AmarSELMANI/algerian-coast-marine-bot

Daily marine weather Telegram bot for the Algerian Mediterranean coast. Deterministic safety assessment in JavaScript; an LLM writes the prose but cannot change a rating. Built on n8n and Open-Meteo, verified against METAR observations. # Algerian Coast Marine Weather Bot A Telegram bot that sends a daily marine outlook for six cities on the Algerian Mediterranean coast at 07:00 `Africa/Algiers`, rating conditions for **sailing**, **shore fishing**, **small-boat fishing** and **swimming**. Built on n8n, Open-Meteo and the Google Gemini free tier. ``` 🌊 Marine outlook — Oran 📅 2026-08-27 | Forecast window: 06:00-19:00 (Algiers time) ⛵ Sailing: CAUTION 🟡 Gusts hold near 24 kt through the afternoon over 0.9 m waves at 4.8 s — short, steep going for a small boat. The morning is calmer. Better window: 06:00-11:00 (GOOD) 🎣 Fishing: • Shore: GOOD 🟢 — operational comfort and safety only • Small boat: AVOID 🔴 — operational safety only 🏊 Swimming: CAUTION 🟡 0.92 m of wave height with a 26.4°C sea; the chop is the limiting factor. Key forecast: Wind 10 km/h (5.4 kt) W, gusts to 62.6 km/h (33.8 kt); waves 0.92 m, period 4.8 s; swell 0.76 m; sea 26.4°C. ⚠️ strong wind, short, steep chop. ⚠️ Official marine warnings were NOT checked — no machine-readable feed exists for this coast. Check them yourself. Forecast only — not an observed sea state. Verify official marine alerts, harbour guidance, lifeguard flags and conditions at the launch or beach before going out. ``` --- ## The idea Marine forecasts are easy to find and hard to read. "1.2 m at 4 s" and "1.2 m at 9 s" are the same number and completely different days on the water. This turns the numbers into an answer, without pretending to more certainty than a forecast can carry. Two rules shape the whole design: **Code decides, the model writes.** Every rating comes from plain JavaScript over the forecast data. The LLM receives an assessment that is already final and turns it into readable prose. It cannot change a rating — not because it is asked not to, but because the ratings are overwritten from the deterministic assessment after the response is parsed. **Nothing positive is ever asserted on thin evidence.** Missing essential data produces A …