Methodology
Every rooftop on Rooftop Scene clears the same four-stage pipeline before it becomes a live page. The pipeline is built around one principle: the directory is only as trustworthy as its weakest field. Below is exactly what happens between a venue being discovered and a page being indexed.
Candidate venues are pulled from multiple independent sources: Google Places for current operating status, Yelp for category and tagging, and a curated list of editorial seeds for venues that don't index well in API responses. Same-venue results across sources are merged using a deterministic identity algorithm (geo proximity + name similarity), so a single venue isn't accidentally listed twice under variant spellings.
Every candidate's Name, Address, and Phone are reconciled across the source set. A venue must clear a confidence floor — built from address-match score, phone-match score, and source agreement — before it proceeds. The actual presence of a rooftop is confirmed via a separate judgment step that looks at floor information, review language, and venue-supplied photography. Anything ambiguous routes to human review.
Confidence is the weakest link, not an average. A venue with perfect address agreement but a contested phone number is treated as low- confidence on the contested field — we do not smooth that away.
Verified venues are enriched with the practical detail visitors actually need: opening hours (parsed into structured weekly windows), happy hour windows (kept as actual ranges, not booleans), price tier, dress code, pool, accessibility, dog policy, signature drinks. Each field is gated on extraction confidence; required fields that fall below the floor flag the record into review rather than getting a "best guess" answer.
Editorial copy — TL;DR, "The Place" body, FAQ — is generated against the verified record and immediately checked for slop (vague claims, hedge words, repeated phrases). Anything that fails slop detection is rewritten or kicked to human review. The final venue page is published only when all four stages return clean.
Every published venue is re-verified every 30 to 60 days. If a venue's NAP fields no longer agree across sources, it's pulled back into review. If a closure signal appears, it's auto-unpublished pending confirmation. The "Verified" badge on each page shows the last re-verification date — that's the truth date, not the publish date.
Several signals route a venue out of the autonomous path and into the human-review queue: conflicting NAP across sources, missing required fields, closure flags, slop failures, and any case where the photo judge can't confirm a rooftop perspective. These are not failures of the pipeline; they are the pipeline working. Better an honest review queue than a guessed page.
Spot something wrong on a venue page? Email seo@bullseyestrategy.com. We re-verify and correct within 7 days, and the next monthly re-verification will pick it up automatically if missed.