A one-time job. Now finished.
We want to be plain about this, because "AI slop" is a fair concern to raise and we'd rather lay it out than dance around it.
The compendium was bootstrapped by AI. That phase is complete. Every article you see was AI-drafted, then reviewed and edited by a human before it shipped to the live site. Every change made since shipping has been made by a human community member.
The AI gave us a draft. A human shipped it. The community owns it from here.
FuckCombustion ran for fifteen years before going dark. That's a vast archive of accumulated knowledge — temperature sweet spots, maintenance tricks, stem compatibility, which device revisions have which quirks — written by people who actually used these vapes for years. When the site went offline, that knowledge was effectively lost. No human team was realistically going to read 22,000+ archived threads from scratch and turn them into something searchable.
So we built a one-time pipeline. It ran between April 9 and May 11,
2026 — roughly a month of active scraping, then done. Every page
our pipeline fetched came from the Wayback Machine
(web.archive.org) and the Internet Archive's public CDX index.
Our scraper never made a single HTTP request to a
fuckcombustion.com server. The FC URLs in our data are lookup
keys against the public archive — not destinations. If you've heard claims
that this project had anything to do with FC going down, those claims are
not based on what actually happened: we hit the Internet Archive's mirror,
not FC. If you want to audit the scraping code yourself, email
guardiansoftheround@gmail.com
and we'll share it.
Once we had the archived threads, the pipeline grouped them by device and fed the aggregated posts to Claude Opus with one instruction: summarize what the community actually said, not invent things. The model's output was a draft — never the published article. A human reviewer read every draft, cut hallucinations, fixed the model's misreadings, and edited until the article was accurate. Only then did it ship. That review work is what makes this the seed of a wiki and not an AI content dump.
The choice was: lose the FC community's knowledge entirely, or AI-draft + human-edit one synthesis pass to preserve it. We chose the second. That pass is now complete.
Every article has a Suggest an Edit link. Click it, the body becomes editable text and photos, your submission goes to a moderation queue, and approved edits land on the live site within minutes — committed to the site's public git history under your submitted display name. No AI rewrites. No AI passes "to clean up." Human edits stay as you wrote them.
Compendium articles are for facts: specs, techniques, maintenance, what worked and what didn't in the community's experience. The moderation queue treats them that way. Profanity, off-topic rants, and graffiti-style edits don't make it through. Edits that genuinely improve the factual content do.
If a new device emerges that we don't have an article for, we'll add it the way you'd add any new entry to any wiki: a human writes a stub, the community fills it in, contributors get credited.
This one matters and we want to be plain about it.
The FC archive is publicly browsable on the Wayback Machine, including every poster's username. We could, in theory, list every FC user whose posts went into the synthesis of each article. We chose not to, on purpose.
Those users posted to a niche forum with forum-scale audience expectations, often under pseudonyms they used for years and never offered up for indexing or republication elsewhere. They didn't consent to being credited on a wiki. Aggregating their names across articles — "thanks to user X for sharing on threads Y and Z" — would drag their forum identity into a different context than the one they participated in.
So we treat the FC archive the way you might treat a long Reddit thread or a private mailing list quoted in good faith: we use the public information, we point people back to the original source on Wayback so anyone can read the discussion in context, but we don't lift names out of that context into a roster.
Removal policy. If you contributed to FuckCombustion and would prefer specific content not be used in our synthesis, email guardiansoftheround@gmail.com with the article URL and a description of your concern. We will review every request in good faith and remove or revise the article as appropriate. We do not require proof of authorship to honor a request.
All source material we drew from was publicly-archived forum content. We are not the original publisher of any of that material, and we do not claim copyright over what the community discussed.
People who submit edits and additions through The Round Table know up-front their contribution lands on the wiki. That's a different consent context. Every approved community edit is already committed under the submitter's display name in the site's public git history — so the credit exists today, it's just in the commit log rather than on the article page.
We're working on surfacing that more visibly: a per-article "Contributors" strip near the footer of each compendium page, listing every person who's made an approved edit to that article by their submitted display name and the date of their first contribution. Opt out is always available — submit with the name field blank and you'll appear as "Anonymous contributor."
| Use | Model | Status |
|---|---|---|
| Article drafting | Claude Opus 4.6 (Anthropic) | Phase 1 — complete |
| Thread classification | Claude Haiku 4.5 (Anthropic) | Phase 1 — complete |
| Device-metadata extraction | Claude Haiku 4.5 (Anthropic) | Phase 1 — complete |
The classification and metadata pipelines may run again only if new archived data is ingested — never to rewrite existing articles.
If an article reads as generic, hallucinated, or just wrong, the Suggest an Edit link is right there at the top of every page. Or email guardiansoftheround@gmail.com. The site's git history is public; nothing is a black box.