Files
pages/docs/design/implementation-roadmap.md
T
Anya Tokugawa 19e12f96a8
Build and publish container images / images (push) Successful in 13m39s
Release container images / release (push) Successful in 18s
Add private CardDAV synchronization and contact revision recovery
2026-11-02 08:41:15 +09:00

48 KiB
Raw Permalink Blame History

Architecture Implementation Roadmap

Task productivity and private contacts (2026-11-01)

  • Add idempotent task capture API and neomutt configuration reference.
  • Extend pages-cli with task planning, start/stop records, reports and private contact lookup.
  • Add per-user defaults, accomplishment heatmap and opt-in daily SMTP task digest.
  • Add home search commands with project/calendar suggestions and explicit creation confirmation.
  • Add private contacts CRUD, trash/restore, vCard import/export and unknown-property preservation.
  • Add private CardDAV serving with isolated credentials, conditional writes, bounded reports and sync tokens; revision preview/restore and whole-book export (2026-11-02).
  • Shared address books, external CardDAV connectors, physical-device compatibility, work-record corrections and manual digest retry.

See CardDAV operation and limits. The private profile is not full CardDAV compliance; production rollout is tracked separately.

See contracts and remaining scope. These checks describe the implemented slice; production rollout and test evidence are recorded separately.

Current highest priority (2026-10-31)

The user-prioritized daily UX / browser / editor / CLI backlog takes precedence over the feature backlog below. It covers type-to-search, virtual Path navigation and browser commands, editor/plain-text and code styling, theme switching, session timeouts, contacts, City labels and a session-aware terminal client. Preserve authorization and storage safeguards. Track implementation, tests and deployment separately; a source checkpoint includes the pre-existing dirty worktree.

Principles

Named architecture components (2026-11-01)

  • Adopt stable responsibility-based names for eight components; preserve UI terminology, public APIs, persisted IDs, event consumer checkpoints and deployment boundaries.
  • Extract Grammaratur: ordinary and protected authoring conversions, compatibility facades and application connections. Verification is recorded separately from deployment.
  • Extract Dregarnuhr clock/loop/admission/retry contracts and the queue adapter; register the 14 existing business handlers outside the core. Preserve IDs/schema and atomic claims.
  • Add Dregarnuhr expiring leases, fencing, multi-process scheduling and application shutdown integration; these are not provided by the initial extraction. See contract and remaining work.
  • Extract Geduldh local payload storage and bounded materialization; preserve the Document Store contract through aliases, connect application consumers, and test layout/Range/backfill/DR compatibility. External backends and placement changes remain separate work; see contract and remaining work.
  • Extract Mestionora SQLite search/index/City lookup with explicit read-policy callbacks, legacy aliases, candidate/snippet budgets, plan and authorization tests; see search contract.
  • Preload lazy user-space read policies before Page search; retain snapshot consistency while removing nested connection requirements. Keep current-only/Domain-partitioned and language-aware indexing as later migrations.
  • Extract Ordoschnellic bounded Web Push dispatch, retry policy and encrypted transport; preserve transactional Inbox enqueue, authorization, event IDs and subscription checkpoints. See delivery contract and remaining work.
  • Add distinct Push claim generations and subscription-version fencing, Retry-After and delivery metrics. Initial extraction retains existing retry/lease limitations, not exactly-once delivery.
  • Extract Schutzaria pure Path/locality/group/publication/user-space evaluation and approval state transitions; retain authenticated fact loading, alias checks and transaction ownership in the application. See policy contract and remaining work.
  • Complete consistent authorization snapshots, lazy user-space preloading, reviewer inheritance/quorum, and publication revalidation. State-machine extraction alone does not enable the reviewer workflow.
  • Extract Kunstzeal built-in adapter selection, bounded Kroki rendering and diagram hash/cache; retain source authorization/CSP and add redirect refusal plus atomic cache replacement. See rendering contract and remaining work.
  • Add provider-versioned cache keys, render concurrency/coalescing, unified disable/metrics and additional diagram/slide providers. Initial extraction does not implement nwdiag or slides.
  • Extract Seheweit resource collection, typed snapshots and history window/retention policy; retain APIs/schema/event IDs, connect foreground and background paths, bound proc reads, guard storage arithmetic, propagate write cancellation and reject read-only sample writes. See observation contract and remaining work.
  • Add source availability, bounded/coalesced collection, rollups/retention budgets, component/provider metrics and instance-scoped HA/SNMP collectors. These are not part of the initial extraction.

See component names, boundaries and acceptance criteria. All eight initial component extractions are implemented; unchecked items describe remaining capabilities. Released and deployed as v0.2.0 on 2026-11-01.

  • Priority follow-up: eliminate AddEventLog deferred read-to-write transaction contention; preserve atomic raw-log/bucket writes and test independent concurrent writers. Production start-log failures are now observable; job execution succeeded, but log completeness is not guaranteed.

Storage portability and operations (2026-09-27)

Implemented and fixture-tested; included in the v0.2.0 production release. This does not mark the unchecked backend/HA capabilities below as implemented.

  • Immutable local Document Store contract, bounded materialization and streaming archive access.
  • Resumable document backfill with store identity, UUID cursor and batch checkpoints.
  • SQLite migration ledger with version/checksum, atomic adoption and current-schema assertions.
  • Managed projection, resource, broadcast, navigation, adapter-session, import and event-queue schemas.
  • Read-only pageswiki schema status|check with JSON output, pending steps and explicit legacy exclusions.
  • Remaining legacy schema/repository contracts and context propagation; configurable materialization limits/metrics.
  • PostgreSQL/MySQL, S3/SFTP and mounted-store contracts, optional Redis cache.
  • Coordinated backup/restore, portable export/import, Web operations and initial setup wizard.

See the Japanese design and acceptance criteria and preflight / deployment procedure. Passing managed-schema checks does not replace an isolated full-startup rehearsal.

Task management first slice (2026-09-24)

2026-09-25 task CalDAV (0.8.0, implemented and fixture-tested, deployed to app/app2 on 2026-09-26 at 00:24 JST; deployment record): private fixed VTODO collection, revocable read/read-write app passwords with UI, conditional writes and archive deletion. Real vdirsyncer 0.21.0 / Todoman 4.7.0 discovery and bidirectional sync tests. Date-only deadlines; not full CalDAV conformance. See task CalDAV contract and CLI setup.

2026-09-25 task follow-up (0.7.0, implemented and deployed to app/app2 at 07:07 JST): daily/tomorrow planning, explicit carryover rescheduling, hierarchy-aware pagination and parent context, and atomic Wiki-note creation with UUID-bound view/edit actions. See daily task / Wiki note contract. Capacity, bulk family planning, manual sorting and shared tasks remain open.

2026-09-25 follow-up (implemented, fixture-tested and deployed to app/app2 at 05:59 JST; deployment record):

  • Inline title/status/date editing, independently dated two-level subtasks, bounded monthly day schedule, home shortcut and Today navigation modal (0.6.0).
  • Inbox presentation preferences per account/browser, focused expiring toasts, opt-in return modal, sound presets/volume, notification-only Service Worker.
  • Opt-in Web Push: transactional Inbox outbox, revocable session-bound subscriptions, VAPID, vendor allowlist/DNS pinning, bounded scheduler delivery/retry/expiry and focus acknowledgements.
  • Live OS/browser delivery acceptance with an explicitly opted-in test device (Chrome/Firefox/Safari), iOS home-screen installation and OS power-saving behavior.

See active notification contract and inline task/day schedule contract.

  • organizer.tasks private per-principal/Domain items and projects, CAS payloads, atomic head/revision/change journal, conditional writes and idempotent creation.
  • Today, Inbox, Backlog, Someday, Review, Completed and Archive; project/task detail, quick capture, due/planned/review dates, priority, estimate and readable Wiki UUID links.
  • Bounded lists/history, EN/JA, theme-aware desktop/mobile dialogs, extension kill switch.
  • Typed independent outbox subscriptions; atomic task/project events, bounded journal migration through the scheduler, no City jobs for task-only changes.
  • Shared collections/assignees, export/recovery and bulk review.
  • Opt-in due/planned-date task reminders, atomic reservations/cancellation, bounded background Inbox delivery, live metadata authorization and badge refresh.
  • Private local calendar time-block links, durable retry intents, current event time/ETag lookup, unlink tombstones; imported/shared/public calendars excluded.
  • Persistent Pomodoro: start/pause/resume/finish/cancel, one active session per principal, conditional transitions, bounded scheduler completion and Inbox notification.
  • Today bounded estimates/current local time blocks/recorded focus intervals, overlap and incomplete-total indicators. Calendar event summaries include an explicitly approved task-title snapshot, never notes.
  • Personal capacity, carry-over and period/project reports; historical focus-span backfill; calendar reminders, notification defaults/quiet hours/digests; Pomodoro personal defaults/cycles and routines.

See time-block and Pomodoro contracts for API, privacy, two-store recovery, polling limits and remaining work. Canonical calendars and task DB metadata both require backup. Production app/app2 deployment completed on 2026-09-25 JST; see the deployment record for backup, verification and the observed transient scheduler lock contention.

  • Limited private VTODO / CalDAV with vdirsyncer + Todoman file synchronization.
  • Time-of-day deadlines, recurrence/alarms, shared task collections, incremental sync-token and additional CalDAV clients; dependency graph, tickets and offline UI.

See Japanese task implementation contract. The existing /v/plan calendar default is unchanged; tasks start at /v/plan/today.

Calendar subproject (2026-09-22)

  • Small-screen/touch calendar navigation, readable week columns and day selector, scroll-first gestures, compact month counts and VisualViewport-aware editor. Mobile contract: fixture-tested and deployed 2026-09-25; iPhone Safari acceptance pending.

  • Initial file-canonical calendar/month snapshots, rebuildable SQLite index, conditional event CRUD and bounded CalDAV time-range REPORT.

  • /v/plan year/month/week/day/events/timeline, event dialogs and per-calendar collaborators.

  • Per-user default calendar overlays and explicit comma-separated pretty_name selection.

  • Week/day time grids, configured-timezone day boundaries, zoom through 400% with 5-minute precision, ETag-protected move/start/end resizing without refetch, equal overlap lanes, drag-range creation, horizontal rescheduling and week/list.

  • Read-only ICS polling with interval/conditional GET and bounded external CalDAV REPORT/multiget; background event queue, encrypted Basic credentials, SSRF guards and failure-safe snapshots.

  • Outlook published-ICS projection for METHOD:PUBLISH, VTIMEZONE/TZID, Windows timezone names, bounded RRULE expansion, EXDATE/RDATE, moved/cancelled RECURRENCE-ID instances and orphaned exception projections, plus multi-megabyte synthetic fixtures.

  • Recurrence PERIOD/THISANDFUTURE, alarms/durations, provider validation, discovery/OAuth, configurable synchronization windows and conflict-aware bidirectional synchronization.

  • Full iCalendar temporal semantics (recurrences, timezones, exceptions), CalDAV discovery/client compatibility, application credentials, synchronization tokens and maintenance/metrics jobs.

  • Group sharing, audited administration, ownership recovery and calendar scheduling/notifications.

See Japanese implementation contract and limitations. The initial supported subset is not a claim of full CalDAV conformance.

System-service plugins are distinct from document adapters. See System extensions for implementation boundaries and the next slices.

  • Responsive wider bookmark-card grid (viewport-dependent columns).
  • Built-in resource.sample event, SQLite history, retention and bounded graphs without foreground collection.
  • Separate system-extension catalog and settings area for trusted built-ins.
  • Unified execution/resource monitor (running jobs only), paginated job history with lazy logs.
  • Protect group owners from ordinary member downgrades and atomically retain an owner during concurrent removal.
  • Transactional activity outbox and Domain/principal recent-edit projections.
  • Diary/blog initial built-ins: Page UUID metadata, calendar/recent entries, categories, read-filtered indices, per-Domain enable/disable and registration UI. See initial scope; this is authenticated, existing-Page registration only.
  • Diary/blog atomic Markdown new-post composition, transactional metadata/Revision, embedded-asset freezing, idempotent retry and create/write enforcement.
  • Diary/blog: anonymous metadata approval/feeds, recent comments and safe resumable tDiary import; declarative time/OS/sunset theme rules.
  • Diary/blog per-Domain prefixes, optional keyword URL planner, multiple articles per day, and authenticated Domain-root diary/blog entry view (2026-09-17). Legacy Page registration remains available.
  • Domain entry-settings dialog and signed manifest v2 provisioning of layout, diary/blog enablement and prefixes; atomic destination apply and managed edit guards.
  • Signed enrollment capability proof and Campus-wide approved-peer v2 preflight before entry-setting changes; unknown/legacy peers block with a visible explanation.
  • Entry provisioning inside Domain creation, controlled-settings release/retention, applied-setting acknowledgments and freshness policy.
  • Diary/blog dated viewer routes /v/{kind}/yyyy/mm[/dd], own presence filter, legacy-query redirects, and canonical calendar/filter/pagination navigation.
  • Local same-Campus diary/blog index/calendar projection without copying articles; source and destination read-policy filtering, source body/edit links, target write guard.
  • Project article bodies on destination host, approved anonymous projection, cross-instance source transport/revocation and source-setting federation.
  • Rankings, reactions and moderated Page/inline comment services.
  • Linkding-compatible bookmark-source injection with scoped credentials.
  • Explicit external-search fallback and PagesWiki/GROWI/PukiWiki connectors.
  • S3-compatible Document Store; read-only WebDAV/SSH external collections.
  • Optional SNMP resource collector with target/OID allowlists and SNMPv3 secrets.
  • Third-party system-plugin signature validation, isolated runtime and lifecycle operations.

vim.wasm runtime fixes are on hold by user request; its existing configuration UI is retained.

Prioritized execution order (2026-09-14)

Feature checkboxes above are not an instruction to bypass the hardening gates. The execution order is maintained in Implementation status.

  1. Completed this slice: unified monitor/history UI, bounded/indexed job/log retrieval; first P0 correction: owner downgrade and concurrent last-owner removal protection.
  2. Completed bounded hardening: OIDC/JWKS, inherited-admin revocation, browser mutation/purpose checks, Domain/Object scope and read filtering; see the execution ledger.
  3. Completed recursive authorized snapshots, literal prefix transactions and signed manifest/checkpoint atomicity. Remaining write-policy fencing and content-author trust are gates before public/shared collaboration.
  4. Completed PDF/archive metadata jobs, bounded read-policy snapshots, Object/Revision pagination and improved search; continue lazy City/directory projection, decoder isolation and durable entry extraction.
  5. Transactional activity outbox, actor attribution, idempotent URL reconciliation and recent-edit projections.
  6. Diary/widgets/import and themes, then comments/rankings, external search/bookmarks, storage connectors and SNMP. HA leases/fencing gate multi-node deployment; external plugin installation remains disabled until isolation/security gates pass.
  • Migrations must preserve existing UUID and path reachability.
  • New architecture is introduced in vertical slices with regression tests.
  • Reserved authorization features must not be simulated with insecure allow-all placeholders.
  • New collaboration and extension features must use the stable Page UUID, immutable Revision, Domain-scoped policy, projection-adapter, event, and scheduler contracts described in Platform Extension Architecture.
  • SQLite remains a supported single-node metadata backend. HA work must add a replicated metadata strategy or explicit single-writer fencing; sharing a live SQLite file over a network filesystem is not an HA design.

Phase 1 — Identity and projection foundations

  • Separate Object creation from Semantic Path binding.
  • Rename the implemented District model to Domain.
  • Separate Revision UUID from Document SHA-256.
  • Add Revision language metadata and fallback selection.
  • Move immutable payloads to documents/sha256/ with a transitional SQLite compatibility cache.
  • Add rsync-like and interactive path: / uuid: / rev: / asset: data administration.

Phase 2 — Knowledge City topology

Current review: City reconciliation and navigation. The completed items below are initial implementations. The next City slices are:

  • Audit City/autoplan/gen-city and design URL-driven event reconciliation.

  • CITY-A named links: parent-qualified Street URLs, ambiguity-safe lookup, shared search/map/macro links, ancestor authorization, lightweight details, literal Page URLs and map focus/locality-search regression coverage.

  • CITY-A: per-result authorization, literal Paths, scoped links/names, shallow/portal inference, manual ownership and idempotent application.

  • CITY-B: transactional mutation outbox → debounced, durable City queue consumer; scheduled/manual repair through the same pipeline.

  • CITY-C: stable fabric, Worker/cache, connected roads, bounded lazy projection and capacity-safe picking targets.

  • CITY-D: icon/modal creation/settings, arrow/WASD/HJKL navigation, focus/history handling and i18n browser regressions.

  • City label follow-up: visible resting labels, horizontal Street names, retained hover/focus highlighting and working label toggle; normal/popup tests.

  • Manual City organization: display-title modal, Ctrl/Command selection, existing/new Spot merge, Street relocation retaining Page URLs, transactional validation and manual-inference overrides; see contract.

  • Immersive City follow-up: focused Spot courtyards and surrounding Page buildings, nonoverlapping bounded/paginated close-ups, themed architectural SVG, modal utility menu and directory navigation; normal/popup regressions. See projection and limits; CITY-A–D gates remain open.

  • Filter City map/details/macros/badges by readable boundaries and Pages; recompute visible portals/counts/relations and use private responses.

  • Local gen-city Worker with time/input/output/cache bounds, cancellation and retry, plus spatial Arrow/WASD/HJKL navigation in normal/popup modes; see worker and navigation.

  • One-connection-safe bounded City snapshots and literal Path containment/overlap/context/health; true lazy Page pagination remains open.

  • Define Campus identity and defaults.

  • Define Domain ownership by Campus.

  • Scope Semantic Paths by (domain_uuid, path).

  • Add local/remote Domain placement.

  • Add absolute cross-Domain link helpers.

  • Add standalone/controller/managed Campus authority modes.

  • Scope non-topology configuration by Domain.

  • Scope Recent Objects by selected Campus and Domain.

  • Record controller-selected Domain instance FQDN.

  • Add hashed enrollment tokens and fingerprinted instance registration.

  • Add pending/approved/revoked instance placement selection.

  • Add external Campus bootstrap enrollment.

  • Add Managed Campus health status and detach actions.

  • Add signed Campus manifest reconciliation.

  • Implement ZoneTree nodes for Quartier and Street.

  • Implement Spot identity and ZoneTree pointers.

  • Implement SpotGraph Page bindings and MagicPage traversal.

  • Define Quartier planning, Street routing, Spot locality, topology invariants, projections, APIs, policy inheritance, and delivery slices.

  • Add the initial topology schema, monotonic Domain topology version, topology events, and Quartier/Street/Spot creation stores.

  • Add the initial interactive /v/city SVG guide map and authorized map API.

  • Add SpotPage/MagicPage edges, Viewer locality badges, Spot detail Page lists, and initial typed SpotGraph relation visualization.

  • Add Quartier/Street/Spot detail projections, transactional Street/Spot moves, ordering, and reversible SpotPage/SpotRelation edge removal.

  • Add planning preview/apply, reversible before/after event payloads, and city health analysis.

  • Add /v/city map and locality breadcrumbs.

  • Add locality-scoped search and topology macros.

Phase 3 — Projection runtime

  • Formalize the initial Viewer projection adapter registry and contract.
  • Route Viewer and semantic editor selection through adapters.
  • Route preview through adapters, including bounded text and CSV projections.
  • Route bounded print layouts through adapters at /_/view/print/<address>.
  • Route the semantic Viewer through the initial adapter registry.
  • Add bounded text retrieval backed by seekable Document Store reads.
  • Defer PDF page rendering to idempotent scheduler jobs with retrying Viewer delivery and bounded PDF page-count commands.
  • Defer non-ZIP archive metadata and entry extraction to cached, bounded scheduler projections.
    • Persist PDF page-count and non-ZIP metadata via shared scheduler jobs, bounded immutable-hash cache and localized pending/retry views.
    • Persist seekable non-ZIP extracted entries with quota/lease/Range and archive-bomb controls; move ZIP central-directory work off the request path.
  • Add bounded structural JSON retrieval by immutable Revision and lazy JSON-Pointer expansion without unmarshalling the complete payload.
  • Separate authoring and projection macro transpilation so HTML-to-Markdown round trips preserve children while Markdown previews render it.
  • Generalize the authoring/projection macro contract to typed extension macros with versioned round-trip and projection tests.
  • Add ZIP/archive directory and Debian metadata adapters.
  • Add a bounded CSV table adapter.
  • Add the bounded @excalidraw/excalidraw hosted Viewer plugin with an isolated immutable-Revision contract and safe built-in Canvas fallback.
  • Separate built-in projections from Domain-scoped hosted adapter plugins.
  • Restrict hosted adapter targets to HTTP(S) services such as vim.wasm or a separately deployed editor gateway.
  • Define and implement one-use launch tickets and optimistic Revision write-back with a reference hosted-editor gateway.
  • Connect an internal code-server instance to the hosted-editor Gateway, with session workspace materialization and autosave Revision write-back.
  • Connect a cross-origin-isolated vim.wasm frontend to the hosted-editor Gateway contract with hierarchical MEMFS loading and Revision write-back.
  • Evaluate a local helper, temporary workspace, and executable allowlist before enabling native applications such as Vim or VS Code.
  • Freeze embedded mutable Asset HEADs into immutable Document hashes on Revision save and retain authorization references from the containing Page.
  • Scope mutable Asset addresses and management metadata by Domain.
  • Add typed Asset MIME/filename/size metadata and immutable HEAD history.
  • Add administrator HTTP create/update/delete with ETag preconditions.
  • Add the initial Asset management panel under Domain configuration.
  • Add Asset HEAD history/restore and Page/Revision dependency-detail UI.
  • Add Asset retention controls, policy-matrix capabilities, and general audit-outbox integration as specified in asset-model.md.
  • Add a capability-checked mutable Asset authoring-preview endpoint and typed srcset/CSS resource resolution without allowing live Asset URLs into stored Viewer output.
  • Add authorized immutable object retrieval at /_/obj/sha256:<hash>, limited to Documents referenced by a readable Path in the current Domain.

Navigation and home layouts

See Home layouts and sidebar design.

  • Vendor Hack Nerd Font for Browser and Tree without a runtime CDN dependency.
  • Add Domain-scoped classic, centered-search, and search/bookmark home layouts.
  • Keep home search global to the current Domain even when rendering /top.
  • Scope /v/tree/{path} to the requested subtree with breadcrumbs and cached expansion.
  • Limit Browser revision lookups to displayed files and repair missing revision indexes.
  • Define a shared sidebar layout, state ownership, and bounded children API contract.
  • Implement the shared sidebar and authorized, paginated lazy navigation API.
  • Group configuration navigation and split scheduler, queue, logs, resources, authentication providers, and adapter-specific settings into separate panels.
  • Add accessible themed switches and a Domain bookmark manager for Paths/URLs.
  • Restore public application assets and align compact/legacy editor paths.
  • Manage bookmarks from home; section cards and permission-filtered async child links.
  • Refine bookmark management into an icon-launched modal and section-card editor with group operations.
  • Add admin-only Domain vimrc/colorscheme upload, download/reset, session-scoped delivery and safe-mode recovery.
  • Explicitly load hosted Vim configuration, bundle its font and add a pure-Vimscript navigation plugin; verify the real Wasm runtime. See home navigation and hosted Vim.
  • Add bounded, visibility-aware live resource graphs without a CDN dependency.
  • Repair Asset GET authorization and align metadata/history/reference reporting.
  • Add versioned adapter-specific settings schemas and secret-field handling; see Configuration panels.
  • Add a transactional directory index and batch authorization for large hierarchies.

Phase 4 — Identity and authorization

  • Normalize password, TOTP, OIDC, and magic-link authentication into Principals.
  • Add the initial direct Membership, Role, and positive-grant schema.
  • Add the initial Domain-level Capability evaluation with default deny.
  • Replace capability inference from URL words such as delete and rename with a centralized typed endpoint policy classifier; deny undefined mutations in reserved namespaces and method-close exact API routes.
  • Complete typed object/path target resolvers for derived views and remove the remaining handler-specific authorization exceptions.
    • Resolve raw, windowed text, PDF-page, preview-cache, print, Object, Revision, history, and diff reads through their source Object/Path grants.
    • Filter Tree, Browser, Search, Preview listings, and asynchronous children results per visible Path without returning hidden child counts.
    • Scope the initial Object list and Revision list API responses to the current Domain and readable Object/Path grants.
    • Declare JSON/multipart target resolution as handler-owned policy and remove route-name bypasses from middleware; enforce every expanded source and destination Path for recursive rename, copy, and delete operations.
    • Bind hosted adapter sessions to the issuing Principal and re-check current read/write grants on every workspace/content operation.
    • Move multipart/JSON source-and-destination mutation checks into typed decoded-request objects shared by middleware and handlers.
  • Add cookie-session CSRF and credential-type separation (legacy unprefixed sessions expire normally); POST-only logout and live membership checks.
  • Verify OIDC issuer JWKS signatures/claims/browser state and reject ambiguous email matches or conflicting external-identity binding.
  • Add Domain-scoped LDAP/LDAPS authentication with service search bind, escaped exact-one user lookup, user-DN password bind, encrypted secrets, immutable provider identity, bounded time/concurrency, optional member-only provisioning, saved-configuration test UI, and Japanese/English labels.
  • Revoke derived Campus-admin grants transactionally without deleting independent local grants or regenerating removed bootstrap authority.
  • Scope foreign/unbound Object authority and record atomic API creation provenance; check all local aliases for global Object purge.
  • Revalidate zombie eligibility within the purge write transaction; scope HTTP/config cleanup by Domain provenance and reject orphan-producing writes.
  • Bound Object/Revision metadata APIs with cursor pagination and batched read checks; compact search exposes full Path and safe matching excerpts.
  • Common inert HTML projection boundary, bounded macro expansion, application CSP and cross-account browser denial tests; preserve canonical authoring bytes. See policy and compatibility. No executable mode.
  • Gate the first anonymous semantic/raw slice on explicit current-revision manifests, inert HTML, mutation invalidation and denial/revocation tests.
  • Extend those gates to anonymous discovery/dependencies and delegated shares with full impact-aware mutation/alias handling.
  • Add Domain groups, group Role bindings, and /g/ management surfaces.
  • Add explicit OIDC/LDAP group mappings and policy-version invalidation; LDAP authentication must remain usable without enabling group synchronization.
  • Add auditable policy decisions.
  • Add an authorized effective-permission inventory for Campus, Domain, Quartier, Street, Spot, Page, share, group, Role, and Principal scopes.
  • Add explicit anonymous-read policy grants without weakening default deny.
  • Add signed PSK federation of selected Principal, group, Role, and grant records between Domains, including origin, sequence, conflict, revocation, and campus_admin inheritance policy.

Cross-phase enabling work — Extension seams

These items must land before the later feature that depends on them. They are listed separately because they replace concrete single-feature assumptions without changing user-visible Page identity.

  • Put filesystem DocumentRoot access behind the seekable, content-addressed Document Store interface before adding S3 or multi-node payload placement.
  • Generalize topology/event-job records into a versioned transactional Domain event outbox before notifications, feeds, metrics, comments, or DR consume mutation events.
  • Add a typed extension manifest and provider registry before adding third- party transpilers, widgets, scanners, storage backends, or themes.
  • Inventory process-local sessions, caches, preview slots, scheduler state, adapter workspaces, and signing material before declaring LB/HA support.
  • Add typed annotation/reaction/assessment stores keyed by Page and Revision UUID before exposing comments, likes, or quality evaluation in Viewer HTML.
  • Add privacy, retention, moderation, audit, export, and deletion contracts before collecting view history, engagement, comments, or notification data.

Phase 5 — Delegation

  • Add Macaroon root-key rotation and verification.
  • Implement attenuation caveats.
  • Add server-rendered OGP/X link cards for already-public Page/home/City views, absolute configured-host URLs and a cached PNG fallback; retain authentication and private metadata suppression. See social cards.
  • Add read-only share projections under /s/.
  • Add Quartier/Spot owner appointments and governance dialogs; subordinate publication policies may only narrow superior ceilings. Keep permission to publish separate from actual publication and preserve authenticated ACLs. See detailed ownership/publication design.
  • Add named user/group read and read/write settings with ancestor caps, stable identities, next-request membership checks and recovery authority. Initial namespace guards freeze protected moves/aliases and City structure; replace them with impact-aware mutation workflows in the next slice.
  • Add bounded explicit current-revision approvals and a separate inert anonymous semantic Viewer/raw-media path with safe OGP and revocation tests. No automatic activation of drafts or future edits. See exact scope and follow-ups.
  • Public Page breadcrumbs, asynchronous children, read-only tree and filtered City navigation; database-generation cache invalidation and fresh authenticated read filters. Remove public download controls; reserve isolated presentation for share URLs. See navigation contract.
  • Add owner-managed /u/{username} audiences for profile metadata, avatar, Wiki root and descendants, plus narrowing subtree overrides, a dedicated anonymous reader/navigation snapshot, immediate superior-ceiling revocation, alias/UUID/Revision fences and atomic rename tracking. Keep /u/ out of the global anonymous tree. See user-space publication.
  • Harden anonymous navigation: public root/virtual-folder landing, read-only directory, Page/children round trips, tree expand/collapse/filter/keyboard controls, mobile header and City exits, cancellable bounded children loading, retries and progressive pagination; retain exact-publication boundaries.
  • Explicit Street/Spot automatic publication with inherited manual-approval boundaries, confirmed impact previews, live publisher/ACL checks and cache invalidation. Existing scopes remain opt-in; see implemented slice and limits.
  • Immutable review requests, scoped approvers, durable outbox delivery and reviewer notification Inbox. Continue the remaining AP-1 through AP-6 work in approval workflow design; automatic publication does not implement the human approval workflow or silently activate drafts.
  • Complete dependency/alias-aware publication previews, large-site review and per-capability audiences; current approvals exclude aliases, restricted user/group scopes and dependencies rather than implicitly publishing them.
  • Fence locality publication against alias/move/merge/archive/import bypass, and share one anonymous read evaluator across safe Viewer/media, City/search, children and OGP. Gate enablement on HTML isolation and revocation tests.
  • Add share management, expiry, revocation, access audit, optional password, download prohibition, language/path restrictions, and QR-friendly links.
  • Ensure share credentials are separate from login, adapter, federation, and API bearer credentials.

Phase 6 — Activity, notification, and collaboration

  • Editor resilience E1–E5 initial delivery: local drafts, conditional language-head saves, conservative merge/conflict review, revision restore, language forks, mobile input, macro completion, and protected HTML authoring blocks. See Japanese implementation contract.

  • Editor E6: authorized cross-instance editing presence and feature-gated CRDT collaboration. Follow-up: block-level HTML protection, hosted-editor/CLI migration, and extensible insertion palette.

  • Formalize a Domain-sequenced event outbox for Page, Revision, Path, topology, policy, share, comment, reaction, quality, and federation events.

  • Add Page/Spot/Street/Quartier subscription and change-notification rules.

  • Add a per-Principal notification inbox with unread state, grouping, filtering, retention, cursor pagination, and authorization-safe deep links.

  • Add delivery channels for Web inbox, email digest, webhook, and future push adapters, with retry/dead-letter state in the event scheduler.

  • Add recently edited and recently viewed Page projections; keep private activity private and provide retention/opt-out controls.

  • Add privacy-aware Page view counters with bot filtering, idempotent event aggregation, time windows, and “most viewed” projections.

  • Add Page reactions/likes without modifying immutable Revision payloads.

  • Add document-quality assessments with rubric/version, reviewer identity, score history, and machine/human provenance.

  • Add Revision-anchored inline comments with text-range anchors, context hashes, orphan/rebase handling, resolution state, and permissions.

  • Add Page-level threaded comments below the document with moderation, mentions, subscriptions, edit history, and soft deletion.

Phase 7 — Knowledge City services

  • Complete Campus landing, Domain catalog, controller status, announcements, shared navigation defaults, audit, export, and delegated Campus governance.
  • Complete Quartier charter, entrances, localized signage, curators, announcements, scoped activity/search, subscriptions, health, merge/split, export/import, and policy-default surfaces.
  • Complete Street route intent, sections, Inbox, landing, activity, maintenance queue, subscriptions, merge/split/promotion, and policy override.
  • Complete Spot generated portal, curator workflow, dynamic sections, watchers/digests, backlinks, shortest routes, health states, split/merge, archive/successor, and machine-readable context payload.
  • Add Campus/Quartier/Street/Spot notification and permission inventory integration without turning locality into Page identity.

Phase 8 — Controller disaster recovery and high availability

First slice: Japanese DR design and operator runbook. This is instance-scoped offline recovery, not automatic controller HA.

DR-D1 follow-up is design complete, implementation pending (2026-09-27): authority/rejoin, protocol/synchronization, delivery plan and acceptance matrix. No DB/schema/code changes were made for this design-only follow-up.

  • Specify separate controller/data authority, shared-writer coupling, authenticated rejoin/probe, Learning state, checkpoint verification, fenced handover/preempt, API contracts, rollout gates and acceptance tests.

  • DR01–DR03: implement validated models, pure state machine and observation-only authenticated probe; discovery must not grant write access.

  • DR04–DR05: connect restart write gates and authority providers only after the separately managed storage interfaces and all writer paths are agreed.

  • DR06–DR10: connect consistent checkpoints, bounded resumable transfer, epoch-aware control manifests, verified fencing and manual group transitions.

  • DR11–DR13: certify preempt policy, scoped operator UI and failure drills; automatic lossless failover requires an independently verified durability boundary.

  • Node-local primary/standby configuration; read-only SQLite startup without migrations, key generation, scheduler, GC or managed-campus reconciliation.

  • Offline snapshot/verify/restore CLI: committed WAL via SQLite Backup API, canonical files and keys, payload/reference verification, atomic new generations.

  • Offline operator-attested promotion, old-primary identity/data-loss checks, promotion audit and explicit restored-job resumption. External fencing itself is not implemented or independently verified by these commands.

  • Reject standby writes and stale reads; separate read/write readiness; coordinate server and data CLI with an instance maintenance lock.

  • Rebuild-free read-only organizer/CalDAV paths, standby UI/role indicators, DR settings panel and instance-operator authorization.

  • Online cross-store snapshot barrier, GC pins, resumable bounded delta transfer, encrypted/signed generations, retention and measured restore drills.

  • Restore controller sequence safely across accepted newer manifests; local audit epoch alone does not solve managed-instance replay protection.

  • Add controller manifest journal replication to a designated DR instance, including signing-key rotation/escrow, encrypted backup, sequence verification, last-known-good snapshots, RPO/RTO reporting, and restore drills.

  • Add controller leader election or manual promotion with fencing so two controllers cannot publish the same Campus sequence concurrently.

  • Add managed-instance controller failover using a pinned set of controller identities and an auditable promotion epoch.

  • Separate process-local cache/session/job state from canonical state and define shared/replicated backends required by multiple application nodes.

  • Add health, readiness, graceful drain, idempotency, forwarded-origin validation, and node identity endpoints for load balancers.

  • Add an LB reference deployment with TLS termination, WebSocket support, upload limits, no mandatory sticky session, and failure tests.

  • Add an HA metadata backend or a fenced single-writer/replica design; never mount the writable SQLite database concurrently from multiple nodes.

  • Add Document Store replication/read-repair and background-job leases so only one node performs a claimed task.

  • Exercise node loss, controller loss, network partition, stale replica, key rotation, and rollback in automated fault tests.

Phase 9 — Content composition and extension runtime

  • Evaluate tasks/tickets, an organizer with calendar/contact interoperability, RSS/Atom reading and FreshRSS/newsboat integration, IMAP mail-to-Wiki and incremental HTML editor improvements. Full WYSIWYG remains an open decision, not a commitment. See the design-only organizer and editing ideas; these are not part of the current implementation batch. The detailed design has since been expanded in the Japanese handbook, chapters 12–17; implementation remains unchecked and follows the dependency-ordered plan.

  • Add a versioned transpiler registry that converts PukiWiki, Redmine, DokuWiki, MediaWiki, and future source syntaxes into canonical HTML while preserving the original source as immutable metadata or a related Revision.

  • Complete Markdown round-trip guarantees and record compiler ID/version, warnings, source map, and canonicalization result on every conversion.

  • Add Excalidraw and draw.io canvas projections with immutable asset freezing and configured hosted-editor actions.

  • Add Kroki and PlantUML diagram adapters with bounded input, pinned renderer version, cached immutable output, timeout, and network policy. Existing bounded Kroki execution is now in Kunstzeal; renderer-version keys and complete provider network/resource policy remain pending rather than marking this whole item complete.

  • Add nwdiag as a bounded network-diagram projection and editor adapter; support static SVG fallback, pinned renderer versions, and explicit source import/save without executing authored code.

  • Add a widget/oEmbed registry for Spotify, YouTube, SoundCloud, and future providers with Domain allowlists, consent placeholders, CSP sandboxing, privacy mode, and no arbitrary document-authored script execution.

  • Add administrator-packaged widgets with signed manifests, semantic versioning, capability declarations, resource quotas, and disable/rollback.

  • Add a generated top-page composer using authorized queries, static blocks, recent work, recent views, popular Pages, featured City entrances, and feeds.

  • Add initial diary/blog template/query/navigation projections over ordinary Pages, not a new Page identity or date-dependent storage hierarchy. Registration metadata follows Page UUID; atomic new-post composition, local Domain projection, dedicated public article routes/navigation and RSS 2.0 are implemented. Remaining federation/metadata/Atom work is in the follow-up plan.

  • Add directory-scoped RSS/Atom feed widgets with scheduler fetching, conditional requests, sanitization, per-host limits, caching, and provenance.

  • Add a bounded CSV table macro and CSV editor projection. RFC 4180-compatible CSV remains the interchange baseline; ODS is a separate multi-sheet adapter.

  • Add Reveal-compatible slide projection adapters for HedgeDoc and Slidev, plus a CSS-only scroll-snap/print editor fallback with explicit import/save.

  • Add custom theme packages with inheritance, tokens, preview, validation, CSP-safe assets, versioning, export/import, and one-click rollback.

Phase 10 — External storage, inspection, and ecosystem

  • Define a Document Store backend interface preserving SHA-256 identity, seek/range reads, atomic put-if-absent, integrity verification, and streaming.
  • Add S3-compatible immutable Document storage with multipart upload, server-side encryption, credential isolation, lifecycle policy, replication, local cache, and read-after-write verification.
  • Keep metadata, mutable configuration, and secrets out of the Document Store; external object keys must never replace Document SHA-256 identity.
  • Add asynchronous file inspection states (pending, clean, suspicious, malicious, error, expired) and quarantine-aware authorization.
  • Add MWDB/Karton integration through a bounded inspection adapter and event queue; Viewer/download policy must fail closed where a Domain requires audit.
  • Record scanner engine/version, signatures, hashes, timestamps, verdict provenance, rescan policy, administrator override, and audit events.
  • Add an extension catalog/API for projection, editor, transpiler, widget, notification, inspection, storage, search, and theme providers.
  • Add typed IPAM, DNS zone/record, and inventory/asset extensions with approval-gated provider writes, dry-run, etag checks, secrets, audit, and bounded queue execution.
  • Add extension signature verification, capability review, compatibility ranges, install/upgrade/disable/rollback, health, logs, and resource quotas.

Each phase requires schema migration, rollback analysis, and tests before the next phase begins.