Methodology

Where the facts come from

Every clone's structured facts — stars, contributors, license, engine, last-commit date, platforms — are fetched live from the source it's actually hosted on: the GitHub or GitLab API for repositories, the Steam and itch.io storefront APIs for games distributed there, and SourceForge for the projects still hosted on it. Original-game facts (release year, developer, publisher, genre) come from Wikipedia and Wikidata. Nothing here is typed in by hand or guessed.

Every raw API response and README is saved before anything is written to the database, so a later correction can be checked against the original source instead of a second network fetch — that archive is what makes the verification pass below possible without re-scraping the internet.

How the written content is produced

The narrative text on each clone and original-game page (the description, how it differs from the original, getting-started notes, history) is written by an AI model reading that project's own README, repository metadata, and the original game's Wikipedia article — never from general knowledge about the game. A field is left blank rather than filled with a plausible-sounding guess when the source material doesn't actually support an answer; nothing is invented to complete a section.

Verification

Beyond the AI-writing step, a separate offline pass re-derives each original game's developer, genre, engine, and entity identity from the archived source material and diffs it against what's stored — catching cases like an original linked to the wrong Wikidata entity, or a clone linked to the wrong original game. This runs in bounded rounds, and every flagged discrepancy is reviewed and corrected by hand before anything changes; nothing is auto-corrected from a model's output alone.

What a human decides

A few things are deliberately never automated end-to-end: a gameplay-video pick for an original game is never shown publicly until a person has watched it and approved it, and any AI-content correction found during verification is applied by a person reading the actual flagged case, not by a script.

Corrections

If you spot something wrong — a mislinked original, a stale fact, a broken link — use the Contact page and it'll be checked against the source and fixed.