Most surf forecasts hand you a wall of numbers — wave height in feet, period in seconds, wind speed and direction, tide state — and leave the interpretation to you. That's real information, but it isn't a decision. Standing in a kitchen deciding whether to load a board into the car, nobody wants to reconcile twelve numbers against their own experience. They want an answer.
Boardcall, an iOS app whose App Store subtitle is simply "Which board to bring today," answers with one of four words: fire, fun, marginal, or skip. That's the entire verdict — no score out of ten, no percentage, no star rating. The app's own design goal is a six-second read: glance at the screen and know whether to go and what to grab. A number invites interpretation. A word like "marginal" doesn't — it means only if you're keen, and there's nothing left to translate.
What actually decides the word
The board pick — which surfboard actually gets recommended — comes out of a deterministic pipeline that runs the same way every time. The verdict word is a separate question: Boardcall's briefing can be written by a plain, template-based writer that computes the verdict directly from the same conditions, or by a language model that's handed those same numbers and writes the verdict itself as part of the prose. Either way, the board pick underneath doesn't depend on which one wrote the words — that part never involves an AI.
First, the app converts the raw offshore swell reading — height, period, direction — into an estimate of the actual breaking wave face height at a specific spot, weighted by how exposed that spot is to open-ocean swell and how well the swell direction lines up with the way the coastline faces. From there it computes a target board volume for that surfer, based on their weight and stated skill level, adjusted further for the period and size of the swell. One detail worth calling out because it runs against intuition: both very small, weak swell and very large swell push the target volume up, not down — small waves need more float to catch at all, and big waves need more paddle power to get into them early. It isn't a straight line from "smaller wave, smaller board."
Every board the surfer owns is then scored against that target — split evenly between how well its volume fits and how well its shape category fits the day's conditions — with small, capped adjustments for details like fin setup, and an optional nudge based on a surfer's own logged sessions that is capped so it can never override the top-scoring pick. All of that happens before a single sentence of prose gets written.
What the AI is, and isn't, allowed to do
When a language model writes the briefing, it's handed the pipeline's output — the computed face height, the target-volume math, the ranked boards — and its job is to write two or three plain-English sentences, a headline, and the handful of short fields that make up the briefing, including the verdict word and a one-line reason for the board it names. It's allowed to name a different board than the deterministic top pick, but only if it states a reason; it isn't allowed to invent one. The style is enforced explicitly — no emoji, no exclamation marks, at most one number in the whole briefing, and a banned-phrase list ("Get out there," "Dawn patrol," "Send it," "stoked") that would otherwise make it read like marketing instead of an answer. The model is also barred from calling its own pick "top-ranked," "first," or "best" — the surfer never sees a ranking, so referring to one would describe the machinery instead of giving a straight answer.
The part that isn't a suggestion
The design decision worth dwelling on is what happens on a day flagged as beyond what a given surfer, at their own stated skill level, should be paddling into. On that day, the verdict is held to "marginal" at best — and for the single most serious case the rule accounts for, a beginner-level surfer facing real size over a shallow reef, it's held all the way down to "skip." Either way, it is not permitted to read "fire" or "fun," regardless of what the AI-generated text around it says. That rule sits between the model's output and the screen as a separate, non-generative check. It doesn't rely on the model reading the conditions correctly or staying in character. A language model can misjudge tone or simply generate an overly upbeat pass on a given try; a hard limit on the verdict can't be written around by a good sentence.
That's a narrow, unglamorous piece of engineering, and it generalizes past surfing. Any product that lets a generative model produce advice with a real physical consequence attached — go or don't go — benefits from separating what the model is free to write from what claim the output is actually allowed to make. The prose can vary. The limit shouldn't.
Where the real edge is, stated plainly
Boardcall isn't the first app to suggest a board for a given spot. Surf Forecast: Quiver, a separate app in the same App Store weather category, already does per-spot board recommendations. What's different here is that Boardcall models a surfer's actual quiver by real dimensions and volume rather than issuing a generic size-based suggestion, and wraps that pick in a briefing that's constrained the way described above — honest about risk even on a day that reads well on paper.
It's also honest about its own limits: it collects an anonymous device identifier to limit abuse of the AI-generation feature, and each AI-written briefing sends that day's conditions along with the surfer's skill, weight, and ranked boards to generate the prose — that's real data leaving the device, not "nothing." The spot catalog covers 193 surf breaks, not an exhaustive global list. And the underlying board-matching and safety logic isn't a paid feature — it runs the same way for every surfer; what a paid tier changes is only how often the AI-written version of the briefing gets generated.