The Toolbench
Every AI tool named in the 353 videos the operator had saved to his TikTok Favorites, as an index: one row per tool, the videos that named it attached as receipts, and a fit score the desk labels as what it is — a cheap model's opinion, useful for sorting and nothing else. Same instrument as the political compass: fixed method, labeled guesses, checkable receipts. The piece that reads this page is The Accessory Channel.
353 videos saved to the operator's TikTok Favorites were fetched with yt-dlp, transcribed with faster-whisper (small.en, CPU), and screened by deepseek-chat through the desk's budget gateway for whether they present a tool; 208 did. Each tool row carries every video that named it as a receipt. Relevance 0-5 is the screener's estimate of fit to this desk's stack; it is an opinion, not a measurement.
Could not be fetched and are listed as absent, not zero: @camnotfunnyy (login required), @stosnai (blocked by address), @delightfuljanice (extractor error), @godemb (login required).
Method: each saved video was fetched, its audio transcribed on the desk's own hardware, and the transcript plus caption screened by the cheapest model on the desk's gateway for whether it presents a tool and which. Names were merged where they were plainly the same product. Nothing on this page was verified beyond the video that said it; the receipts are the video. Dataset CC BY 4.0: toolbench-v0.json.