the-nexus/experiments/results_5user_concurrent.md

Experiment: 5-User Concurrent Session Isolation

Date: 2026-04-12 Bridge version: feat/multi-user-bridge (5442d5b) Hardware: macOS, local aiohttp server

Configuration

Parameter	Value
Concurrent users	5
Messages per user	20
Total messages	100
Rooms tested	Tower, Chapel, Library, Garden, Dungeon
Bridge endpoint	http://127.0.0.1:4004

Results

Throughput & Latency

Metric	Value
Throughput	9,570.9 msg/s
Latency p50	0.4 ms
Latency p95	1.1 ms
Latency p99	1.4 ms
Wall time (100 msgs)	0.010s
Errors	0

Session Isolation

Test	Result
Independent response streams	✅ PASS
5 active sessions tracked	✅ PASS
No cross-user history leakage	✅ PASS
Per-session message counts correct	✅ PASS

Room Occupancy

Test	Result
Concurrent look returns consistent occupants	✅ PASS
All 5 users see same 5-member set	✅ PASS

Crisis Detection Under Load

Test	Result
Crisis detected on turn 3	✅ PASS
988 message included in response	✅ PASS
Detection unaffected by concurrent load	✅ PASS

Analysis

The multi-user bridge achieves sub-millisecond latency at ~9,500 msg/s for 5 concurrent users. Session isolation holds perfectly — no user sees another's history or responses. Crisis detection triggers correctly at the configured 3-turn threshold even under concurrent load.

The bridge's aiohttp-based architecture handles concurrent requests efficiently with negligible overhead. Room occupancy tracking is consistent when users are pre-positioned before concurrent queries.

Reproduction

# Start bridge
python nexus/multi_user_bridge.py --port 4004 &

# Run benchmark
python experiments/benchmark_concurrent_users.py

# Kill bridge
pkill -f multi_user_bridge

JSON Results

{
  "users": 5,
  "messages_per_user": 20,
  "total_messages": 100,
  "total_errors": 0,
  "throughput_msg_per_sec": 9570.9,
  "latency_p50_ms": 0.4,
  "latency_p95_ms": 1.1,
  "latency_p99_ms": 1.4,
  "wall_time_sec": 0.01,
  "session_isolation": true,
  "crisis_detection": true
}