docs: Add L4 Feed architecture documentation

l1-identity/qvl/feed.md

@@ -0,0 +1,99 @@
+# L4 Feed — Temporal Event Store
+
+> Social media primitives for sovereign agents.
+
+## Overview
+
+L4 Feed provides a hybrid storage layer for social content:
+- **DuckDB**: Structured data (posts, reactions, follows)
+- **LanceDB**: Vector embeddings for semantic search
+
+## Architecture
+
+```
+┌─────────────────────────────────────────┐
+│           L4 Feed Layer                 │
+├─────────────────────────────────────────┤
+│  Query Interface (SQL + Vector)         │
+├─────────────────────────────────────────┤
+│  DuckDB        │        LanceDB        │
+│  (time-series) │        (vectors)      │
+│                │                       │
+│  events table  │  embeddings table     │
+│  - id          │  - event_id           │
+│  - type        │  - embedding (384d)   │
+│  - author      │  - indexed (ANN)      │
+│  - timestamp   │                       │
+│  - content     │                       │
+└─────────────────────────────────────────┘
+```
+
+## Event Types
+
+```zig
+pub const EventType = enum {
+    post,       // Content creation
+    reaction,   // Like, boost, etc.
+    follow,     // Social graph edge
+    mention,    // @username reference
+    hashtag,    // #topic categorization
+};
+```
+
+## Usage
+
+### Store Event
+
+```zig
+const feed = try FeedStore.init(allocator, "/path/to/db");
+
+try feed.store(.{
+    .id = snowflake(),
+    .event_type = .post,
+    .author = my_did,
+    .timestamp = now(),
+    .content_hash = hash(content),
+    .embedding = try embed(content), // 384-dim vector
+    .tags = &.{"libertaria", "zig"},
+    .mentions = &.{},
+});
+```
+
+### Query Feed
+
+```zig
+// Temporal query
+const posts = try feed.query(.{
+    .author = alice_did,
+    .event_type = .post,
+    .since = now() - 86400, // Last 24h
+    .limit = 50,
+});
+
+// Semantic search
+const similar = try feed.searchSimilar(
+    query_embedding,
+    10 // Top-10 similar
+);
+```
+
+## Kenya Compliance
+
+- **Binary**: ~95KB added to L1
+- **Memory**: Streaming queries, no full table loads
+- **Storage**: Single DuckDB file (~50MB for 1M events)
+- **Offline**: Full functionality without cloud
+
+## Roadmap
+
+- [ ] DuckDB schema and connection
+- [ ] LanceDB vector index
+- [ ] Event encoding/decoding
+- [ ] Query optimizer
+- [ ] Replication protocol
+
+---
+
+*Posts are ephemeral. The protocol is eternal.*
+
+⚡️