How the work is made
How an idea is tested
In the author’s words, from a draft series introduction:
I’m J. Hugh Holden. For most of my life I’ve noticed things that didn’t sit right: a planet with ice that’s too clean, a carbon number that never drops, a sky that seems to have run out of air. I’d chase each one until I found a mechanism that made sense, and then ask how it could be tested. These papers are the result. Some of them question settled answers. I’m not asking you to agree, only to check.
The same draft describes the shared method this way: every paper starts with an observation that does not fit. It then builds a physical mechanism, often from basic water-and-gas physics, and ends with predictions someone can check. Consensus “guardrails” are treated as questions to test, not answers to defend. The style is forensic: follow the missing mass to where it went.
Ideas are also checked against published data, run through numerical tests, and given an adversarial review that looks for weaknesses before anything is posted.
How humans and AI tools work together here
The standing disclosure, used on this site and meant to appear with the papers, is:
AI tools assist with literature search and editing; all claims and conclusions are the author’s.
The author’s own account, from the same series introduction:
A note on how they’re written: peripheral neuropathy means I can no longer type or write by hand. Every paper here was built by voice, working with AI tools (Gemini, Copilot, ChatGPT and Grok) to draft, structure and check numbers. The ideas and arguments are mine, the tools did the typing, and I’ve tried to strip out their interpretations. Where a claim still lacks a source, I say so.
AI tools are not listed as authors. An AI tool is also used as an adversarial reviewer, to look for weaknesses. The public site does not treat software as a co-author or a named colleague.
Sources and corrections
Where a claim still lacks a source, the author says so. If you find an error, please write using the contact page. If the error is real, a dated correction note should go on the paper’s page, and the archived copy should be updated when a new version is posted.