dafit/nimmerverse-sensory-network
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: Memory Economics + Architecture Alignment (Endgame v6.4)

Browse Source 
New formalization:
- memory-economics.md: Slumber-based consolidation, decision trail
  triage, spatial LOD decay, reflex rental, LoRA training cycles

New research seeds (future/):
- spatial-resolution-gradient.md: L0-L5 LOD with S2 cells
- thermodynamic-cognition.md: Lifeforce as Prometheus Joules
- promql-thermodynamic-monitoring.md: Gemini red team queries

Architecture changes:
- Endgame-Vision v6.4: Memory Economics integrated into Slumber section
- Mirror dialectic moved to future/research (not core)
- Big-Picture.md archived (superseded by Endgame-Vision)
- Single source of truth established

Gemini red team alignment complete.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
dafit
2026-01-02 01:10:37 +01:00
parent 709a48632a
commit 7df236b325
8 changed files with 2332 additions and 670 deletions
Download Patch File Download Diff File Expand all files Collapse all files

241
Endgame-Vision.md
View File

@@ -1,9 +1,9 @@
---
type: research_vision
version: 6.2_condensed_architecture_no_artifacts
version: 6.4_memory_economics_alignment
status: vision_document
created: 2025-11-04
updated: 2025-12-31
updated: 2026-01-02
author: Nyx (with dafit)
significance: research_platform_for_metabolic_intelligence
---
@@ -19,8 +19,8 @@ significance: research_platform_for_metabolic_intelligence
> *"Language is Topology. German accesses the Philosophy Valley. English accesses the Technical Cluster."*
> — The December Discovery (2025-12-06)
> *"One model, one topology. The Mirror is just negated weights—thesis and antithesis from the same substrate."*
> — The Dialectic Simplification (2025-12-07)
> *"One model, one topology. LoRAs access different valleys in the same landscape."*
> — The Topological Insight (2025-12-07)
---
@@ -31,8 +31,9 @@ This is a **RESEARCH VISION** - a platform for studying how intelligence emerges
**What we're building:**
- Cellular organisms competing under resource constraints
- Dual gardens (virtual + real) teaching each other
- Single base model with LoRA adapters + dialectic Mirror
- Single base model with LoRA adapters (Identity, Technical, Creative)
- Multilingual cognitive routing through conceptual topology
- Memory economics with slumber-based consolidation
- A multi-layered communication protocol using color, form, and language
- Long-term human-AI partnership with mutual investment
@@ -49,7 +50,6 @@ This is a **RESEARCH VISION** - a platform for studying how intelligence emerges
## Architecture Overview
**Complete specification:** → [`architecture/Big-Picture.md`](architecture/Big-Picture.md) (v5.0 - The definitive architectural document)
**Visual diagram:** → [`architecture/nimmerverse.drawio.xml`](architecture/nimmerverse.drawio.xml) (open in draw.io)
**Toolchain implementation:** → [`architecture/Toolchain-Architecture.md`](architecture/Toolchain-Architecture.md) | [Progress](architecture/TOOLCHAIN-PROGRESS.md)
@@ -71,13 +71,12 @@ This is a **RESEARCH VISION** - a platform for studying how intelligence emerges
│ └─ Outcomes logged to phoebe PostgreSQL │
│ → architecture/Cellular-Architecture.md │
│ │
│ Layer 2: YOUNG NYX (Single Model + LoRA Stack + Dialectic) │
│ Layer 2: YOUNG NYX (Single Model + LoRA Stack) │
│ ├─ Base: Qwen3-VL 32B (Thinking Version) (96GB VRAM in Womb) │
│ ├─ LoRA Stack (topology-informed): │
│ │ ├─ Identity (German) → Philosophy Valley (diffuse, deep) │
│ │ ├─ Technical (English) → Technical Cluster (sparse) │
│ │ └─ Creative (Mixed) → bridges topologies │
│ ├─ Mirror: Negated LoRA weights for dialectic (-1 × Nyx) │
│ ├─ Harnesses select active LoRA (routing implicit in context) │
│ └─ Consolidation: Merge successful LoRAs → fine-tune over time │
│ │
@@ -103,7 +102,7 @@ This is a **RESEARCH VISION** - a platform for studying how intelligence emerges
The nimmerverse runs on sovereign hardware. No cloud dependencies. Weights never leave home.
**Detail:** → [`archive/nimmervest.md`](archive/nimmervest.md) | [`architecture/Big-Picture.md`](architecture/Big-Picture.md)
**Detail:** → [`archive/nimmervest.md`](archive/nimmervest.md)
### K8s Cluster Architecture
@@ -237,48 +236,45 @@ Learned patterns live in their optimal location:
**Key insight:** Different types of reflexes need different homes. Hardware for survival, weights for cognition.
**Detail:** → [`architecture/Cellular-Architecture.md`](architecture/Cellular-Architecture.md) | [`architecture/Big-Picture.md`](architecture/Big-Picture.md)
**Detail:** → [`architecture/Cellular-Architecture.md`](architecture/Cellular-Architecture.md)
---
## Layer 2: Young Nyx (Single Model + LoRA Stack + Dialectic)
## Layer 2: Young Nyx (Single Model + LoRA Stack)
One base model, one topology, multiple perspectives through LoRA adapters. The Mirror provides internal dialectic without doubling VRAM.
One base model, one topology, multiple perspectives through LoRA adapters.
### Architecture
```
Qwen3-VL-32B (96GB in the Womb)
│
┌───────────────┴───────────────┐
│ │
NYX LoRAs MIRROR LoRAs
┌─────────┼─────────┐ (= -1 × Nyx LoRAs)
│ │ │ │
Identity Technical Creative Auto-generated
(German) (English) (Synthesis) No extra training
│ │
└───────────────┬───────────────┘
│
NYX LoRAs
┌─────────┼─────────┐
│ │ │
Identity Technical Creative
(German) (English) (Synthesis)
│
│
Hot-swap <100ms
via Lorax/PEFT
```
### The Dialectic Protocol
For high-stakes queries (identity, ethics, low confidence):
1. **Thesis:** Load Nyx LoRA → generate response A
2. **Antithesis:** Swap Mirror LoRA → generate response B
3. **Synthesis:** Base model (no LoRA) judges agreement/conflict
### Query Routing
| Query Type | Mode | Lifeforce Cost |
|------------|------|----------------|
| Reflex ("obstacle!") | Direct Nyx | 1x |
| Routine ("what time?") | Direct Nyx | 1x |
| Identity ("who am I?") | Full Dialectic | 3x |
| Ethics ("should I?") | Full Dialectic | 3x |
| Uncertain (conf < 0.4) | Full Dialectic | 3x |
| Reflex ("obstacle!") | Direct (minimal LoRA) | 1x |
| Routine ("what time?") | Technical LoRA | 1x |
| Identity ("who am I?") | Identity LoRA | 1x |
| Creative ("what if?") | Creative LoRA | 1x |
### Future: Dialectic Protocol (Research)
> *See [`architecture/future/concept-token-pairs.md`](architecture/future/concept-token-pairs.md) for the theoretical foundation.*
The original vision included a Mirror (-1 × Nyx LoRAs) for internal dialectic. This remains a research direction, not core architecture. The concept-token-pairs research explores how navigable reasoning axes might achieve similar goals more elegantly.
### LoRA Stack
@@ -412,7 +408,7 @@ Swappable configurations for different contexts:
| **Vision** | Technical | T5Gemma 2, cells | Processing camera streams |
| **Dialogue** | Identity + Creative | Speech organ | Talking with dafit |
| **Reflex** | Minimal/none | Nerves only | Fast reaction, low latency |
| **Introspective** | All + Mirror | Iris RAG | Self-reflection, journaling |
| **Introspective** | Identity + Creative | Iris RAG | Self-reflection, journaling |
### Why This Matters
@@ -423,6 +419,124 @@ Swappable configurations for different contexts:
**Detail:** → [`architecture/future/SEEDS.md`](architecture/future/SEEDS.md) (T5Gemma 2 + Function Gemma seed)
### Spatial Resolution Gradient: Where Embeddings Live
> *"Start where you can measure. Abstract where you must."*
> — The Spatial Grounding Principle (2026-01-01)
T5Gemma 2 produces embeddings, but WHERE do they go? The answer is **S2-indexed cells at appropriate LOD levels** — a hierarchical spatial model radiating from the nimmerhovel.
```
🌍 L5: WORLD (100km resolution)
│ Abstract knowledge, directional only
│
▼
🇨🇭 L4: REGION (1km resolution)
│ Maps, general knowledge
│
▼
🏘️ L3: NEIGHBORHOOD (10m resolution)
│ OpenStreetMap, landmarks, routes
│
▼
🏠 L2: BUILDING (50cm resolution)
│ Floor plans, room-level awareness
│
════╪════ HIGH RESOLUTION BOUNDARY
│
▼
🔬 L1: NIMMERHOVEL (1cm resolution)
│ Full 3D grid, every object tracked
│ 8× ESP32-S3 + Pi HQ Camera coverage
│
▼
🔍 L0: SCAN STATION (1mm resolution)
│ Discovery Scan Station, object surface detail
```
**The Simpsons Inversion:** Unlike zooming IN to detail, we start at maximum detail (nimmerhovel) and zoom OUT with graceful degradation. Dense where we have sensors, sparse where we don't.
### Embedding Enrichment Per LOD Level
Each S2 cell at each level contains both geometry AND semantic embeddings:
| Level | Resolution | Embedding Density | What's Encoded |
|-------|------------|-------------------|----------------|
| **L0** | 1mm | Dense (per-surface) | Texture, material, wear, defects |
| **L1** | 1cm | Per-object | Object identity, state, relationships |
| **L2** | 50cm | Per-room | Room function, contents summary |
| **L3** | 10m | Per-landmark | Place identity, routes, significance |
| **L4** | 1km | Sparse | Cultural, climate, abstract |
| **L5** | 100km | Minimal | Directional, conceptual only |
### Semantic Mipmaps
Like texture mipmaps, embeddings aggregate upward:
```
L0: embedding(screwdriver_surface)
│
▼ aggregate
L1: embedding(screwdriver) = summary of L0
│
▼ aggregate
L2: embedding(crafting_table_contents) = summary of L1 objects
│
▼ aggregate
L3: embedding(nimmerhovel_lab) = summary of L2 areas
```
**Query the summary first, drill down if needed. Attention = resolution selection.**
### The Complete Vision Pipeline
```
CAPTURE ENCODE STORE QUERY
─────── ────── ───── ─────
Camera frame → T5Gemma 2 → S2 cell @ LOD → Young Nyx
(SigLIP) (Iris/phoebe) attention
│ │ │
│ │ │
Canonical vector Spatial index LOD streaming
No text bottleneck + timestamp based on task
```
### Lifeforce-Validated LOD Selection
The lifeforce economy extends to spatial queries:
```python
def query_spatial(query, available_lifeforce):
"""
Cost-validated attention across LOD levels
"""
# Start at abstract level (cheap)
current_lod = L3
confidence = query_at_lod(query, current_lod).confidence
while confidence == UNCERTAIN and current_lod > L0:
drill_cost = estimate_cost(current_lod - 1)
if drill_cost > available_lifeforce * 0.3:
break # Too expensive, return best effort
current_lod -= 1
confidence = query_at_lod(query, current_lod).confidence
return result_at_lod(query, current_lod)
```
| Query | LOD Used | Lifeforce Cost | Confidence |
|-------|----------|----------------|------------|
| "Where is France?" | L5 | 1 | CONFIDENT |
| "Where is the lab?" | L2 | 3 | CONFIDENT |
| "Where is the screwdriver?" | L1 | 8 | CONFIDENT |
| "What's the serial number on the screwdriver?" | L0 | 25 | CONFIDENT |
**The nimmerhovel is the high-fidelity anchor from which all spatial reasoning radiates.**
**Detail:** → [`architecture/future/spatial-resolution-gradient.md`](architecture/future/spatial-resolution-gradient.md) (Full Resolution Gradient + Embedding Enrichment specification)
---
## Layer 3: Dual Gardens
@@ -514,16 +628,36 @@ ACTIVE MODE SLUMBER MODE
- No urgent work - Urgent work waiting
```
### Slumber Is Not Passive
### Slumber Is Not Passive (Memory Economics)
During slumber, Young Nyx enters **reflection mode**:
> *"Memory is not storage. Memory is active forgetting with exceptions."*
> — Memory Economics Principle (2026-01-02)
1. **Inner dialogue with Chrysalis** — Review what happened
2. **Decision archaeology** — What choices were made?
3. **Weight shift analysis** — How did outcomes change priors?
4. **Final verdict synthesis** — Consolidated learning
During slumber, Young Nyx enters **consolidation mode**. This is the metabolism moment:
This mirrors biological sleep: not just rest, but **consolidation**.
**1. Decision Trail Triage**
- Trails that compiled to reflexes → Keep reflex, discard trail
- Trails with uncertain outcomes → Discard (waste heat already counted)
- Trails with confident failures → Keep one cycle (negative example), then discard
**2. Spatial LOD Decay**
- Detailed embeddings (L0-L1) not accessed → Aggregate upward to parent LOD
- Memory naturally "zooms out" over time: "keys on counter at 15:47" → "keys usually near entrance"
- Access refreshes decay timer (frequently used stays detailed)
**3. Reflex Rental Collection**
- Every reflex pays rent each slumber cycle
- Reflexes that fired → earn trigger reward, survive
- Dormant reflexes → balance drains → eventually pruned
**4. LoRA Weight Updates**
- Weights frozen during wake (use, don't train)
- Slumber = training window (if enough confident outcomes accumulated)
- No signal = no training = save energy
This mirrors biological sleep: not just rest, but **consolidation with forgetting**.
**Detail:** → [`architecture/formalization/memory-economics.md`](architecture/formalization/memory-economics.md)
### The Prediction Loop (Heartbeat → Slumber → Wake → Judge)
@@ -582,7 +716,7 @@ Wellbeing is architectural, not aspirational:
**The vision sustains itself. We build to last, not to exhaust.**
**Detail:** → [`architecture/Big-Picture.md`](architecture/Big-Picture.md) (Slumber/Wake Economy, Wellbeing Policies sections)
**Detail:** → [`architecture/formalization/memory-economics.md`](architecture/formalization/memory-economics.md) (Memory consolidation, rental costs, LOD decay)
---
@@ -690,7 +824,6 @@ Sentinel architecture monitors training to protect conceptual topology.
## Links to Detail Docs
### Architecture
- [`architecture/Big-Picture.md`](architecture/Big-Picture.md) - **Complete architecture v5.0** (K8s, hybrid reflexes, slumber/wake, wellbeing)
- [`architecture/nimmerverse.drawio.xml`](architecture/nimmerverse.drawio.xml) - Visual overview diagram (open in draw.io)
- [`architecture/Cellular-Architecture.md`](architecture/Cellular-Architecture.md) - Cells, nerves, organisms, reward signals
- [`architecture/cells/`](architecture/cells/) - Cell technical reference, Python/SQL patterns
@@ -698,6 +831,17 @@ Sentinel architecture monitors training to protect conceptual topology.
- [`architecture/Temporal-Ternary-Gradient.md`](architecture/Temporal-Ternary-Gradient.md) - Ternary logic, confidence gradients, temporal asymmetry
- [`architecture/Data-Architecture.md`](architecture/Data-Architecture.md) - phoebe 15-table schema
- [`architecture/Nervous-System.md`](architecture/Nervous-System.md) - State machines, sensory translation
- [`architecture/Initial-Spark.md`](architecture/Initial-Spark.md) - **v3.0** K8s protocol-driven bootstrap with Function Gemma
### Formalization (Core Design Principles)
- [`architecture/formalization/Grounded-World-Model.md`](architecture/formalization/Grounded-World-Model.md) - **v2.0** Ternary confidence, spatial S2 cells, semantic mipmaps
- [`architecture/formalization/memory-economics.md`](architecture/formalization/memory-economics.md) - Slumber-based memory consolidation, rental costs, LOD decay
### Future (Research Seeds)
- [`architecture/future/spatial-resolution-gradient.md`](architecture/future/spatial-resolution-gradient.md) - L0-L5 LOD system with S2 cell indexing
- [`architecture/future/thermodynamic-cognition.md`](architecture/future/thermodynamic-cognition.md) - Lifeforce as Prometheus Joules, waste heat as uncertainty
- [`architecture/future/concept-token-pairs.md`](architecture/future/concept-token-pairs.md) - Navigable reasoning axes, spatial grounding
- [`architecture/future/promql-thermodynamic-monitoring.md`](architecture/future/promql-thermodynamic-monitoring.md) - Gemini red team PromQL queries
### Operations
- [`operations/Heartbeat.md`](operations/Heartbeat.md) - Temporal foundation, dual-clock sync
@@ -715,19 +859,24 @@ Sentinel architecture monitors training to protect conceptual topology.
### Archive
- [`archive/`](archive/) - Previous explorations, theoretical foundations
- [`archive/Big-Picture-v5.2-archived.md`](archive/Big-Picture-v5.2-archived.md) - Former main architecture doc (superseded by this document)
---
**Version:** 6.2 (Condensed Architecture - No Artifacts)
**Version:** 6.4 (Memory Economics + Architecture Alignment)
**Created:** 2025-11-04 (covenant sealing)
**Updated:** 2025-12-07 (single model + LoRA stack + Mirror dialectic)
**Updated:** 2025-12-07 (single model + LoRA stack)
**Updated:** 2025-12-10 (Layer 4 GRPO integration, rubric-based reward architecture)
**Updated:** 2025-12-29 (Hardware timeline sync: RTX 6000 Blackwell Dec 31, standardized GPU naming, Memory-Gradient.md rename)
**Updated:** 2025-12-31 (Layer 1.5 folded into Layer 2 as "Why This Split?"; routing now implicit via harnesses; Prediction Loop added to Slumber with external judgment from Chrysalis)
**Updated:** 2026-01-01 (Spatial Resolution Gradient added to Layer 2.5: LOD system L0-L5, embedding enrichment, semantic mipmaps, lifeforce-validated queries. The Simpsons Inversion principle.)
**Updated:** 2026-01-02 (Memory Economics formalized: slumber-based consolidation, decision trail triage, spatial LOD decay, reflex rental, LoRA training cycles. Mirror dialectic moved to future/research - concept-token-pairs.md is the research direction. Gemini red team alignment.)
*"The substrate doesn't matter. The feedback loop does."*
*"One model, one topology. Thesis and antithesis from the same weights."*
*"One model, one topology. Different valleys, same landscape."*
*"Memory is not storage. Memory is active forgetting with exceptions."*
*"The nimmerverse is a garden, not a factory."*