I build and write SpaceOdysseyHub alone, and I use AI to do it. A disclosure is only worth reading if it is specific, so this page names every place a language model touches this site, how tightly each one is constrained, and what that does and doesn't guarantee.
I use AI to sweep sources, pull threads together and produce first drafts. I choose every topic. I check the figures against the agency's own page — NASA, ESA, ISRO, JAXA, the filing, the press kit — not against whatever the draft asserted. I review and edit every article before it publishes.
AI-generated statements are never treated as a source. If a claim can't be traced back to a primary document, it doesn't run, and where a figure is genuinely uncertain or disputed the page says so rather than picking the tidier number.
Four features call a language model — Llama 3.1, running on Cloudflare Workers AI. Every one of them is labelled where it appears. They are not equally constrained, and the difference matters, so they're described separately.
On every mission replay at /replays
Ask a question while a replay is running and the guide answers from that mission's record only. Its instructions are explicit: use ONLY the mission record, say so plainly when the record doesn't contain the answer, point to the most relevant chapter instead, never guess, never use outside knowledge, and never invent dates, figures, names or quotes. Quotes may only be reproduced from the transcript it was given.
These are written from structured data this site already holds, with an instruction to stay strictly factual and avoid marketing language. They are not restricted to that data the way the Mission Guide is, so they can draw on the model's own knowledge — the orbit brief explicitly asks for a current trend, which requires it. Treat them as orientation, not as a citable source. Every figure that matters on those pages is rendered from the underlying data beside them, not from the summary.
Imagery.Not one image on this site is AI-generated. Every photograph comes from a space agency archive, a company media kit or a public-domain collection, and every one is credited. If I can't source a real image, the piece runs without one.
Anything presented as a record. Mission timelines, telemetry, transcripts and orbital data come from agency archives and public data feeds. Interactive experiences may simplify a complex subject for clarity — 3D scenes use artistic body scaling by default, with a true-scale toggle — but simplification is disclosed and never substitutes invention for the record.
I won't claim I catch everything. One person working at this volume will miss things, and some pages here have had closer scrutiny than others. That is the real cost of running a site this size alone — so rather than pretend otherwise, I keep a public corrections log, and I would genuinely rather hear from you than not.
Why I work this way. One person cannot cover eleven space agencies at this depth without help. The alternative isn't slower, purer work — it's no coverage at all. What I won't trade away is checking the numbers, so that is where the human hours go.
Found something wrong, or think a label is missing? Tell me — or read the full editorial policy.
Last updated: July 2026