Add Phases 20-24 to roadmap: scalability, AI, voice, growth, sustainability

- Phase 20: Scalability & Performance Hardening (v1.9)
  Horizontal scaling, federation perf, voice resilience, large-server tools
- Phase 21: Advanced AI & Intelligence Layer (v2.0)
  Semantic search, proactive assists, moderation AI, voice AI, governance
- Phase 22: Enhanced Voice & Real-Time Collaboration (v2.1)
  Video grid, live streaming, breakout rooms, spatial HRTF, low-latency
- Phase 23: User Growth & Retention (v2.2)
  Onboarding personalization, cross-device continuity, gamification, offline-first
- Phase 24: Long-Term Sustainability & Extensibility (v2.x)
  Protocol versioning, governance tools, security audits, documentation

Updated appendix: What Nexus Does Better (+4 rows), Feature Gap Analysis
(+v2.x Horizons table), Community Most-Wanted Top 10 (March 2026 refresh)

Author: Zajfan

.planning/ROADMAP.md

@@ -886,6 +886,197 @@ Phantom is an infant today. This phase will happen when it is ready, not before.
 
 ---
 
+## Phase 20: Scalability & Performance Hardening (v1.9) 🔲 Planned
+
+> **High priority for large and federated communities.** Everything built so far works for small-to-mid instances. This phase ensures Nexus can handle 100k+ member servers and dense federation meshes without degradation.
+
+### 20-01: Horizontal Scaling
+
+- [ ] ScyllaDB message partitioning improvements (shard by channel + time bucket, automatic compaction strategy tuning)
+- [ ] Redis clustering support (Sentinel and Cluster modes for pub/sub, rate limiting, and session store)
+- [ ] Load-balanced WebSocket gateways (sticky sessions via session-id header, coordinated dispatch across N gateway nodes)
+- [ ] Multi-instance voice SFU (region-aware routing, seamless handoff between SFU nodes)
+
+### 20-02: Federation Performance
+
+- [ ] Batched federation event delivery (coalesce presence/typing/reactions into fewer HTTP round-trips)
+- [ ] Push-based federation subscriptions (persistent WebSocket between peered instances, fallback to polling)
+- [ ] Automatic health-based routing (prefer lowest-latency peer path for multi-hop federation)
+- [ ] Federation event deduplication and idempotency (handle retries without duplicate processing)
+
+### 20-03: Voice Network Resilience
+
+- [ ] Adaptive bitrate with Opus FEC tuning (dynamically adjust codec parameters based on packet loss / jitter)
+- [ ] Multi-region SFU peering (optional self-hosted relay nodes for geographically distributed communities)
+- [ ] Graceful SFU failover (detect unhealthy SFU node, migrate participants to backup with minimal interruption)
+- [ ] Bandwidth estimation and quality indicators (expose per-participant connection quality to client)
+
+### 20-04: Large-Server Tooling
+
+- [ ] Member list pagination and search optimization (cursor-based, indexed by role/join-date/activity)
+- [ ] Role hierarchy visualization (tree view in admin panel, conflict detection)
+- [ ] Auto-prune inactive members (configurable inactivity threshold, grace period DM notification before removal)
+- [ ] Slow mode enhancements (per-role overrides, escalating cooldowns for repeated violations)
+
+### 20-05: Operational Excellence
+
+- [ ] Auto-scaling metrics and alerting (Prometheus rules for connection saturation, message throughput, voice capacity)
+- [ ] Zero-downtime rolling upgrades (connection drain on gateway/SFU, migration-safe DB schema changes)
+- [ ] Distributed tracing (OpenTelemetry spans across API → gateway → SFU → federation)
+- [ ] Capacity planning dashboard (project growth based on current trends, recommend infrastructure changes)
+
+---
+
+## Phase 21: Advanced AI & Intelligence Layer (v2.0) 🔲 Planned
+
+> **Medium-high priority.** Builds on the Phase 16 on-device AI assists foundation. All AI features remain fully opt-in per-server and per-user, with zero cloud dependency by default.
+
+### 21-01: Semantic Search
+
+- [ ] AI-powered query understanding (natural language → structured search, e.g. "last week's memes about cats" → cross-channel media results)
+- [ ] On-device embedding generation (lightweight sentence-transformer model, stored in Tantivy alongside keyword index)
+- [ ] Hybrid search ranking (combine BM25 keyword score + vector similarity for relevance)
+- [ ] Cross-federated search (query forwarded to peered instances with consent, merged results)
+
+### 21-02: Proactive Assists
+
+- [ ] Context-aware reply suggestions (thread/forum context → relevant response options, privacy-first on-device inference)
+- [ ] Auto-tag and summarize long threads/forum posts (generate TL;DR when thread exceeds N messages)
+- [ ] Intelligent notification filtering (learn per-user engagement patterns → surface urgent items, suppress noise)
+- [ ] Meeting notes from voice channels (auto-generate summary from captions after voice session ends)
+
+### 21-03: Moderation AI Evolution
+
+- [ ] Toxicity/sentiment scoring for reported messages (flag for human review with confidence level, never auto-action)
+- [ ] Auto-categorize reports by type (spam, harassment, NSFW, off-topic — suggested label for moderator)
+- [ ] Spam pattern learning per-server (adaptive rules that improve from moderator accept/dismiss decisions)
+- [ ] Raid detection and auto-response (detect coordinated join + spam patterns, auto-enable slow mode, alert admins)
+
+### 21-04: Voice AI
+
+- [ ] Real-time speaker identification / diarisation in recordings and captions (attribute transcript segments to speakers)
+- [ ] Emotion-aware noise suppression (detect voice stress → prioritize speech clarity)
+- [ ] Voice command integration (hands-free mute/deafen/channel-switch via keyword trigger)
+- [ ] Auto-translate voice captions (real-time caption language conversion, on-device where possible)
+
+### 21-05: AI Governance
+
+- [ ] All AI features opt-in at both server and user level (double consent)
+- [ ] Fully on-device or self-hosted LLM endpoints only — no default cloud calls, ever
+- [ ] Model provenance transparency (show which model generated each suggestion, version, license)
+- [ ] AI audit log per server (track which AI features are active, who enabled them, usage statistics)
+
+---
+
+## Phase 22: Enhanced Voice & Real-Time Collaboration (v2.1) 🔲 Planned
+
+> **Medium priority.** Capitalizes on SFU maturity from Phase 3 and drawing/whiteboard infrastructure from Phase 17. Makes Nexus the strongest voice platform in the open-source ecosystem.
+
+### 22-01: Video Grid Polish
+
+- [ ] Dynamic layouts (spotlight/active-speaker, mosaic/gallery, side-by-side, custom drag-and-drop arrangement)
+- [ ] Picture-in-picture mode (floating mini-player when navigating away from voice channel)
+- [ ] Mobile-optimized video grid (swipe between participants, pinch-to-zoom on mobile)
+- [ ] Virtual backgrounds and blur (on-device processing, no cloud dependency)
+
+### 22-02: Live Streaming
+
+- [ ] Go-live from voice channels (broadcast to channel followers, announcement-style distribution)
+- [ ] Stream viewer list with count (opt-in visibility, streamer controls who can watch)
+- [ ] E2EE opt-in recording of live streams (encrypted at rest, decryptable only by participants)
+- [ ] Low-latency HLS/DASH fallback for high viewer counts (SFU → media server fan-out)
+
+### 22-03: Collaborative Sessions
+
+- [ ] Shared screen with pointer/highlight tools (remote cursor visible to all participants)
+- [ ] In-call whiteboard (extend Phase 17 `DrawingCanvas`, synchronized via gateway events)
+- [ ] Breakout rooms (temporary sub-channels within a voice session, moderator-controlled split/merge)
+- [ ] Collaborative document editing in voice (open Canvas doc during call, cursor presence for all participants)
+
+### 22-04: Spatial & Immersive Audio
+
+- [ ] HRTF personalization (head-related transfer function calibration per-user for realistic spatial positioning)
+- [ ] 360° audio positioning in virtual stages/events (place speakers in a virtual room layout)
+- [ ] Ambisonics support for immersive environments (first-order ambisonics encoding/decoding)
+- [ ] Spatial audio presets (conference table, auditorium, campfire — pre-configured positioning templates)
+
+### 22-05: Low-Latency & Quality
+
+- [ ] "Competitive" voice preset (minimal buffering, aggressive jitter control, sub-50ms target)
+- [ ] Echo and feedback auto-calibration (detect and eliminate acoustic feedback loops automatically)
+- [ ] Per-participant audio normalization (consistent volume across different microphones/setups)
+- [ ] Network quality overlay (real-time per-user latency/jitter/packet-loss badges in voice UI)
+
+---
+
+## Phase 23: User Growth & Retention (v2.2) 🔲 Planned
+
+> **Medium priority — post-mobile adoption focus.** These features drive stickiness after users arrive. Reduces churn and makes Nexus feel personal and indispensable.
+
+### 23-01: Onboarding Personalization
+
+- [ ] AI-assisted server recommendations (interest tags + activity patterns → suggested public servers, privacy-respecting)
+- [ ] Quick-join previews with sample messages (read-only peek into active channels before committing)
+- [ ] Adaptive onboarding flow (skip steps the user has already completed, remember preferences across servers)
+- [ ] Friend-of-friend discovery (opt-in: suggest servers where your contacts are active)
+
+### 23-02: Cross-Device Continuity
+
+- [ ] Seamless voice handoff (start call on desktop → transfer to mobile without dropping, via SFU session migration)
+- [ ] Unified read state across all clients (real-time sync, not poll-based — gateway `READ_STATE_SYNC` event)
+- [ ] Clipboard sync across devices (copy on desktop → paste on mobile, E2EE in transit)
+- [ ] "Continue where you left off" (deep-link to exact scroll position and channel on device switch)
+
+### 23-03: Gamification Lite
+
+- [ ] Server-level XP and levels for participation (opt-in, configurable per-server, no central leaderboard)
+- [ ] Custom reaction pack unlocks tied to level milestones
+- [ ] Activity streaks (daily/weekly engagement badges, gentle and non-predatory)
+- [ ] Server achievements (configurable milestones: "First 100 Members", "1000 Messages", "Host an Event")
+
+### 23-04: Offline-First Improvements
+
+- [ ] Full offline compose and search (local SQLite cache, queued messages sent on reconnect)
+- [ ] Conflict resolution for offline edits (last-write-wins with merge notification for simultaneous edits)
+- [ ] Cached federated content (store recent messages from peered servers for offline browsing)
+- [ ] Background sync (delta-based message sync on reconnect, minimize bandwidth after long offline periods)
+
+---
+
+## Phase 24: Long-Term Sustainability & Extensibility (v2.x Ongoing) 🔲 Planned
+
+> **Ongoing maintenance and evolution.** Ensures Nexus stays secure, interoperable, and community-governed as the ecosystem grows.
+
+### 24-01: Protocol Evolution
+
+- [ ] Versioned federation API (backward-compatible negotiation, deprecation path for old protocol versions)
+- [ ] Plugin protocol extensions for custom gateway events (plugins can define and dispatch new event types)
+- [ ] Stable bot API versioning (v1, v2… with migration guides, no breaking changes without major version bump)
+- [ ] Federation capability negotiation (instances advertise supported features, graceful degradation for mismatched versions)
+
+### 24-02: Community Governance Tools
+
+- [ ] Decentralized voting/polls for server policies (binding votes on moderation rules, role changes, etc.)
+- [ ] Contributor badges tied to GitHub/development activity (opt-in, verified via signatures)
+- [ ] Community proposal system (RFC-style discussions for server-wide changes, integrated with forum channels)
+- [ ] Transparent moderation logs (opt-in public audit trail for server actions — community trust building)
+
+### 24-03: Security & Threat Modeling
+
+- [ ] Regular post-quantum readiness audits beyond Phantom (track NIST PQC standards, update as needed)
+- [ ] Metadata minimization in federation (reduce leaked timing/size information in S2S traffic)
+- [ ] Automated dependency vulnerability scanning in CI (cargo-deny + advisory DB refresh, block merges on critical CVEs)
+- [ ] Penetration testing framework (self-hosted security assessment tooling for instance admins)
+
+### 24-04: Documentation & Education
+
+- [ ] Interactive threat model explainer (visual walkthrough of how E2EE, federation, and Phantom protect users)
+- [ ] "Why Nexus?" migration guide (comparison with Discord 2026 changes, Matrix/Element, Revolt, Guilded, TeamSpeak)
+- [ ] Video tutorials for self-hosters (setup walkthroughs, federation peering, plugin development)
+- [ ] Contributor onboarding guide (architecture overview, crate map, "your first PR" tutorial)
+
+---
+
 ## Appendix: Platform Comparison & Competitive Positioning
 
 > Reference document for feature prioritisation. Updated after full audit of Nexus vs. IRC, Discord, Telegram, Slack, Microsoft Teams, Signal, WhatsApp, Zulip, and Guilded.
@@ -945,6 +1136,10 @@ Phantom is an infant today. This phase will happen when it is ready, not before.
 | Privacy as policy promise | Mathematical privacy via Phantom (Phase 11) |
 | Matrix bridge complexity / latency | True federation without bridges or compatibility shims |
 | Docker-only self-hosting (most alternatives) | Single-binary lite mode — download, run, invite friends |
+| TeamSpeak/Mumble basic voice only | Full voice ecosystem: spatial audio, live captions/transcripts, on-device processing |
+| Guilded/Slack lightweight productivity | Built-in productivity suite: tasks, Canvas docs, calendar sync, file versioning |
+| Emerging cloud-dependent AI trends | On-device/edge AI only — summaries, smart replies, auto-mod with zero cloud leakage |
+| Discord/Stoat age-gating & ID verification | No teen restrictions, no default age-gating — mathematical anonymity path via Phantom |
 
 ### Feature Gap Analysis (Priority)
 
@@ -1001,17 +1196,36 @@ Phantom is an infant today. This phase will happen when it is ready, not before.
 | Discord / Slack import tools | ✅ Complete | Phase 19 |
 | Plugin marketplace (federated) | ✅ Complete | Phase 19 |
 
-### Community Most-Wanted Top 10 (2026 Update)
-
-> Updated to reflect what the community is asking for now that Phases 1–14 are complete.
+#### v2.x Horizons (scalability, AI, voice evolution)
 
-1. **Robust mobile apps** — top request everywhere; no features matter if people can't use Nexus on their phone
-2. **Stronger default privacy** — E2EE everywhere + Phantom integration; the unique Nexus edge
-3. **Creator tools & monetization** — tip jars, subscriptions, supporter tiers without a central cut
-4. **Better voice** — spatial audio, mobile parity, voice notes with transcripts
-5. **Collaborative docs & tasks** — Canvas + task management for teams and knowledge communities
-6. **Stories & ephemeral content** — 24h status updates, casual engagement for younger users
-7. **Import / migration ease** — ✅ Discord, Slack, Matrix import tools + bulk invite (Phase 19)
-8. **On-device AI helpers** — ✅ message summaries, auto-moderation, smart replies (Phase 16)
-9. **Advanced search** — cross-federated full-text search, semantic search, media filters
-10. **Voice channel captions & transcripts** — ✅ live captioning + TTS (Phase 18)
+| Feature | Status | Phase |
+|---|---|---|
+| Horizontal scaling (ScyllaDB sharding, Redis cluster, multi-SFU) | Not yet built | Phase 20 |
+| Federation performance (batched events, push-based subscriptions) | Not yet built | Phase 20 |
+| Large-server tooling (auto-prune, role tree, member search) | Not yet built | Phase 20 |
+| Semantic / AI-powered search | Not yet built | Phase 21 |
+| Proactive AI assists (thread summaries, notification filtering) | Not yet built | Phase 21 |
+| Moderation AI (toxicity scoring, raid detection) | Not yet built | Phase 21 |
+| Video grid polish (layouts, PiP, virtual backgrounds) | Not yet built | Phase 22 |
+| Live streaming from voice channels | Not yet built | Phase 22 |
+| Breakout rooms & collaborative sessions | Not yet built | Phase 22 |
+| Cross-device voice handoff & continuity | Not yet built | Phase 23 |
+| Gamification lite (XP, levels, achievements) | Not yet built | Phase 23 |
+| Offline-first compose & search | Not yet built | Phase 23 |
+| Versioned federation protocol | Not yet built | Phase 24 |
+| Community governance tools (voting, proposals) | Not yet built | Phase 24 |
+
+### Community Most-Wanted Top 10 (March 2026 Update)
+
+> Updated to reflect community priorities now that Phases 1–19 are complete. Items 1–2 are clear blockers; 3–10 are post-mobile evolution priorities.
+
+1. **Robust mobile apps** — still the #1 blocker; no features matter if people can't use Nexus on their phone (Phase 10)
+2. **Phantom / mathematical privacy default** — E2EE everywhere + anonymous networking; the unique Nexus edge (Phase 11)
+3. **Advanced AI intelligence** — semantic search, proactive thread summaries, intelligent notifications (Phase 21)
+4. **Scalable large-server & federation performance** — horizontal scaling, batched federation, multi-region SFU (Phase 20)
+5. **Immersive & collaborative voice** — video grid polish, breakout rooms, live streaming, spatial audio HRTF (Phase 22)
+6. **Semantic / cross-federated search** — natural language queries across servers and peered instances (Phase 21)
+7. **Proactive retention & personalization** — AI-assisted onboarding, cross-device continuity, gamification lite (Phase 23)
+8. **Deeper creator monetization** — ✅ tip jar + subscriptions done (Phase 15); next: integrated self-hosted tipping workflows
+9. **Edge / on-device processing** — voice AI, privacy-first inference, zero cloud dependency for all AI features (Phase 21)
+10. **Large-community moderation at scale** — auto-prune, raid detection, AI-assisted report triage (Phases 20 + 21)