Skip to content
Sunset Beach, NC
Get Support

How the app judges things

Every judgement Sunset Beach makes is a sum of named, signed contributions that the app prints back verbatim, so you can always see why. Four activities (Swimming, Surfing, Beach walk & shelling, Sunset watching) each start at 50 and add or subtract points for rip risk, water temperature, surf, wind, rain, sun and air quality; some conditions cap a score outright, such as swimming after dark. The headline is weighted toward the best of them. The same scoring runs hour by hour to find the day’s best time, and the summary line at the top of Today is written by Claude Haiku from those computed facts, then checked by code before it is shown. The sunset and sunrise outlook is judged on the layered cloud profile at that hour, not total cloud cover. Rip and swim verdicts come straight from the NWS surf zone forecast’s rip level. Rocket-launch visibility is a geometry-and-lighting model built on the Worker. Almanac windows score this occurrence against a typical one of its kind. All of the weights are, in the code’s own words, opinionated guesses tuned against nothing but judgement about this beach.

Activity scores (“What it’s good for”)

Section titled “Activity scores (“What it’s good for”)”

Each activity starts at 50 and its contributions are added, then clamped to 0–100. Ratings: great ≥ 75, good ≥ 55, fair ≥ 35, otherwise poor. Tap an activity on Today to see every contribution, with a Start 50, total line so you can check the arithmetic.

ActivityWhat adds or subtracts pointsCeiling
SwimmingRip level None +15 / Low +5 / Moderate −25 / High −50 · water ≥ 78°F +20, ≥ 72 +12, ≥ 65 −5, colder −25 · surf midpoint ≤ 2 ft +12, ≤ 4 ft −5, higher −20 · ≥ 50% rain −12 · burn time ≤ 20 min −8 · air quality (see below)10 after dark (“After dark — nobody can see a swimmer in trouble”) — bath-warm water is still true at 2 AM, so darkness caps the score rather than merely deducting. 54 with no rip forecast (the top of fair, “Rip current forecast unavailable”)
Surfingsurf 3–6 ft +25, ≥ 2 ft +8, smaller −25 · swell period ≥ 10 s +20, ≥ 7 s +8, shorter −10 · wind < 5 mph +15, offshore +18, onshore > 12 mph −20, otherwise −5 · steepness printed at 010 after dark (“After dark — you can’t see the waves coming”)
Beach walk & shellingtide ≤ 1 ft +25, ≤ 2.5 +12, ≤ 4 −5, higher −20 · falling tide +8 · wind > 20 mph −20, > 14 −8, else +8 · rain ≥ 60% −25, ≤ 15% +8 · air 65–88°F +10, > 95 or < 45 −15 · feels-like ≥ 8° hotter −8 · air qualitynone — after dark is a −10 (“After dark — fine for a walk, but shelling needs a flashlight”), because a low-tide walk at 11 PM is a real reason to go out
Sunset watching(sunset model score − 50) × 0.6, with the high-cloud and low-cloud percentages printed at 0 · ≥ 50% rain −15 · wind < 12 mph +5 · falls back to total cloud (25–70% +15, else −5) if the layered profile is missing10 once tonight’s sunset has passed (“The sun has already set”), lifting at midnight

A cap is a ceiling, not another negative row: no amount of warm water makes swimming in the dark something to recommend, so the two should not be able to cancel out. The activity screen shows a capped score as Capped at N, with “Without the cap these conditions would score M.”, and puts the cap above the contributions under What decides it — for example “This overrides everything below until <sunrise>.” The rows are then headed What went into the M, so they still add up. A cap is only reported when it changed the number: a surf score of 4 at night was decided by the surf, not the dark.

Air quality applies to swimming and walking: US AQI ≤ 50 +5, ≤ 100 −3, ≤ 150 −15, worse −30. Wind and its direction come from buoy 41024 first and the NWS grid second; the shore normal is taken as 190°, so wind from about 10° is offshore. “Dark” means the sun is below −6° (past civil twilight).

A warning rides along with the swimming score and is shown even when the headline is positive: at rip level 3, “High rip current risk — NWS advises staying out of the water.”; at level 2, “Moderate rip current risk and no lifeguards. Stay where you can stand, and know how to escape a rip.”; with no rip forecast at all, “No rip current forecast right now, and no lifeguards. Stay where you can stand, and know how to escape a rip.” With none of those, a burn time of 20 minutes or less gives “Strong sun: bare skin burns in about N minutes.”

Out of season NWS stops issuing the surf zone forecast, but it still issues a Rip Current Statement when it expects life-threatening rips. With no forecast and a statement in force, the rip level is taken as High (level 3): the swimming score and warning above follow it, and the Rip risk tile on Today reads High with “NWS statement” under it. Today’s hour-by-hour scores follow it too, so the best time and the Plan tab’s picks for today agree with the tile. When there is a forecast, its own level is used.

The overall score is 0.6 × the best activity + 0.4 × the mean of all four — a flat day is still a great walking day, and that is the answer the question wants. If any live NWS alert is active (drills and tests are filtered out) the headline is Active alert: <event>; otherwise ≥ 75 → “Go. Best for …” (or “Good day for swimming.” when swimming is the pick — the app never says “Go.” beside swimming), ≥ 55 → “Worth the trip — good for …”, ≥ 35 → “Mixed. … is the pick if you go.”, else “Not a great day for the beach.” With no rip forecast, swimming is never the pick. If nothing could be scored at all, the score is 0 and the headline reads “No data” (or the active alert).

This headline is what the widgets show. On Today, from version 1.1, the summary line described below takes its place at the top of the screen.

Every daylight hour is scored by the same function as the activity tiles, fed that hour’s forecast instead of this minute’s readings — so the best time and the tiles can only differ by what the forecast says about the hour. Three activities are scored per hour: swimming, beach walk & shelling, and surfing. Sunset watching is left out; it is one moment a day.

Per hour, the inputs are the NWS gridpoint forecast for wind, air temperature, cloud and rain (Open-Meteo’s hour when the gridpoint is missing), the tide height and trend at that hour, the hour’s UV, the sun’s elevation (so caps for dark apply), and the day’s surf zone forecast and water temperature. Swell period and air quality are observations, so they only count for today.

The best window is built around the single best hour of any activity:

  1. Find the best hour. Ties go to the earlier hour.
  2. If that hour scores below 55 (good), there is no best time today.
  3. Otherwise extend the window either side while that same activity stays within 8 points of its peak and at or above 55. The window ends at sunset at the latest.
  4. List every activity that is good somewhere in the window, best first.

Today’s best time runs from now on. Once today’s daylight is spent — under an hour left before sunset — it is tomorrow’s instead, and Today says Best time tomorrow. The Plan tab’s per-day Swimming pick uses the same procedure for swimming alone.

Who writes the summary at the top of Today?

Section titled “Who writes the summary at the top of Today?”

The headline and sentence at the top of Today (for example “Good beach afternoon.” then “Low rip risk and 79° water. Best: shelling, 1:30–5 PM.”) are written in three tiers:

TierWriterUsed when
1Claude Haiku 5.5Normally.
2gpt-oss-120b on Cloudflare Workers AIClaude’s call fails, or its draft fails the check.
3A fixed templateBoth fail.

No writer computes anything. The server works out every figure first — rip risk, water and air temperature, tide height and next high or low, rain and thunder times, tonight’s sunset and outlook, the alerts already in force, the best window, every activity score — and the writer is only given those. Rain or thunder that has already started is given as now, not as a clock time, and an alert that starts later (an advisory beginning tomorrow morning) is left out until it is in force. Then code checks every draft before it is shown. A draft is rejected if:

  • it contains a number or time that is not in the facts, or a real one next to the wrong thing (a tide time given as the time to swim);
  • it denies forecast rain or thunder, or leaves out the thunder;
  • it states a rip risk other than the forecast’s, any rip risk when there is no forecast, or no rip risk or forecast when there is one;
  • it mentions swimming without saying not to, when the rip risk is moderate, high or unknown, or it is night;
  • an active NWS alert isn’t named in the headline, or it names an advisory, warning, statement or weather watch that is not in force;
  • the headline names the wrong part of the day, or either line is too long (headline 40 characters, sentence 220).

So the worst a model can do is be replaced by the template. The template is plain: “Good beach morning.”, “Decent beach afternoon.”, “Not much of a beach evening.”, “Night at the beach.”, or “<Alert> in effect.”, then the rip risk and water, any thunder or rain, and the best time (or the next tide).

A summary is kept while it still passes the current facts. It is rewritten when the facts change in substance, at most once an hour, and replaced within 15 minutes if the facts move out from under it (the tide height it quoted has changed). Between rewrites you may see the template wording; that is expected. How the models were chosen is shown on More › How it’s built.

Judged on the Open-Meteo cloud profile at the sunset (or sunrise) hour — never on total cloud cover, because “99% cloud” that is all high cirrus is close to ideal while 60% low stratus kills it. The score starts at 35:

FactorPoints
The canvas — high cloud × 0.8 + mid cloud × 0.4< 10 → +5 “clear sky with little color” · < 25 → +18 · 25–75 → +35 “broken high cloud to catch the color” · ≤ 90 → +18 “thick high cloud that may mute the color” · higher → +8 “solid high overcast, likely to stay gray”
The blocker — low cloud≤ 15% → +20 “clear western horizon” · ≤ 40 → +5 · ≤ 70 → −15 “low cloud likely to block the sun” · higher → −30 “thick low cloud”
Visibility≥ 9 mi +5 · < 4 mi −10 “hazy”

Ratings: spectacular ≥ 78, promising ≥ 60, average ≥ 40, else flat. The same function scores the sunrise hour with “eastern horizon” in the wording. If the sunset hour is more than two hours outside the forecast the outlook is not scored at all. Burn time is 200 ÷ UV index minutes, never under 5.

The rip level is read from the NWS Surf Zone Forecast’s words for the “Ocean Isle Beach West” sub-zone: None = 0, Low = 1, Moderate = 2, High = 3. That level drives the swimming score above, the Swim verdict on each day’s screen under the Plan tab’s Seven days, and the safety push (a rise to High). That per-day swim call is: level 3 → poor “stay out of the water”; level 2 → caution; water below 68°F → caution “cold enough to cut a swim short”; otherwise good. Days beyond the sub-zone rows carry the whole-zone forecast and say so; days with no rip forecast at all only get a verdict if the water is below 68°F (caution). Good needs a rip forecast, so with none, warmer water gets no verdict.

Only launches from Cape Canaveral / KSC and Wallops Island are modelled. For each, the Worker infers a launch corridor from the target orbit (ISS-class flies northeast with high confidence; GTO east/southeast; 35–50° inclination is ambiguous and both corridors are tried; unknown orbit tries 45°, 90° and 135° at low confidence), propagates a representative ascent for 10 minutes in 10-second steps, and at each step asks three things: how high above Sunset Beach’s horizon the vehicle is, whether its plume is still in sunlight, and how dark your sky is.

  • Below 1° elevation a moment counts for nothing.
  • Lighting dominates: full daylight scores 2; sun just down 25 (8 if the plume is in shadow); the twilight sweet spot (sun −3° to −18°) 55 lit / 22 unlit; full dark 45 / 20 — plus a geometry term that keeps rising with elevation.
  • Range: best moment under 500 km +10, under 800 +6, under 1200 +3.
  • Cloud is a hard gate: ≤ 20% +10, ≤ 50% +5, ≤ 80% −5 “Mostly cloudy (N%) — cloud could easily hide it.”, higher −25 “Overcast (N%) — very likely hidden.” More than 7 days out there is no cloud forecast and the note says so.
  • Daylight is a ceiling of 12, whatever the geometry.

Verdicts: ≥ 70 “Excellent. Look …”, ≥ 50 “Good chance …”, ≥ 30 “Marginal …”, else “Unlikely to be seen from here.”, plus “Not visible — stays below the horizon.” and “Daytime launch — not visible from here.” A launch is visible — and eligible for the Next visible launch card, the Live Activity and launch alerts — only when it clears the horizon, is not in daylight, and scores ≥ 30. Every launch carries where and when to look (compass point, elevation, T+ seconds) and a confidence that drops when the corridor is ambiguous or the peak is under 3°.

Each “When to go” window (Shelling, Bird Island & the Kindred Spirit mailbox, Dolphin watch, Ghost crabs, Turtle hatch watch, Right whale watch, Shorebirds, Moonrise over the ocean, Sunrise, Sunset) starts at 50 and scores only what varies between occurrences — the tide height and range for a shelling low, low cloud on the horizon for a moonrise, rain for a hatch watch. Things true of every occurrence (“Nests almost always hatch after dark”) print at 0 points so the reasoning still shows. Ratings use the same 75/55/35 bands. A window whose contributions are all zero is unscored and shows its reasoning without a rating. Every window also states its certainty — deterministic (tide, sun, moon: knowable years out, so it can be scheduled as a local alert) or forecast (rests on weather and decays with horizon) — and what it promises, which says exactly what it guarantees: a shelling window scores the conditions that expose shells, never what is lying there.

For example, a Shelling low scores the low-tide height (a minus tide +24, ≤ 0.5 ft +17, ≤ 1.2 +8, ≤ 2 −2, higher −18), the tidal range (≥ 5.5 ft spring +13, ≥ 4.5 +6, neap −8), that day’s surf (≥ 4 ft +14, ≥ 2.5 +6, calmer −6), an early low within 3 hours of dawn +10, and the hour’s weather. The Sunset and Sunrise windows carry the sunset model’s own score, unaltered, so the almanac and the Today card can never disagree.