Research a venue across several AI models into venues_new. Existing matches surface first. Fields where the models agree are written; splits, lone claims and drift surface in the consensus grid for review.
Sort
Unapproved venues whose latest run left fields empty β no model emitted them, and no curator cleared them. A re-run with fresh sources fills these; your clicks do not. Open a venue and research it again, or wait for the gap-fill batch.
Sort
Resolves a Google listing for venues that have none, and queues each one for your confirmation. Nothing is written until you confirm.
One web search per source and category (β 40 for a curated city, β 24 for an unknown one), hard-restricted to that source, then one Google Places lookup per candidate for the address, website and whether it still trades. Names that two sources agree on rank first; rows already in venues_new and permanently closed places are marked. Nothing is researched: keep the rows you want and send them to the paste list.
The bar before a city nobody has curated: recall β₯ 80 % on Basel, Dubai and Edinburgh. Keep / drop is remembered on the server; Send kept to paste list hands the kept rows to the Phase 1 batch (max 50 per paste).
Venue name ; City ; Address ; Website β semicolons or tabs (paste straight from a spreadsheet; a header row is skipped). Name + city required; address + website sharpen the match & research. Up to 500 rows; a list longer than 50 becomes batches of 50 β the first starts, the rest wait for you to start them in turn.
0 / 500 rows
π Researching 0 / 0batch β Β· 0 tokens
Runs on the server, one venue at a time (about 70 s each; a full 50 β an hour). You can close this tab: the batch keeps going, and opening this page again in this browser shows where it is. Pause any time; it pauses itself after 3 consecutive failures. Duplicates are skipped unless overridden per row. Results land in the Review queue.
Catalogue venues
Pick the real venue β its website & Google identity seed the research and link google_place_id onto the curated row.
Search results
Open an existing record, or research it as a new venue.
π
Reading the venue's siteβ¦
Researching across the web + several AI models, then reconciling β this can take ~10β30s.