What's New¶

Recently added and updated content — in reverse chronological order.

February 2026¶

Diagrams & Tooling¶

• SVG Workflow Generation Best Practices — Mermaid-first, artifact-driven approach to producing committed SVG diagrams: diagram-as-code principles, repo conventions, style guide, CI-ready rendering workflow, and a full data platform example.

• Mermaid → SVG Workflow Pipeline (Tutorial) — Step-by-step guide to setting up @mermaid-js/mermaid-cli, rendering .mmd sources to .svg artifacts, and embedding them in MkDocs pages. Includes troubleshooting for font, viewBox, and Puppeteer issues.

• Diagram Style Guide — When to use Mermaid vs SVG, diagram type selection by content, formatting conventions (orientation, subgraphs, node text limits), accessibility requirements, and a reusable Mermaid snippet library (control plane/data plane, pipeline stages, monolith vs microservices, MQTT broker, ML training vs serving).

• ADR 0015: Standardize Diagrams on Mermaid — Mermaid becomes the site standard for architecture, flow, state, and sequence diagrams; SVG is reserved for geospatial illustrations and cases where Mermaid's layout engine is insufficient.

Site Doctrine & Structure¶

• Start Here — Architectural Compass — Decision-tree navigation organized by problem domain: embedded systems, data pipelines, microservices, ML systems, geospatial, and infrastructure. Each path sequences 4–6 essays with context.

• Reading Tracks — Six curated reading sequences: Modern Data Architecture, Distributed Systems & Scale, Embedded Systems Doctrine, Observability & Operations, Infrastructure Economics, and Spatial Systems.

• Decision Frameworks — Aggregated decision frameworks from nine deep dives, each summarized in 5–10 lines: Kubernetes, Microservices, Serverless, Real-Time, Storage, Analytical Systems, Metadata Governance, ML Deployment, GPU Infrastructure, and Spatial Architecture.

• Anti-Patterns Index — Six architectural anti-patterns with diagnostic signals and deep dive links: Premature Microservices, Overusing Kubernetes, Real-Time by Default, Data Swamp Formation, Serverless Cargo Cult, and Distributed Systems for Small Teams.

• Philosophy of the Site — Five principles: Restraint Over Hype, Economics Over Fashion, Determinism Over Abstraction, Governance Over Chaos, Discipline Over Novelty.

• Systems Thinking Glossary — Operationally-oriented definitions for: control plane, data plane, abstraction debt, blast radius, data gravity, eventual consistency, metadata debt, operational entropy, and schema contract.

• ADR 0014: Elevate Site to Systems Doctrine — Structural decision to add reading tracks, decision frameworks, anti-patterns, philosophy, and glossary pages to transform the site into a navigable intellectual framework.

New Deep Dives¶

• Why Most Kubernetes Clusters Shouldn't Exist — Orchestration overhead, etcd fragility, networking complexity, organizational maturity requirements, and the portability illusion. Decision matrix by team size and workload. (Themes: Infrastructure · Architecture · Economics)

• The End of the Data Warehouse? — The warehouse is not dying — it is mutating. Lakehouse convergence, open table formats, DuckDB compute fragmentation, governance implications, and a tiered decision framework. (Themes: Data Architecture · Economics · Ecosystem)

• The Economics of GPU Infrastructure — GPU scarcity, utilization patterns (~21% realistic vs 100% theoretical), CUDA/driver operational complexity, PCIe vs NVLink interconnect economics, training vs inference cost structures, and a buy-vs-rent decision matrix. (Themes: Infrastructure · Economics · ML Systems)

• The Myth of Serverless Simplicity — Serverless relocates complexity rather than eliminating it. Cold starts, IAM explosion, observability fragmentation, vendor lock-in, and the economics of per-invocation billing at scale. (Themes: Infrastructure · Economics · Architecture)

• Why Most ML Systems Fail in Production — Training/serving mismatch, data drift, feature skew, silent degradation, organizational misalignment, ML observability beyond accuracy metrics, and when heuristics are the right choice. (Themes: Data Architecture · Organizational · Economics)

• The Physics of Storage Systems — HDD rotational latency through NVMe through cloud object storage. IO amplification, throughput vs IOPS, range requests, cold tiers, and a storage tier decision framework by workload type. (Themes: Storage · Infrastructure · Economics)

• The Operational Geometry of Spatial Systems — Quadtree vs R-tree vs H3 hex grid indexing, COG vs GeoParquet format interaction, H3 partitioning strategies, routing graph vs raster cost surface trade-offs. (Themes: Spatial · Architecture · Data Formats)

• The Hidden Cost of Metadata Debt — Catalog drift patterns, control plane collapse, governance vs bureaucracy, the economic cost of duplicate pipelines and compliance failure, and a progressive enforcement framework. (Themes: Governance · Data Architecture · Economics)

• ADR 0012–0013: Deep Dive Scale Governance and Curation Model — Formal curation criteria (novelty, substance, durability, and non-tutorial tests), cluster soft maxima, cross-link requirements, and obsolescence handling for the now-29-essay Deep Dives corpus.

ASCII → Mermaid Diagram Conversions¶

Converted ASCII diagrams in four deep dives to rendered Mermaid blocks:

• Prefect vs Airflow — Airflow control plane / execution layer; Prefect API vs worker data plane separation
• ESP32 vs Raspberry Pi — Side-by-side execution model stacks (bare-metal vs Linux kernel layers)
• Why Most Data Pipelines Fail — Monolithic orchestration fan-out; DAG spaghetti coupling; shared warehouse coupling; event-driven chaos
• Observability vs Monitoring — Sampling trade-off (100% / tail-1% / head-0.1%); data pipeline observability dimensions (freshness, completeness, correctness, lineage)

Recursive Cathedral Generator (Kotlin + Processing)¶

• Recursive Cathedral Generator — Grow Gothic cathedral silhouettes from recursive L-system rules. Deterministic seeds, bilateral symmetry, exportable PNG frames.

• Pi-Based Sample Library Server with Live Audition — Raspberry Pi sample server with SQLite indexing, Nginx, FastAPI, WebAudio API, and USB MIDI live audition. Full 10-section deep dive.

Late 2025¶

• Vibe → Agentic LLMs — Moving from prompt experimentation to production-grade agentic LLM architectures.

• Fractal Art Explorer (JavaScript) — Real-time Mandelbrot/Julia explorer with GPU acceleration, custom palettes, and orbit traps.

• OSC + MQTT + Prometheus + SuperCollider — Let your metrics sing. Wire Prometheus exporters into SuperCollider via OSC for live data sonification.

• MCP + FastAPI Full Stack — Model Context Protocol integrated with FastAPI for production LLM toolchains.

• Cross-Domain Identity Federation — Unified identity federation architecture across all system layers and environments.

• Environment Promotion Drift Governance — Controlled environment promotion with drift prevention and release channel management.

• Data Quality SLA Validation — Comprehensive data quality governance with SLAs and multi-layer validation for tabular, geospatial, and ML data.

Mid 2025¶

• IAM & RBAC Governance — Identity and access management across heterogeneous stacks. One of the most-linked pages on the site.

• Release Management & Progressive Delivery — Safe deployment strategies for coordinating changes across applications, databases, data pipelines, and ML systems.

• Holistic Capacity Planning — Workload modeling, scaling economics, and resource prediction frameworks.

• ONNX Browser Inference — Run ML models directly in the browser with ONNX Runtime Web. No server required.

• RKE2 on Raspberry Pi Farm — Full Kubernetes cluster on ARM hardware. Practical, reproducible, and genuinely fun.

• PostGIS Geometry Indexing — Spatial index strategies, operator classes, and query plans for production PostGIS deployments.