Workflow on Werk Space

Git Workflow

Thu, 06 Aug 2026 00:00:00 +0000

The repository

The entire site lives in one public GitHub repository, gary-dalton/public-documents. Git provides three things here: versioning for every document, transfer between working machines and the host, and the trigger for deployment. Every push to master causes a fresh build of the live site (see Hosting).

If you need to install git first, see the development environment instructions.

The working cycle

Routine work is a short loop:

Edit content in content/en/.

Preview locally with drafts visible:

hugo server -D --watch --poll 700ms --bind 0.0.0.0 --liveReloadPort=1313

Commit with a message that says what changed and why.
Push to master. The host then builds and deploys automatically.

Publication is controlled entirely by the front matter: a page with draft: true renders in the local preview (because of the -D flag) but is excluded from the production build. Flipping draft: false and pushing is the act of publishing.

AI Multi-Engine Workflow in VS Code

Wed, 28 Jan 2026 00:00:00 +0000

AI Multi-Engine Workflow Introduction

This document summarizes the operational configuration for using GPT(ChatGPT), Gemini, Claude, and GitHub Copilot together in a coordinated research and engineering workflow for modeling and code development using VS Code.

Engine Roles

Engine	Model	Role
GitHub Copilot	GPT-4.1	Inline coding, refactors, reliable Apply
ChatGPT (OpenAI)	GPT-5.x	Deep reasoning, model design, refactor planning
Gemini Pro	Gemini 2.5 Pro	Data pipeline, schema reasoning, ETL logic
Claude	Sonnet / Opus	Architecture comprehension, conceptual critique, writing

Operational Principle

Copilot maintains continuous full-repo awareness and applies edits.
ChatGPT performs high-level reasoning and mathematical/system design.
Gemini engineers data ingestion and validation pipelines.
Claude critiques architecture and synthesizes narrative outputs.

Rule: Only Copilot performs direct code application. Other engines produce plans or diffs for manual application.

Research and Publishing Workflow

Thu, 06 Aug 2026 00:00:00 +0000

Two paths to publication

Most articles are written directly on this site. A page starts as draft: true in the content tree, gets written and revised in place, and publishes when the flag flips. A second path exists for complex or in-depth work: a private research repository holds the piece until it is ready, and it arrives here only at publication. Which path a piece takes depends on one question: does it matter if someone reads the work before it is finished?

Use of AI

Thu, 06 Aug 2026 00:00:00 +0000

Where AI sits in the system

I use AI as a drafting and management assistant inside the workflow. The ideas, the positions, and nearly all of the published language are mine. AI accelerates the parts of the work that are mechanical: gathering research, outlining, producing first drafts, restructuring the site, and building supporting material such as datasets and figures.

What AI does, and what I do

Work	AI does	I do
Research	Collects sources, summarizes dockets and reports, and organizes reference material in the private research repository (see the publishing workflow). For deep research this means multiple agents, sometimes ten at once, splitting search, analysis, and collation.	Direct the agents and triage the results. My reading queue is long, so I review the output and discard mercilessly. What survives gets verified against primary sources: every reference, figure, and dataset row. Nothing publishes with an open `[verify]` flag.
Argument	Assembles evidence for the positions I set.	Choose the thesis and review every argument closely. Drafts often frame positions I do not hold. The draft gets corrected, not the position.
Outlining	Develops a drafting outline interactively with me, often for more complex documents.	Write the document myself, from the outline.
Drafting	Produces first drafts against the voice guide. A draft is raw material.	Rewrite into my voice. I expect to rewrite a third of a draft or more, and for some pieces nearly all of it. My own edits are applied verbatim.
Editing	Works through a document with me, often line by line or paragraph by paragraph. Cleans rough passages I throw out, and makes spelling and punctuation corrections on my rewrites.	Supply the rough material, correct the cleaned version, and settle the final wording.
Reorganization	Restructures a document when I find it needs it: moves material within and between articles, and checks that a line in one article is not contradicted elsewhere.	Decide that a reorganization is needed, and review the moved result.
Site engineering	Restructures sections, builds redirects, edits configuration, and drafts documentation pages like this one, working directly in the repository.	Review every change before commit.
Code documentation	Reads code and documents what it does: template overrides, build scripts, and the procedures around them, in whatever language they are written. The Theme Overrides page is an example.	Ask the questions, and check the answers against the running site.
Supporting artifacts	Builds curated datasets, maps, and figures. These carry draft markings.	Verify every element before the draft marking comes off.
Publication	Nothing.	Decide what publishes, where, and when.

The reader might notice a pattern in that table. Every entry in the AI column is assembly: reading, collecting, drafting, moving. Every entry in mine is judgment against ground truth: the primary sources, my positions, my voice, the running site.