First published 2026-07-26

New-to-me routines

After every change,

• run make black_out
• run make mypy_out
• run make pytest_web

Perpetual TOOD

• decrease number of open bugs on github
• increase the coverage of tests
• Sync the three pdg.cypher files: in version control, on local, on website.

Priorities

• Lean integration for checking steps.
• agentic workflow for ingest of derivations.

The agentic workflow is a realistic path to accomplishing the definition of done for the Physics Derivation Graph. I did not imagine having a reasonable path to completion 5 years ago. Similarly, I didn't foresee having a way to practically use Lean because I didn't expect to become proficient at proofs. Using LLMs to debug and leverage tools has expanded what techniques I view as accessible for use in the Physics Derivation Graph. (Being constrained to StackOverflow was the bottleneck before having LLMs that could reason.)

Lean and SymPy provide feedback on whether the LLM's automated ingest is producing true statements.

Claude Sonnet has been crucial for creating the HATEOAS-compliant API. In addition to allowing me to avoid the coding workload, I've been able to be less concerned about learning the details of HATEOAS requirements.