Pattern is Everything

A visual field guide to patterns in machine learning, markets, and ideas.

Live site · Start here · Support

Pattern is Everything is a free educational platform for learning complex concepts through interactive visual references. The core library covers 257 topics across machine learning and markets, each paired with a canvas-based visualization, practical explanation, formulas, and code-oriented context.

The project also includes essays, real-data mini-cases, notebook companions, sandboxes, and a public donation log for mental-health-related giving.

At a Glance

Metric	Count	Notes
Core topics	257	Machine Learning + Markets topic pages, excluding collection overview pages
Core visualizations	257	One interactive visual reference per core topic
Universes	4	Machine Learning, Statistics & Data Analytics, Markets, and Pattern Essays
Collections	10	4 ML collections, 1 Statistics & Analytics collection, 4 Markets collections, 1 Essays collection

Universes

Machine Learning - 118 topics

Collection	Topics	Covers
ML Math	38	Linear algebra, gradients, backprop, attention, transformers, diffusion, LoRA, RLHF
LLM Engineering	30	Tokenization, transformer internals, RLHF, DPO, RAG, inference optimization
MLOps & Production ML	25	Serving, drift, feature stores, pipelines, monitoring, governance
Timeseries Engineering	25	Stationarity, ARIMA, Prophet, LSTM, transformers, anomaly detection, forecasting workflows

Statistics & Data Analytics - 39 topics

Collection	Topics	Covers
The Toolkit	39	Evaluation, feature importance, statistical foundations, data analytics, backtesting, decision support

Markets - 100 topics

Collection	Topics	Covers
Chart Patterns	25	Head and shoulders, double tops, flags, wedges, cup and handle
Technical Indicators	25	Moving averages, RSI, MACD, Bollinger Bands, Fibonacci, volatility tools
Market Psychology	25	Biases, fear and greed, herding, sentiment cycles, behavioral traps
Risk & Portfolio	25	VaR, expected shortfall, position sizing, hedging, performance attribution

Pattern Essays

Long-form visual essays that connect pattern thinking across machine learning, markets, language, systems, and decision-making.

What Makes It Useful

• Visual first - diagrams and animations make abstract ideas easier to inspect.
• Practical by default - formulas, examples, pitfalls, and code-oriented notes sit beside the visuals.
• Real-data case studies - four cases on public data (California housing, credit card default, household electricity, a century of US stock returns), every step runnable in the browser; built by scripts/build-cases.py from scripts/cases/.
• Notebook companions - Jupyter notebooks under notebooks/ for executable case walkthroughs.
• Guided learning path - the Start Here flow helps visitors move from beginner concepts to applied workflows.
• Sandbox areas - focused playgrounds for ML, stats, deep learning, markets, and chaos concepts.
• Fast static site - the public site is plain HTML, CSS, and JavaScript with no app build step.
• Respectful experience - no ads, no paywalls, and privacy-friendly analytics only.

Project Map

Pattern Portal/
├── index.html                 # Homepage
├── css/main.css               # Shared design system
├── js/                        # Shared homepage and learning-path behavior
├── start/                     # Guided beginner path
├── stats/                     # Statistics & Data Analytics universe (The Toolkit)
├── ml-math/                   # ML Math collection
├── llm/                       # LLM Engineering collection
├── mlops/                     # MLOps & Production ML collection
├── timeseries/                # Timeseries Engineering collection
├── markets/                   # Market hub and market collections
│   ├── charts/
│   ├── indicators/
│   ├── psychology/
│   └── risk/
├── essays/                    # Pattern Essays universe
├── cases/                     # Real-data mini-cases and local CSV samples
├── notebooks/                 # Notebook companions
├── sandbox/                   # Interactive sandbox surfaces
├── support/                   # Giving page
├── impact/                    # Public donation log
└── vercel.json                # Headers, cache rules, and master-only deploys

Run Locally

Use any static file server from the repository root.

python -m http.server 4173

Then open:

http://localhost:4173/

You can also use npx serve . or another static server if that is your normal workflow.

Checks and Reviews

node scripts/check.mjs is the drift guard: deep links, counts, pre-rendered pages, runnable code, the case studies re-run against their pages, and more. A GitHub Action runs it on every push and pull request; locally it needs the Python in scripts/requirements.txt (pip install -r scripts/requirements.txt).

Every topic's entry in its collection's TOPIC_DATA carries two fields the check enforces:

• evidence - mathematical, statistical, practice, heuristic or debated, as defined on /about/. It belongs to the topic, not the collection.
• reviewed - the date the topic was last checked against its sources ('2026-10-02'). It appears on the page and as its dateModified.

node scripts/reviews.mjs lists the topics due for another review: six months after the last one in LLM Engineering and MLOps, eighteen elsewhere. A weekly GitHub Action fails when any are due. After reviewing a topic, update its reviewed date and run node scripts/prerender.mjs <collection>.

node scripts/links.mjs checks every external link (sources, references, data sets). A monthly GitHub Action fails on dead links (404, 410, host gone); links that only refuse scripts (403, 429, timeouts) are listed for a look in a browser.

Every topic, essay, guide and case ends with Report a mistake, which opens the issue form in .github/ISSUE_TEMPLATE/correction.yml with the page filled in. Close a report with the commit that fixes it.

Deployment

The site is deployed on Vercel through its GitHub integration. Only master deploys: pushes to any other branch build nothing, so pull requests do not get preview URLs.

git push origin master
node scripts/deploy-status.mjs     # confirm it actually went live

That rule lives in vercel.json under git.deploymentEnabled. It exists because the account is on Vercel's Hobby plan, where every build - production or preview - draws on one shared deployment quota. Every commit pushed to a PR branch was building a preview, and once those exhausted the quota Vercel refused production builds too, while git push kept reporting success.

Two things follow from that:

• A push is not a deploy. Vercel can refuse a build after the push has succeeded, and it does not retry a refused commit by itself - nothing builds until the next push to master. node scripts/deploy-status.mjs compares what is committed, what is on GitHub, the last production deployment and the live site, and exits non-zero when they disagree.
• Batch changes before pushing to master. Each push is one production build. If a build was refused for the rate limit, node scripts/push-when-allowed.mjs waits for the window given in the refusal, pushes, and reports whether Vercel accepted the build.

Giving

Pattern is Everything is free to use. Donations are intended to support mental health organizations. The receiving organization is confirmed before transfer, and received amounts, transfers, and kept amounts are documented in the public donation log.

License

MIT