diff --git a/docs/architecture/COMPANION_BRAIN_ARCHITECTURE.md b/docs/architecture/COMPANION_BRAIN_ARCHITECTURE.md
index be53af9ba..ba699e349 100644
--- a/docs/architecture/COMPANION_BRAIN_ARCHITECTURE.md
+++ b/docs/architecture/COMPANION_BRAIN_ARCHITECTURE.md
@@ -1804,9 +1804,14 @@ Code:
   - Webapp-Staging-Effect (`server-iteration-staging.ts`) übersetzt
     eingehende `source:'server'` Iterationen in lokale Proposals pro
     PlanStep mit AI-Actor-Attribution; idempotent via proposalId-Marker
-  - **Offen für Future Work**: Server-side Input-Resolver (notes/kontext
-    Projections), Contract-Test zwischen `AI_AVAILABLE_TOOLS` und Webapp-
-    Policy, Materialized View bei >100 Usern
+  - Server-side Input-Resolver (`db/resolvers/`) für plaintext Tabellen
+    (goals); encrypted Tables bleiben privacy-by-design browser-only
+  - Contract-Test via `@mana/shared-ai`'s `AI_PROPOSABLE_TOOL_NAMES` +
+    Runtime-Drift-Guard im Service
+  - `mana_ai.mission_snapshots` — inkrementeller Snapshot, `listDueMissions`
+    ist ein indexed SELECT statt O(N) LWW-Replay
+
+**Die Workbench-Roadmap ist damit funktional abgeschlossen.**
 
 ### 20.5a Symmetrische Registries: Resolver vs. Indexer
 
diff --git a/services/mana-ai/CLAUDE.md b/services/mana-ai/CLAUDE.md
index e59ce3a87..a30dd1f89 100644
--- a/services/mana-ai/CLAUDE.md
+++ b/services/mana-ai/CLAUDE.md
@@ -18,9 +18,12 @@ What works end-to-end:
 - [x] **Write-back**: `db/iteration-writer.ts` appends the server-produced iteration to `Mission.iterations[]` via a `sync_changes` INSERT under an RLS-scoped `withUser` transaction. Row is attributed with actor `{kind:'system', source:'mission-runner'}`.
 - [x] Webapp staging effect (`server-iteration-staging.ts`) picks up the synced iteration and translates each PlanStep into a local Proposal with full AI-actor attribution (missionId + iterationId + rationale). Idempotent via durable `proposalId` markers.
 - [x] **Server-side input resolvers** for plaintext tables — `db/resolvers/` with a pluggable registry + single-record LWW replay (`record-replay.ts`). `goals` resolver ships by default. Encrypted tables (notes, kontext, tasks, events, journal, …) are intentionally **not** resolved server-side; those missions depend on the foreground runner which decrypts client-side. See `resolvers/types.ts` for the privacy rationale.
+- [x] **Materialized mission snapshots** — `mana_ai.mission_snapshots` table with per-tick incremental refresh (`db/snapshot-refresh.ts`). `listDueMissions` is now a single indexed SELECT; the prior O(N changes) LWW replay stays only in `mergeAndFilter` for tests. Idempotent `migrate()` on boot creates the schema.
 
-Intentionally **not yet** implemented (future work):
-- [ ] Materialized mission snapshot — full LWW replay per tick is O(N changes). Fine pre-launch; revisit when user count grows.
+All roadmap items shipped. Future polish (not blockers):
+- Multi-instance deploy with advisory locks on snapshot refresh (today single-process)
+- Read-only `/internal/missions/:userId` endpoint for ops inspection
+- Metrics endpoint for Prometheus (tick latency, plans/hour, parse-failure rate)
 
 ## Port: 3066