How PulseScore works

Every recommended order on PulseNYC carries a 0–100 PulseScore. It is a fixed, public formula — no editorial thumb on the scale, no paid placement. Here is the entire model.

The score, component by component

Protein efficiency

0–55 pts

Calories per gram of protein. ≤7 cal/g earns the full 55 points; the score steps down through 10, 14, 18, 25, and 35 cal/g. This is the dominant factor: food that feeds you without overshooting your day wins.

Absolute protein

0–15 pts

≥40g earns 15, ≥30g earns 12, ≥20g earns 6, ≥10g earns 3 — so a tiny-but-efficient item can't outrank a real meal.

Fiber density

0–18 pts

Fiber per 100 calories: ≥2.5% earns 18, stepping down through 1.5%, 0.8%, and 0.4%.

Calorie reasonableness

0–15 pts

≤300 cal earns 15, ≤450 earns 12, ≤600 earns 8, ≤800 earns 4. Above 800 earns nothing.

Sodium penalty

0 to −6 pts

>1,800mg costs 6 points, >1,500mg costs 3, >1,200mg costs 1.

Sugar penalty

0 to −6 pts

>20g costs 6 points, >12g costs 3, >6g costs 1.

The components sum and clamp to 0–100. When you filter by meal (breakfast, lunch, dinner), items matching that meal get a 10-point ranking bonus, and the headline “best order” must be a real meal — at least 200 calories and not a beverage. The best drink, if any, is shown separately so a 5-calorie cold brew can never claim a restaurant's top pick.

The under-$15 anchor. The default venue ranking adds the price wedge on top of PulseScore: orders estimated at $15 or less (or venues in the $/$$ band when the exact order price is unknown) get +8 ranking points, and $$$ venues get −8. Sorting by Protein, Calories, Distance, or Protein per $ bypasses the anchor. Every card shows the estimated order price — an exact “~$11” when we know it, a “~$10–15” band when we don't. Price is never hidden.

Where the data comes from

Venues: NYC DOHMH restaurant inspection data, refreshed hourly. Only venues currently graded A or B appear in ranked picks. Bars, lounges, nightlife, dessert-only shops, and hotel banquet kitchens are excluded from rankings (they stay visible on the map).
Chain nutrition: curated from each chain's published nutrition data for 30+ NYC chains, validated in CI against sanity bounds (a 50-calorie “meal” fails the build).
Local venues: when a restaurant isn't a known chain, picks come from a cuisine template matched to its DOHMH cuisine code and are labeled “est.” — expect ±15% variance from the real menu. If no template fits the cuisine, we show ordering guidance instead of inventing a dish.

Why our data is right

Most food apps get local places wrong because they trust one stale database. We assume every database is wrong somewhere, and cross-check.

Two regulators, both read. NYC restaurants are inspected by the city health department (DOHMH letter grades); bodegas and delis are licensed by NY State Agriculture & Markets and never appear in the city feed. We read both, so the corner deli exists here — and we never invent a letter grade for a venue the city doesn't grade.

A liveness check on every ranked venue. Government permit data outlives real storefronts. Before a venue can rank, we verify it against a live places index: permanently closed venues are excluded and labeled, dead permits with years-old inspections are excluded, and a listing whose real kitchen is a commissary blocks away is excluded rather than pointing you at a door with no restaurant behind it. Opening hours come from the same live source, evaluated on the NYC clock — “open now” is never claimed without data.

Menus verified in person. For our launch neighborhoods we walk in, photograph the menu, and record real dishes and prices. Verified venues carry a dated badge; everything else is honestly labeled an estimate (“est.”, ±15%) — the label tells you which kind of data you're looking at.

A correction loop that actually closes. Every card has a one-tap “this place is closed” report; two independent reports pull a venue from rankings immediately, pending review. When we're wrong, we'd rather be wrong for hours than weeks.

Corrections

Every venue panel has a “Report an error” form. Reports land in a review queue with the venue, field, and your note; data fixes ship in the next deploy.

PulseScore ranks relative nutritional value — it is not medical advice and doesn't know your dietary needs. Nutrition values for non-chain venues are estimates (±15%).