Publishing Infrastructure

The reusable scaffold and deploy pipeline behind every book and guide here: npm-published, dogfooded across projects.

book-scaffold-astro

released

Reusable scaffold for searchable, math-rendered long-form books: Astro + MDX content, Paged.js print output, Tufte-style typography, and a Koller–Friedman chapter structure. Published to npm and consumed as a versioned dependency, so consumer books (double_ml_time_series, ssm-foundations, the guides) inherit upgrades by bumping a version. Maintained through a dogfooding loop — each consumer surfaces scaffold gaps that feed back as fixes.

Stack: Astro · MDX · Paged.js · KaTeX · Pagefind · TypeScript · npm

What's next

A reusable Provenance + audit component (Phase 6a) that surfaces the process-as-artifact trail on every consumer chapter.

deploy-workflows

released

A single reusable deploy workflow that consumer repos call with secrets inherited: it runs the build and a Wrangler deploy to Cloudflare Workers Static Assets. Centralizing it means a deploy fix ships once and propagates to every consumer that pins the tag, instead of drifting across per-repo YAML. This site deploys through it.

Stack: GitHub Actions · Cloudflare Workers · Wrangler · Shell

What's next

Tagged releases carry optional apt-package and PR-preview inputs; an OIDC-auth migration is queued for a later phase.

runpod-deploy

in progress

Infrastructure-as-config for GPU work: parameterized RunPod pod definitions and run configs so training and inference experiments launch reproducibly rather than through ad-hoc console clicks. Built to make the compute side of the research loop repeatable.

Stack: Python · RunPod · Docker

What's next

Active; tightening the config schema and templates as more GPU experiments consume it.