-
-
Notifications
You must be signed in to change notification settings - Fork 2
Experiment Report
Back to Home
Test Date: 2026-03-28 · Server Version: 0.1.1 · Status: Historical baseline scenario
A real-world baseline smoke test of the neurodivergent-memory MCP server, designed as a realistic scenario for managing ADHD executive dysfunction. This page documents the original v0.1.1 validation scenario and remains useful as a conceptual example, but the current release line is v0.2.0 with WAL persistence, concurrency safety, structured error handling, loop telemetry, and benchmark baselines.
For current release status, see Release-Notes, Roadmap, and the repository benchmark artifacts.
Problem Statement: Executive dysfunction (task initiation paralysis) as experienced with ADHD:
- Amygdala hyperactivation in response to task ambiguity
- Time blindness (no dopamine gradient) preventing task urgency perception
- Perfectionism creating fear-based goal setting
- Shame spirals from perceived laziness
Approach: Build an interconnected memory graph that supports someone managing executive dysfunction, testing all five districts, all major tools, and graph connectivity.
The test created 8 memories across all 5 districts:
┌─ logical_analysis (Scholar)
│ └─ Executive dysfunction root causes (amygdala + dopamine)
├─ emotional_processing (Mystic)
│ └─ Shame cycle dynamics (avoidance → guilt → identity)
├─ practical_execution (Merchant)
│ └─ Time-box micro-task intervention strategy
├─ vigilant_monitoring (Guard)
│ └─ Risk assessment (4 failure modes identified)
└─ creative_synthesis (Mystic)
└─ ADHD ≈ Systems Design failure (insufficient feedback loops)
Connections established (6 edges):
-
memory_1 ↔ memory_2(Executive dysfunction ↔ Shame cycle) -
memory_2 ↔ memory_4(Shame cycle ↔ Risks) -
memory_1 ↔ memory_3(Root cause ↔ Intervention) -
memory_4 ↔ memory_3(Risks ↔ Intervention) -
memory_5 ↔ memory_3(Systems insight ↔ Intervention) -
memory_4 → memory_7(Risks → Dependencies) [unidirectional]
| District | Count | Archetype |
|---|---|---|
logical_analysis |
4 | Scholar |
emotional_processing |
2 | Mystic |
practical_execution |
4 | Merchant |
vigilant_monitoring |
4 | Guard |
creative_synthesis |
2 | Mystic |
| TOTAL | 16 |
Note: 16 total due to server restart during testing (8 original + 8 from re-run, demonstrating persistence).
1. memory_1 [1.000] — Executive dysfunction (Scholar)
2. memory_5 [0.360] — Systems insight
3. memory_3 [0.317] — Intervention strategy
4. memory_2 [0.272] — Shame cycle
5. memory_6 [0.154] — Current task
1. memory_4 [1.000] — Risk assessment (Guard)
2. memory_2 [0.461] — Shame cycle
3. memory_1 [0.230] — Executive dysfunction
1. memory_5 [1.000] — Systems insight (Mystic)
District filtering, tag filtering, and relevance ranking all working correctly.
Total connections: 6
5 bidirectional (10 directed edges)
1 unidirectional (1 directed edge)
Connected memories: 9
Orphaned memories: 7
BFS from memory_1 (Executive dysfunction) at depth 2:
Hop 1: memory_2 (shame cycle), memory_3 (intervention)
Hop 2: memory_4 (risks), memory_5 (systems insight)
Total: 4 memories reachable
Related to memory_3 (Intervention strategy):
1. [1.000] memory_1 — Direct connection (root cause)
2. [0.849] memory_5 — Direct connection (systems insight)
3. [0.603] memory_4 — Connected via memory_2
4. [0.153] memory_2 — Semantic relevance
5. [0.153] memory_7 — Semantic relevance (dependencies)
Hybrid ranking (hop proximity + BM25) working correctly.
Updated memory_4 with new risk (burnout) and increased intensity (8 → 9). Search index refreshed immediately after update.
All 8 memories survived server restart. Run 2 produced 16 total memories (8 + 8), confirming JSON persistence works correctly.
| Operation | Time | Status |
|---|---|---|
| Store memory | ~50ms | ✅ Fast |
| Search (10 results) | ~30ms | ✅ Fast |
| Create connection | ~20ms | ✅ Very Fast |
| Traverse (2 hops, 16 memories) | ~40ms | ✅ Fast |
| Update memory | ~15ms | ✅ Very Fast |
| Generate stats | ~25ms | ✅ Fast |
Total test time: ~5 seconds · Memory usage: ~2MB · Success rate: 100% (32/32 operations)
BM25 search correctly identifies conceptually related memories, not just keyword matches:
- Shame cycle → semantically related to perfectionism → risk
- Time blindness → related to dopamine gradient → task initiation
- Systems failures (insufficient feedback) → map to ADHD symptoms
The 5-district + archetype system provides natural storytelling:
- Scholar diagnoses the problem
- Mystic acknowledges the pain
- Merchant proposes solutions
- Guard identifies threats
- Mystic connects across domains
This mirrors how neurodivergent cognition actually works: associative, multi-perspective, integrative.
Memories form a knowledge graph where:
- Root causes connect to emotional impacts
- Emotional impacts inform risks
- Risks motivate interventions
- Interventions generate systems insights
- Insights loop back to reframe root causes
| Goal | Result |
|---|---|
| Neurodivergent-friendly (non-linear, associative) | ✅ Pass |
| Multi-dimensional (all 5 districts) | ✅ Pass |
| Canonical tagging (4-namespace system) | ✅ Pass |
| Semantic search (BM25 relevance) | ✅ Pass |
| Graph connectivity (uni/bidirectional) | ✅ Pass |
| Persistence (survives restarts) | ✅ Pass |
| Performance (sub-100ms operations) | ✅ Pass |
| MCP compliance (JSON-RPC 2.0) | ✅ Pass |
- ✅ Bidirectional connections — both directions traversable
- ✅ Unidirectional connections — one-way traversal works
- ✅ Multi-hop paths — 2+ hop traversal returns correct results
- ✅ Orphaned memories — correctly identified in stats
- ✅ Duplicate memories — server handles identical content separately
- ✅ Tag filtering — district + tag combination filtering works
- ✅ Empty search results — handled gracefully
- ✅ Intensity scaling — update with new intensity value works
The neurodivergent-memory MCP server demonstrated conceptual soundness and early real-world utility at the v0.1.1 research-preview stage. Since this experiment was first run, the project has advanced to v0.2.0 with stronger persistence, concurrency protection, structured operator-facing errors, loop telemetry, and published benchmark baselines.
See also: Architecture · Getting-Started · Release-Notes · Roadmap