# STRATA 2.0: From Intelligence Engine to Research Lab Bench

**Canemah Nature Laboratory**
Technical Note Series

**Document ID:** CNL-TN-2026-043
**Version:** 0.2 (Draft)
**Date:** April 6, 2026
**Author:** Michael P. Hamilton, Ph.D.
**Affiliation:** Canemah Nature Laboratory, Oregon City, Oregon

---

**AI Assistance Disclosure:** This document was developed collaboratively with Claude (Anthropic, claude-opus-4-6) via Cowork. Claude contributed to architectural analysis, capability mapping, workflow design, and document drafting. The author takes full responsibility for the content, accuracy, and conclusions.

---

## Abstract

STRATA 2.0 extends the existing STRATA intelligence engine — currently deployed on Galatea with 13 temporal micro-agents, 7 context builders, and 24 tool-calling endpoints — into a full research lab bench for the Canemah Nature Laboratory. The lab bench orchestrates structured scientific investigations through a seven-phase workflow, from initial seed observation through peer-publishable conclusions, combining local and cloud AI resources, real-time sensor databases, document archives, pattern detection, and computational tools under human-guided collaboration.

This document describes the evolution from STRATA 1.0 (monitoring intelligence) to STRATA 2.0 (investigation lab bench), the relationship between STRATA and the Science with Claude (SWC) publishing platform, the initial research queue, and the architectural direction for development. STRATA 2.0 operates on Data (MacBook Pro M4 Max) as the primary research instrument; SWC operates on Galatea (Mac Mini M4 Pro) as the public-facing publication venue. Together they form a complete investigation-to-publication pipeline.

---

## 1. Introduction

### 1.1 Motivation

The Macroscope project has accumulated substantial infrastructure over four decades of development: 57 sensor tables streaming environmental and biological data, three trained anomaly-detection meshes (SOMA), 13 temporal micro-agents (STRATA), a curated place-based observatory (MNG), and a growing archive of technical documents. What has been missing is a structured environment for conducting investigations that leverage this infrastructure systematically — an environment that treats each research question as a first-class object with tracked phases, transparent AI attribution, and reproducible methods.

STRATA 1.0 proved the concept: modular agents can generate meaningful temporal context from heterogeneous sensor streams. STRATA 2.0 extends this from passive monitoring intelligence into an active research instrument — a lab bench where investigations are designed, executed, and documented.

### 1.2 From Monitoring to Investigation

STRATA 1.0 answers the question: *What is happening right now across the sensor network?* Its 13 micro-agents generate temporal state summaries, its context builders assemble prompts for Claude conversations, and its tool-calling endpoints enable real-time data retrieval during chat.

STRATA 2.0 answers a different question: *What does this pattern mean, and how do we test that hypothesis?* The lab bench adds investigation management, a seven-phase scientific workflow, a measurement log, pattern detection and identification, cost-aware AI routing, and an export pipeline to SWC for publication.

### 1.3 Design Principles

- **Human-guided, AI-assisted.** The investigator defines questions, evaluates results, and makes all scientific judgments. AI handles literature search, data retrieval, pattern detection, computation, and draft generation.
- **Cost-aware.** Tiered AI strategy: Ollama/MLX local models for repetitive agentic processing; Claude API (Opus for synthesis, Sonnet for routine tasks) for heavy lifting. Every API call is logged with cost.
- **Phase-structured.** The seven-phase workflow provides scientific rigor and auditability. Each phase has defined inputs, outputs, and completion criteria.
- **Tool-rich.** Python, PHP, JavaScript, Three.js, Chart.js, YOLO, GD — whatever the investigation demands. Visualization is first-class, not an afterthought.
- **Pattern-first.** STRATA defines the laboratory's pattern detection and identification capacity. SOMA's RBM meshes, transfer entropy for causal discovery, and the micro-agent temporal windows all feed into the lab bench as detection infrastructure.

---

## 2. System Architecture

### 2.1 STRATA 1.0 to 2.0 Evolution

| Capability | STRATA 1.0 (Galatea) | STRATA 2.0 (Data Lab Bench) |
|---|---|---|
| **Core function** | Monitoring intelligence | Investigation lab bench |
| **Agents** | 13 temporal micro-agents | 13 temporal + investigation-phase agents |
| **Context** | 7 context builders for chat | Extended with investigation context, cost context |
| **Tools** | 24 tool-calling endpoints | 24 existing + analysis, visualization, export tools |
| **Pattern detection** | Temporal state summaries | SOMA integration, transfer entropy, anomaly propagation |
| **Workflow** | Conversational (chat) | Seven-phase structured investigations |
| **Data tracking** | Sensor readings | Sensor readings + measurement log + cost log |
| **Output** | Chat responses | Investigations exportable to SWC for publication |
| **Privacy** | 4-domain model (EARTH/LIFE/HOME/SELF) | Inherited; investigations are private by default |
| **AI routing** | Claude API only | Tiered: Ollama/MLX local, Sonnet standard, Opus synthesis |

### 2.2 The STRATA–SWC Relationship

STRATA and SWC are complementary systems with a clear division of responsibility:

|  | STRATA (Lab Bench) | SWC (Publication) |
|---|---|---|
| **Role** | Conduct investigations | Publish investigations |
| **Machine** | Data (MacBook Pro M4 Max) | Galatea (Mac Mini M4 Pro) |
| **Audience** | Investigator (private) | Public readers |
| **AI Access** | Full: Ollama, MLX, Claude API | None (static content) |
| **Database** | strata_db + macroscope + macroscope_nexus | sciencewithclaude_db |
| **Workflow** | Active: run experiments, analyze | Archival: display results |
| **Output** | Measurements, analyses, findings | Formatted investigation narratives |

The pipeline flows from STRATA to SWC: investigations are conducted in the STRATA lab bench, and when findings reach the Conclusions or Reflections phase, they are exported to SWC for public presentation. SWC's four-tab viewer (Story, Technical Specs, Workbench, Publication) maps directly to STRATA investigation data.

### 2.3 Inherited Capabilities from STRATA 1.0

STRATA 2.0 inherits and extends everything from the Galatea production system:

- **13 temporal micro-agents:** Platform-specific state generators for Tempest, Ecowitt, AirLink, Airthings, AmbientWeather, BirdWeather, iNaturalist, Apple Health vitals, Apple Health activity, Apple Health workouts, Apple Health clinical, Withings, and a meta-agent
- **7 context builders:** Modular prompt generators that assemble temporal, spatial, document, and domain context for Claude conversations
- **24 tool-calling endpoints:** Real-time sensor queries, document search, observation analysis, statistical summaries callable by Claude during conversations
- **4-domain privacy model:** EARTH and LIFE (public), HOME and SELF (private) with per-user, per-domain, per-platform granular access control
- **Multi-temporal windows:** 9 temporal windows per sensor (last_hour through rain_year) plus 7 bird-specific windows (including dawn_chorus)
- **Personality system:** 4 AI personas (field_naturalist, technical_analyst, conversational_guide, research_assistant)

### 2.4 New Capabilities in STRATA 2.0

Three major extensions:

1. **Pattern Detection and Identification.** STRATA's micro-agents currently generate temporal state summaries. In 2.0, they gain the ability to detect patterns, anomalies, and correlations across streams — integrating with SOMA's RBM meshes for energy-landscape anomaly detection and adding transfer entropy for causal discovery. STRATA defines the laboratory's pattern detection and identification capacity.

2. **Investigation-Aware Agents.** New agents aligned with the seven-phase workflow: a Priors agent for literature search and context assembly, a Workflow agent for methods documentation, a Testing agent for statistical analysis and visualization, and a Synthesis agent for narrative generation.

3. **Cost-Aware Routing.** An orchestration layer that routes tasks to the appropriate AI tier: Ollama/MLX for classification, embedding, and repetitive operations; Claude Sonnet for routine analysis; Claude Opus with extended thinking for synthesis and complex reasoning. Every call is logged with cost, latency, and model attribution.

### 2.5 Hardware Topology

| Machine | Specs | STRATA Role | Services |
|---|---|---|---|
| **Data** | MacBook Pro M4 Max | Primary lab bench | STRATA 2.0 UI, agents, Ollama, MLX, MySQL, Python |
| **Galatea** | Mac Mini M4 Pro, 1Gb | Production data + publication | Collectors, macroscope DB (authoritative), SWC, MNG, STRATA 1.0 |
| **Sauron** | Intel NUC i9, 2x RTX 3090 | GPU compute | YOLO inference, 3DGS processing, CUDA workloads |

---

## 3. The Seven-Phase Investigation Workflow

Every STRATA investigation follows the same seven-phase structure. Each phase has defined inputs, AI roles, human responsibilities, and completion criteria.

| # | Phase | Purpose | AI Role | Human Role |
|---|---|---|---|---|
| 1 | **Seed** | Capture originating observation or question | Suggest related patterns from sensor data | Define the question from field experience |
| 2 | **Priors** | Literature review, contextual framing | Search literature, summarize relevant work, identify gaps | Evaluate relevance, add domain knowledge |
| 3 | **Proposal** | Testable hypothesis with success criteria | Draft hypothesis, suggest metrics and thresholds | Refine hypothesis, approve success criteria |
| 4 | **Workflow** | Methods, tools, AI role documentation | Generate analysis scripts, configure data pipelines | Review methods, ensure ecological validity |
| 5 | **Testing** | Execute analysis, collect results | Run computations, generate visualizations, report uncertainty | Interpret results, identify artifacts vs signal |
| 6 | **Conclusions** | Synthesize findings, state limitations | Draft conclusions narrative, flag logical gaps | Final scientific judgment, approve findings |
| 7 | **Reflections** | Meta-analysis of the investigation process | Assess collaboration quality, suggest improvements | Evaluate AI contribution, plan follow-up |

Confidence scores (0.0–1.0) are tracked per phase, reflecting the investigator's assessment of completeness and reliability. A measurement log captures every data query, analysis, observation, and computation with timestamp, method, cost, and investigation linkage.

---

## 4. Initial Research Queue

The following investigations define the initial STRATA research agenda, ordered by priority. STR-001 is foundational — building the lab bench that powers all subsequent investigations.

### 4.1 STR-001: STRATA 2.0 Lab Bench Build-Out

**Status:** Active (Phase 2: Priors)
**Hypothesis:** *Extending STRATA's 13 temporal micro-agents with investigation-phase agents, pattern detection integration, and a structured lab bench workflow will enable reproducible AI-assisted scientific research with transparent human-AI attribution and cost tracking.*
**Data Sources:** `macroscope` (57 tables), `macroscope_nexus`, `cnl_archive`, Macroscope-Galatea codebase
**Tools:** Claude Opus, Ollama/Llama3, Python

This investigation is self-referential: building the instrument that will conduct all subsequent investigations. Key deliverables include the strata_db schema, investigation-aware agents, cost-aware routing, pattern detection integration with SOMA, and the lab bench UI.

### 4.2 STR-002: SOMA 2.0 — Domain Mesh Expansion and Temporal Depth

**Status:** Queued (Phase 1: Seed)
**Hypothesis:** *Adding temporal trajectory encoding and relational cross-domain tensioning to the existing RBM mesh architecture will enable detection of anomaly propagation across environmental, acoustic, and biological observation streams.*
**Data Sources:** `macroscope.tempest_observations`, `birdweather_detections`, `ecowitt_readings`
**Tools:** Python/NumPy, MLX, Claude Sonnet

Builds on the three existing SOMA meshes (Tempest 35–100 nodes, BirdWeather 27–50 nodes, Ecosystem 65–100 nodes) by adding temporal depth (trajectory encoding) and a relational layer for cross-domain tension detection. Architectural direction follows Whittaker's gradient analysis: domain meshes maintain independent internal logic. Deployed to Galatea for production inference.

### 4.3 STR-003: Tempest vs. Open-Meteo Microclimate Divergence

**Status:** Queued (Phase 1: Seed)
**Hypothesis:** *Physical weather station measurements at Canemah will show systematic divergence from Open-Meteo grid estimates, revealing microclimate signatures driven by topography, canopy cover, and riparian proximity to the Willamette.*
**Data Sources:** `macroscope.tempest_observations`, `macroscope.openmeteo_hourly`
**Tools:** Python/Pandas, Chart.js, Claude Opus

The most immediately tractable investigation — both data sources are already streaming and the comparison methodology is straightforward. Initial seed observation recorded a 3.2 degrees C divergence during a morning temperature inversion event. This investigation will characterize the systematic biases and their ecological implications.

### 4.4 STR-004: ecoSPLAT 2.0 — Panoramic to 3D Gaussian Terrarium

**Status:** Queued (not started)
**Hypothesis:** *3D Gaussian Splatting can transform panoramic field photography into navigable ecological terrariums suitable for species-in-context visualization and phenological time-series comparison.*
**Data Sources:** ecoSPLAT filesystem (894 GB), Insta360 panoramas
**Tools:** Sauron GPU, Three.js, Python, 3DGS toolkit

Leverages Sauron's dual RTX 3090 GPUs for Gaussian Splatting processing. The goal is to create immersive 3D environments from 360-degree field photography, annotatable with species observations and sensor overlays.

### 4.5 STR-005: Real-time Video Stream Classification

**Status:** Blocked (Linux/CUDA install pending on Sauron)
**Hypothesis:** *Real-time YOLO classification of camera trap video combined with vision-enabled local AI can achieve >85% species identification accuracy for mammals and birds at the Canemah study site.*
**Data Sources:** RTSP camera streams, `macroscope.birdweather_detections` (training reference)
**Tools:** Sauron GPU/CUDA, YOLOv8, Python, Ollama/LLaVA

Blocked on Sauron's Linux/CUDA setup. Once operational, combines YOLO real-time detection with LLaVA vision for species identification from camera trap streams, cross-referenced against BirdWeather acoustic detections for multi-modal species confirmation.

---

## 5. Database Architecture

### 5.1 The strata_db Schema (Planned)

STRATA 2.0 will use a dedicated database for investigation management, separate from but connected to the existing Macroscope databases:

- `investigations` — Core investigation records (id, title, hypothesis, status, priority, dates, investigator, cost tracking)
- `investigation_phases` — Per-investigation phase state (phase number, status, confidence, content, start/end dates)
- `measurements` — Lab notebook: every data query, analysis, computation, observation with timestamp, method, AI model, cost, and investigation linkage
- `data_sources` — Registry of databases, tables, and external sources used by each investigation
- `tools` — Registry of software tools, models, and libraries used by each investigation
- `tags` — Investigation-specific classification
- `cost_log` — Per-call AI cost tracking with model, tokens, latency, and cost
- `strata_agents` — Agent registry: definitions, capabilities, and investigation assignments
- `patterns` — Detected patterns and anomalies with source agent, confidence, and investigation linkage

### 5.2 Cross-Database Architecture

STRATA 2.0 joins three federated databases at the application layer:

- `strata_db` — Investigation management, lab notebook, pattern registry (new, Data-local)
- `macroscope` — Time-series sensor data, 57 tables (synced from Galatea every 5 minutes)
- `macroscope_nexus` — Curated places, categories, media, species, organizations (shared with MNG)

The existing bridge pattern (`macroscope_nexus.monitoring_sources.macroscope_platform_id` linking to `macroscope.sensor_platforms.id`) provides the architectural seam for STRATA to access both spatial identity and temporal data streams.

---

## 6. Tiered AI Cost Strategy

STRATA operates on a principle of cost-awareness: every AI call is logged, and the system routes tasks to the cheapest sufficient model. The three tiers:

| Tier | Provider | Use Cases | Cost |
|---|---|---|---|
| **Local** | Ollama, MLX | Classification, embedding, repetitive agentic loops, structured extraction, draft generation | $0 (electricity only) |
| **Standard** | Claude Sonnet | Routine analysis, literature summaries, code generation, data interpretation | ~$0.01–$0.05/call |
| **Synthesis** | Claude Opus (ext. thinking) | Complex reasoning, narrative synthesis, architectural decisions, cross-domain integration | ~$0.05–$0.50/call |

Environmental cost awareness is also tracked — API calls carry both dollar and estimated compute-energy costs. The STRATA dashboard displays running session costs in real time.

---

## 7. Visualization and Tool Capabilities

STRATA investigations demand diverse visualization. The planned toolkit:

- **Chart.js:** Time-series, scatter, bar, radar charts for sensor data and statistical results
- **Three.js:** 3D Gaussian Splatting viewer, planetary visualization, spatial data rendering
- **Python/Matplotlib:** Statistical plots, heatmaps, correlation matrices, scientific figures
- **D3.js:** Network graphs (SOMA mesh visualization), force-directed layouts, custom interactive visualizations
- **GD/PHP:** Server-side image processing, thumbnail generation, annotation overlays
- **YOLO/LLaVA:** Real-time object detection and vision-language classification for camera trap streams

---

## 8. User Interface Design Direction

The STRATA lab bench dashboard prototype follows the visual language of computer science paper figures — rich saturated panel colors (lavender, peach, mint, sky blue, cream) on a white background with crisp black typography. This aesthetic was chosen for information density, clarity, and visual distinction between functional zones.

Key interface elements:

- **Research Queue:** Ordered investigation list with status badges, phase progress bars, and cost indicators
- **Detail Panel:** Hypothesis display, phase timeline with per-phase color coding, data source and tool registries, cost metrics
- **Lab Notebook:** Real-time measurement feed with type indicators (Analysis, Query, Computation, Literature, Observation, Classification), method attribution, and cost per call
- **System Status Bar:** Machine availability (Data, Galatea, Sauron), AI provider status (Ollama, MLX, Claude API), database sync status
- **STRATA Engine Panel:** Prominent indicator showing agent count, context builder count, and tool count

A working React prototype (`strata-prototype.jsx`) has been developed and validated. The production implementation will use PHP/vanilla JavaScript consistent with the Macroscope LAMP stack.

---

## 9. Next Steps

The following tasks are prioritized for the next development session:

1. **Create strata_db schema** on Data's MySQL. Define tables for investigations, phases, measurements, cost_log, patterns, data_sources, tools, tags, and strata_agents.
2. **Build STRATA 2.0 application scaffold** in Projects/Workbench/STRATA/ with env.php, lib/ directory, admin/ dashboard, and lab bench views.
3. **Implement investigation CRUD** for creating, editing, and managing investigations and their phase progression.
4. **Build the measurement logger** — the lab notebook that captures every data operation with cost tracking.
5. **Begin STRATA 2.0 codebase analysis** (STR-001 Phase 2): review the Galatea STRATA codebase, map agent architectures, and design the extended agent framework.
6. **Deploy SWC to Galatea** (rsync) with maintenance splash page active, and populate the first investigation seed in SWC from STR-003 (Tempest vs. Open-Meteo).
7. **Rename and move prototype** to `strata-prototype.jsx` as the UI reference for LAMP implementation.

---

## 10. References

[1] Hamilton, M. P. (2026). "Macroscope/STRATA and MNG Convergence Plan." CNL-TN-2026-027, Canemah Nature Laboratory.

[2] Hamilton, M. P. (2026). "Organelle Convergence Architecture." CNL-FN-2026-026, Canemah Nature Laboratory.

[3] Hamilton, M. P. (2025). "CNL Technical Note Style Guide." CNL-SG-2025-002 v1.1, Canemah Nature Laboratory.

[4] Whittaker, R. H. (1967). "Gradient analysis of vegetation." *Biological Reviews*, 42(2), 207–264.

[5] Anthropic (2026). "Claude API Documentation." https://docs.anthropic.com (accessed April 6, 2026).

[6] Ollama (2026). "Run Large Language Models Locally." https://ollama.ai (accessed April 6, 2026).

---

## Document History

| Version | Date | Changes |
|---|---|---|
| 0.1 | 2026-04-06 | Initial draft as "IRIS" planning document. |
| 0.2 | 2026-04-06 | Renamed to STRATA 2.0. Reframed from separate system to evolution of existing STRATA engine. Added 1.0-to-2.0 comparison table, pattern detection as core capability, patterns table in schema. Investigation IDs changed from IRIS-nnn to STR-nnn. |