours that reveal and require an override reason. Add real-MongoDB replica-set coverage for the parts that prove persistence and the ordered result write: the tally and rule basis stored on the result, rejection of a contradiction without a reason, an audited override, idempotent retries, refused replacement, encrypted-ballot certification through the decryption path, and a historical result that stays readable and unrewritten. The result API is exercised end to end with the dev session. Refs #28. * docs(governance): record computed ballot outcome validation Describe the computed Yes-versus-No check, its `simple-majority-d9-v1` basis, the stored tally and the audited override in the lifecycle, API and agent guides. Mark known gap 10 as partly resolved and keep #28/L1 Partial because the abstention denominator (O1), quorum and ranked choice remain undecided. * chore(changelog): regenerate the public changelog snapshot * feat(ballots): certify only a closed ballot from a consistent bound tally Refuse certification while a ballot can still accept votes. A result is accepted when the ballot has a closure marker, its close time has passed, or the system has already closed it; a live ballot returns BALLOT_NOT_CLOSED and a ballot with no recorded close is refused rather than assumed closed, because a later vote could contradict an early certification. Tally only votes bound to the ballot and exact revision, and treat the rest as a ballot defect instead of ignoring them: unbound or mismatched-revision votes now return BALLOT_TALLY_UNBOUND_VOTES, unreadable encrypted votes return BALLOT_TALLY_UNREADABLE, and missing or invalid decryption keys return BALLOT_DECRYPTION_UNAVAILABLE (503). The tally query reads only the fields the check needs. Reject an override reason longer than 1000 characters instead of truncating audit evidence, and treat a retry that changes the recorded override reason as a conflict rather than an idempotent success. The refusal message now names the computed outcome and says that any other result, including inconclusive, requires a written reason. Refs #28. * feat(admin): show the enforced tally and recorded outcomes Derive the admin outcome basis from the same bound tally that certification enforces, so a page can no longer show a tally the server will not count. A zero-vote ballot reads "No votes yet — outcome not yet computable" instead of "Unsuccessful", and an unbound or unreadable tally is named instead of guessed. An open ballot shows when voting closes instead of an unusable result form; once a result exists both admin surfaces render the recorded outcome, computed result, tally and override reason instead of a re-submittable form. Revealing the override field moves focus to it and announces the requirement to assistive technology. Show overridden outcomes in Ballots & decisions with the reason available, and add the computed result, rule basis and override to the proposal PDF report. Refs #28. * test(ballots): cover closure, the bound set and override integrity Close fixtures exactly as the scheduler does before certifying, and pin the override rule in the two pre-existing suites by asserting the refusal when the reason is removed rather than only supplying one. Add replica-set cases for a live ballot that cannot be certified and the changed closed tally that is then certified, a vote on another revision or with no revision, amendment-ballot votes excluded from the matter tally, an unreadable encrypted vote, missing decryption keys, an over-length override reason and a retry that rewrites the recorded reason. Cover the form focus/announcement behaviour and the PDF computed/override rows. Refs #28. * docs(legal): correct the terms to describe computed ballot outcomes Terms of Use §9 claimed Engage "does not calculate or apply a passing threshold", which is no longer true. Describe the versioned simple-majority-d9-v1 rule, the tie and zero-vote result, the treatment of abstentions, the absence of quorum or turnout, and the explicit, reasoned and audited override. The acceptance gate stays inactive: the document remains owner-authorised with no effective date. The incorrect sentence was in the Terms of Use, not the Privacy Policy, so the Privacy Policy and its version are unchanged. Roll the Terms to version 1.1 and record the correction and the #40 follow-up in the privacy review. Refs #28, #40. * docs(governance): record closed-ballot certification and the bound tally Document that certification requires a closed ballot, that unbound votes and unreadable votes are refused, that inconclusive is always an explicit override, and that amendment endorsement ballots remain officer-supplied because O2/D18 are undecided. Correct the failure table so a zero-vote unsuccessful result needs no override, refresh the tracker, known gaps and handover status, and point the agent guide at the corrected Terms section. Refs #28. * chore(changelog): regenerate the public changelog snapshot * test(ballots): warm the route database before asserting on it The result-API case resolved the application client only inside the route. A transient connect failure could make the development fallback serve the in-memory database, so the route could not see the fixture and the test failed with a confusing authorisation-style error. Resolve the application client in the test and assert the fixture is visible before exercising the route, so a real connectivity failure fails loudly instead of degrading silently. Refs #28. * chore(changelog): record the review fixes in the public snapshot * fix(ballots): judge each ballot's own close time and keep zero-vote outcomes computable Read the addressed ballot's row before the matter so a second, still-open ballot on a closed matter is refused instead of inheriting the matter's closure. The matter fields remain the fallback for legacy ballots with no ballot row. Return the computed outcome for every computable tally, including a zero-vote ballot (still "unsuccessful" under D9), so the UI can require an override instead of silently withholding the reason field. Separate a missing key ring from unreadable votes: the display summary now reports `decryption-unavailable` on its own, with unbound votes stated first because that is the cause an officer can act on. Add a tolerant key loader for read-only paths so a broken key ring no longer fails a page that renders no encrypted votes; signed-in list and detail reads use it, while casting and certification stay strict. The result route now reports a clean BALLOT_DECRYPTION_UNAVAILABLE (503) instead of leaking key-configuration detail. Refs #28. * fix(admin): keep the visible result status and render every tally state The hidden override announcement was the first `[role="status"]` in the result form, so `querySelector('[role="status"]')` returned it and every success or refusal was written to an invisible node. The announcement now uses `aria-live="polite"` without `role="status"`, and both submit handlers target the visible line through `formStatusRegion`, so feedback remains sighted-visible and still announced. Show the computed outcome for a zero-vote ballot instead of claiming it is not computable, which also restores the override field for `Successful` and `Inconclusive`. Add explicit states for unavailable decryption keys and separate unreadable votes, and withhold the form while the tally is defective rather than offering an action the server will refuse. Refs #28. * test(ballots): cover per-ballot closure, zero-vote overrides and the status line Prove an open second ballot on a closed matter is refused while its closed sibling certifies, and that a zero-vote ballot records successful only with a written reason. Pin the tally-state precedence (unbound before unreadable), distinguish missing keys from unreadable votes, and assert display key loading degrades to null instead of throwing. Guard the UI regression: the announcement is not a `[role="status"]`, both pages use `formStatusRegion`, and that helper returns the visible line when the hidden node comes first in the form. Refs #28. * docs(governance): correct the legal pointer and merge wording The privacy review said the Privacy Policy describes the computed rule; the Privacy Policy is deliberately unchanged and the correction is in the Terms of Use. Correct that sentence, rename the lifecycle row so it describes the zero-vote case rather than implying every untallied result needs an override, qualify the display-versus-enforcement claim, and state per-ballot closure. Align the tracker and handover on "integrated, branch only; not merged". Refs #28, #40. * chore(changelog): record the second-pass review fixes * chore(changelog): include the final verification commits * fix(ballots): let display reads degrade instead of failing on unreadable votes Add an opt-in `tolerateUnreadableVotes` read mode to `listProposals` and `getProposalDetail`. Admin pages use it so a broken key ring renders an explicit "decryption unavailable" state instead of returning 500, and affected rows carry `unreadableVoteCount` so the missing votes are counted and shown rather than quietly dropped. Every other read keeps today's loud failure, and certification is untouched. Refs #28. * fix(ballots): key results by ballot, share closure state and reject stray override reasons Add `proposalBallotResultFilter` so the result attached to a matter's primary ballot is addressed by `ballotId` (legacy results without a ballotId still resolve). Apply it to the admin workspace, the proposal PDF, the public proposal and policy pages and the matter read model, so a result on one ballot no longer shadows another. Add `resolveBallotClosure` as the single closure rule used by the server gate and the pages: the addressed ballot's own status/close time first, the matter as the fallback. Add `ballotOutcomeLabel` for consistent title-case rendering and fix the quadratic vote grouping in `readBoundBallotTallies`. Refuse a written `overrideReason` when the result already matches the computed outcome (`RESULT_OVERRIDE_REASON_UNEXPECTED`, 400) instead of accepting it once and then failing an identical retry, so a reason is never silently discarded. Refs #28. * fix(admin): derive the result and closure state from the balloted record Use the shared closure rule on both canvassing surfaces and the oversight page so the form is offered exactly when certification would accept it, including the case where a sibling ballot's close time differs from the matter's. Look up recorded results by the primary ballot, show "N unreadable" when the display tally is partial, and render computed outcomes in title case alongside recorded results. Refs #28. * test(ballots): cover closure precedence, per-ballot results and stray reasons Add pure cases for closure precedence (ballot row before matter) and outcome labels, and replica-set cases proving an unnecessary override reason is refused rather than silently dropped, that display reads tolerate unreadable votes while strict reads still fail, and that primary and sibling results resolve to their own ballots. Refs #28. * docs(legal): scope the outcome rule to ordinary Yes/No ballots Say that the computed simple-majority rule applies to ordinary Yes/No ballots on a matter, and that amendment endorsement ballots follow a separate procedure recorded by an authorised officer. The publication status and the inactive acceptance gate are unchanged; this stays inside the existing unreleased Terms version 1.1. Refs #28, #40. * docs(governance): record stray-reason refusal and display degradation Document that a reason supplied for a matching result is refused rather than dropped, that a broken key ring degrades admin displays to an explicit unavailable state with unreadable votes counted while certification stays strict, and keep the Terms-of-Use pointer and per-ballot closure precedence accurate. Refs #28. * chore(changelog): record the merge and third-pass fixes

    762f76c ↗
  8. Featuredocs
    Check documentation links in CI (#44) (#85)

    * fix(docs): repair broken relative links Point the deployment verification record at the canonical known-gaps page instead of the removed mvp-roadmap, and drop the link to the uncommitted UI-refinement screenshot directory. * docs(guides): point agents at document status labels Replace the stale-page list with a reference to docs/README.md, which defines the document status labels and canonical replacements, and list the new link check among the verification commands. * feat(docs): check documentation links in CI Add a dependency-free script that resolves relative links in docs/, README.md and AGENTS.md, ignoring external, anchor-only, template and fenced-code links. Wire it into the existing application check job as docs:check-links.

    2a52ef4 ↗

13 November 2026

4 commits
  1. Featureplatform
    Usability, governance lifecycle, contributor feedback & admin reporting (#20)

    * feat(platform): redirect reclaim.fusionparty.org.au to Engage The Reclaim policy site is retired. A Worker bound to the reclaim custom domain now answers every request with a 301 to the same path on engage.fusionparty.org.au, preserving query strings so Engage's existing legacy redirects (pending-vote, all-policies, reclaim-our-*) still resolve. The previous reclaim Worker is left deployed for rollback. * chore(changelog): record Reclaim domain redirect * feat(platform): map Reclaim division paths to Engage proposals Old links to /reclaim-our-economy/<proposal>/ and /reclaim-our-future/<proposal>/ now land on /proposals/<proposal>/ instead of a 404 on Engage. Retired section index pages, whose names are not proposal slugs, resolve to the proposal list. Other paths and query strings are unchanged. * chore(changelog): record Reclaim path mapping * feat(discovery): surface Sense Box and next actions when caught up on ballots * feat(policies): link empty policy registry directly to fusionparty.org.au/policy * feat(governance): add contributor feedback, proposal supersession, and resolution lifecycle support * feat(admin): add ballot progress, branch summary, and privacy-preserving CSV export * chore(changelog): record usability, governance and reporting updates * fix(governance): address review findings and add Reddit-style comment composer * chore(changelog): record review fixes and comment composer update * feat(notifications): surface contribution responses and catch-up alerts in header and My Engage * chore(changelog): record in-app notifications update * fix(governance): address PR #20 CodeRabbit review comments * chore(changelog): record PR #20 review fixes * docs(research): compare participatory governance patterns with Engage * docs(research): record verification and refresh changelog * fix(governance): address PR #20 review findings * chore(changelog): update snapshot after main merge * docs(research): validate governance patterns with REA runtime evidence * chore(changelog): record REA governance research

    94c0caa ↗
  2. Featuresense-box
    Offer accessible geographic destination picker
    5549e0c ↗
  3. Featurebranches
    Let verified members manage branch membership
    e839386 ↗
  4. Featuresense-box
    Collect private branch community priorities
    8ec0cf1 ↗

9 November 2026

6 commits
  1. Featureproposals
    Implement approved reading and voting journey
    b5207f2 ↗
  2. Featureproposals
    Connect policy reading with voting and simplify discovery
    fc8d130 ↗
  3. Featureproposals
    Simplify discovery and expose ballot progression
    1c13678 ↗
  4. Featureadmin
    Add audited platform settings and branch hierarchy validation

    Replace the raw JSON platform settings form with a task-based settings workspace. Administrators can manage visibility, public features, social channels and the branch website directory, with server-side URL validation, read-only infrastructure notes and a recorded change history that supports restoring earlier values. Enforce branch parent kinds on the server so states and territories sit under National and local branches sit under a state or territory. Add coverage for the settings API, audit history, restore behaviour and branch hierarchy rules.

    5bb7e82 ↗
  5. Featureadmin
    Add durable database migrations and standard branch seed

    Add Admin -> Migrations for national administrators. The page previews each additive migration, shows the last recorded run and any error, and applies Standard branches or Branch participation individually or in order. Every run is recorded in data_migrations with run count, timestamps, actor and result. Standard branches is an idempotent seed for Fusion National, VIC, NSW, QLD, WA, SA, TAS, ACT and NT with National as parent; existing branches are never renamed, re-parented or un-recognised. Branch participation remains blocked until the National branch exists. The command-line seed now uses the same service, and the participation migration remains available for deployment automation.

    e2137a5 ↗
  6. Featurebranches
    Separate branch following from verified membership

    Members choose the branches they follow during onboarding and in My Engage. Authorised administrators verify followers as members in a focused Branch governance workspace. Branch subscriptions are now Engage-owned preferences, while verified organisational membership stays in branch_memberships with source, verification and audit metadata. National governance is included for every verified account and never removed by branch preferences. Existing members without branch connections receive national scope silently and a one-time state branch invitation. Eligibility is decided in one service used by proposal pages, ballot submission, pending-vote promotion and digests. Followers can discover branch activity but cannot vote. Revocation requires a reason and preserves historical votes and the person's following preference. The migration backfills legacy membership metadata, adds subscriptions without duplicating them, assigns open proposals without branch context to national governance, and leaves branch-owned and archived records untouched. The Underground CRM boundary is implemented through applyExternalMembership: external membership facts can update organisational membership but never branch subscriptions, and local administrators cannot silently override CRM-sourced records. Tests cover national defaults, follow/unfollow idempotency, source integrity, permissions, migration, parent/child eligibility and new proposal defaults.

    a727732 ↗