Atlantic Salmon Are Not a Bite Forecast

In July we caught the app doing something indefensible.

Atlantic salmon, scored by the same local engine we use when a species has no dedicated model, could come back as a near-perfect day. Cold water. Summer. High confidence. The kind of number that makes someone book a river.

Adult salmon in freshwater are not feeding like pike.

They are in the river to migrate and spawn. They will take a fly. That is not the same biology as a resident predator whose metabolism tracks water temperature and whose day is about hunting. Using a juvenile growth optimum, or a generic cold-water feeding curve, as a bite probability for an adult salmon is a category error. It looks scientific. It is not.

The same trap waits on sea trout

Sea trout on the coast can feed hard. Sea trout in a river, holding or running, are a different phase. One species name is not one forecast.

Until the model can tell those phases apart, freshwater sea trout gets the same caution as salmon. Coastal sea trout does not. Mixing them into one rule would be the other lie: pretending a fish in the sea is already on the redd.

Resident brown trout and rainbow trout stay on the ordinary path. They live in the water you are fishing. They are allowed to be hungry.

What we did instead of a pretty number

We capped the local fallback. The app can still point at a river and say the day is not empty. It cannot present salmon as a 100-out-of-100 feeding event. It cannot let a generic boost push the call back up to certainty.

That is a holding pattern, not a finished salmon model. A proper one has to know the phase: marine feeding, estuary, river migration, holding, spawning. It has to know whether you are even on a salmon river. Until that exists, a modest, guarded call is the honest product.

The alternative is the fishing-app default: never look stupid, never say "we do not know this fish yet," always fill the ring.

Why this belongs in public

People download a fishing app because they want a number they can act on. The trust is spent on the days the number should have been lower.

Salmon was the visible failure. The underlying rule is broader. If the species is not in the model, GILLI should not invent a confident day for it. Unknown stays unknown. A withheld score is better than a perfect one with the wrong biology behind it.

If you fish salmon or sea trout, treat the present call as directional and conservative. The app is not pretending to be a ghillie. It is refusing to fake one.