Stars Need
Receipts.

Wing Science: Consistent Review Rules

1. Resolve the branch: verify name, street address, city/state, official website or phone, operating status and Google Place ID / Yelp business ID. A chain name or search link alone is not a match. Conflicting or ambiguous branches stay out of scoring.
2. Capture a dated snapshot: platform, stable business ID, address, rating, review count, source URL, observation time and retrieval method. Preserve earlier snapshots; select only the newest eligible snapshot per platform and branch. Refresh targets are 90 days for active candidates and 180 days for research leads; old observations must retain their dates.
3. Separate restaurant and wing evidence: aggregate stars may describe service, atmosphere or other food. Wing-specific reviews need review IDs, date, sauce, serving context and theme codes. Never infer wing taste from an aggregate star average.
4. Make comparable summaries: each verified platform receives (n × stars + 100 × 3.5) ÷ (n + 100), then ×20. These are declared priors. Average eligible platforms equally; expose disagreement. Do not combine different branches or count multiple snapshots as separate platforms. Review volume affects shrinkage and an optional confidence factor, not automatic taste points.
5. Report contrarian evidence honestly: retain positive and negative examples, deduplicate review IDs and cross-posted text, distinguish dine-in from delivery, and count independent wing-relevant reviews supporting each theme. A recurring theme requires at least three independent supporting reviews; publish numerator, denominator, dates, source links and selection method. This threshold is a reporting convention, not statistical significance. Small or selected snippets establish anecdotes, not prevalence.

Coverage Is Not Collection

This ledger covers every census record. Pending rows have no verified rating/count pair; they are not completed reviews. Existing ATL 10 review evidence remains available in its branch-scoped dataset. It is not automatically assigned to market-level Sam/Jon records. No Google/Yelp averages or review text were invented to fill gaps.

Google exposes place identity, rating and userRatingCount; Yelp provides business details and a limited review-excerpt endpoint. Limited excerpts and relevance-sorted samples are not a random sample of all customers. Use permitted access and preserve attribution and source dates. The local site requires no API keys; opening a search link does not import data.

National Review Ledger