How SaltCast Works

SaltCast analyzes real-time tidal, weather, lunar, and water temperature data to predict the best saltwater fishing windows — updated every hour for 10 coastal locations.

The Overall Score

Every hour at every location gets a score from 0–100. This score combines four factors, each weighted by how much it actually influences saltwater fish behavior:

🌊 Tides30%

💨 Weather30%

🌙 Moon / Solunar25%

🌡️ Water Temp15%

The raw weighted score is then run through a power curve that pulls average conditions down significantly. This means a 90+ score is genuinely exceptional — not just a good tide on a calm day.

What Each Factor Measures

Tides (30%)

Rising tides score highest — baitfish push into the shallows, predators follow. Falling and slack-low tides score lowest. Data comes from NOAA tide prediction stations closest to each location.

Tide Condition	Score
Rising, upper half	90
Rising, lower half	65
Slack high	40
Falling, upper half	25
Falling, lower half	15
Slack low	5

Weather (30%)

Light winds (under 10 kts) and stable high pressure produce the best scores. High winds chop up the surface, disrupt baitfish, and make fish lock tight to structure. Heavy cloud cover slightly reduces scores as it limits light penetration.

Moon / Solunar (25%)

Solunar theory holds that fish feed most aggressively during major periods (around moonrise and moonset, roughly every 12.4 hours) and less intensely during minor periods (every 6 hours). Between periods, fish tend to be inactive.

Solunar Period	Score
Major period	90
Minor period	50
Between periods	10

Water Temperature (15%)

The sweet spot for most Gulf Coast and Atlantic inshore species is 65–78°F. Below 55°F or above 85°F, fish become lethargic and feeding activity drops significantly.

The Fish Scale

Each hour in the slideout panel shows 0–4 fish icons representing fishing quality for that specific hour:

🐟🐟🐟🐟

Prime

Major solunar + high overall score. Get on the water.

72+ & major

🐟🐟🐟

Good

Strong conditions — active bite expected.

72+

🐟🐟

Decent

Worth fishing — fish are moving.

60–71

🐟

Slow

Below average — be patient.

52–59

— slow

Dead

Poor conditions — consider waiting.

<52

The Heatmap Strip

Each day card in the 7-day forecast shows a 24-segment color strip — one segment per hour. The color tells you at a glance when the bite turns on:

Bright green

Score 70+ — prime fishing window. Plan to be on the water.

Yellow-green

Score 58–69 — decent conditions, worth going.

Amber

Score 45–57 — marginal, fish will be selective.

Dark navy

Score <30 — dead zone, save your energy.

The AI Fishing Forecast

Each day in the slideout panel includes a plain-English forecast written by Claude (Haiku) — Anthropic's AI. The forecast synthesizes the day's conditions into specific advice: which species will be active, what tactics to use, and what time to be on the water. It's generated fresh with every pipeline run.

Data Sources

NOAA Tides & Currents

Hourly tide predictions and water temperature from official NOAA tide gauge stations at each location.

Open-Meteo

Free, open-source weather API providing hourly wind speed, wind direction, surface pressure, and cloud cover.

Solunar Calculation

Major and minor feeding periods calculated from lunar position using the synodic month cycle (29.53 days).

Anthropic Claude

AI-generated plain-English fishing narrative for each day, based on the computed conditions and scores.

Score Alert Emails

SaltCast automatically emails bob@rensho.ai whenever any location scores 90 or above in the next 24 hours. The email includes the location, peak score, best time windows, tide direction, and wind speed — so you can plan a trip on short notice when conditions align perfectly.

Technical Architecture

SaltCast is a fully automated prediction engine built on a Hetzner VPS using n8n for orchestration, Supabase (self-hosted) for storage, and a static HTML frontend served by Caddy.

Stack Overview

n8n

Orchestration
Data pipeline + AI

→

Supabase

PostgreSQL
Single source of truth

→

Caddy

Static frontend
saltcast.rensho.ai

Component	Details
Server	Hetzner VPS — root@5.161.91.134
n8n	Docker container (custom image with ffmpeg). Access at n8n.rensho.ai. Do NOT restart with docker compose — use `docker restart n8n`
Supabase	Self-hosted Docker stack at supabase.rensho.ai. Schema: `saltcast`
Caddy	Docker container. Config: /opt/rensho/caddy/Caddyfile. Reload: `docker exec caddy caddy reload --config /etc/caddy/Caddyfile`
Frontend	/opt/rensho/www/saltcast/ — single index.html

Important: When editing the Caddyfile, always use scp from your Mac to write directly to the original file inode. Using cat > or Python open('w') creates a new inode that the Docker bind mount won't follow.

n8n Workflows

SaltCast — Fishing Prediction Pipeline (HlkH6OgQRoD9BUgJ)

Currently inactive (manual trigger while testing). Set to hourly once stable. Takes 5–8 minutes per full run across all 10 locations. Max concurrency = 1 prevents overlapping runs.

Node	Purpose
Every Hour	Schedule trigger — change interval here
Get Active Locations	Fetches all locations where active=true from Supabase
Loop Over Locations	Iterates one location at a time
Fetch Weather	Open-Meteo API — wind, pressure, clouds (7-day hourly)
Fetch Tides	NOAA Tides & Currents — hourly predictions (7-day)
Fetch Water Temp	NOAA water temperature. continueOnFail=true, retryOnFail=true (NOAA timeouts are common)
Calculate Scores	Core scoring algorithm — weather, tide, solunar, water temp → power curve → overall score
Batch by Day / Upsert Predictions	Upserts 168 hourly predictions per location to Supabase
Build Daily Summaries	Aggregates 24 hours into daily summary stats
Build Narrative Prompts	Constructs plain-text prompt for each day
Generate Narratives	Calls Claude Haiku for each day's AI forecast (max 300 tokens)
Merge Narratives	Merges narrative text back onto summary objects
Upsert Summaries	Writes to daily_summaries table. Empty response = success (204 No Content)

SaltCast — Score Alert (90+) (NFEHqoCMw5GdLaTI)

Active, runs hourly. Queries predictions ≥90 in next 24 hours, sends styled HTML email from saltcast@rensho.ai via Resend SMTP (credential ID: dRfO8HPussxphvHM).

Supabase Schema

All tables live in the saltcast schema. PostgREST is configured to expose this schema via PGRST_DB_SCHEMAS.

locations

      id           uuid PRIMARY KEY

      name         text (e.g. "St. George Island, FL")

      lat          numeric

      lng          numeric

      noaa_station_id  text (NOAA tide gauge ID)

      timezone     text (e.g. "America/Chicago")

      active       boolean (false = excluded from pipeline)

predictions

      id                 uuid PRIMARY KEY

      location_id        uuid REFERENCES locations

      forecast_time      timestamptz

      overall_score      integer (0–100, power-curved)

      weather_score      integer

      tide_score         integer

      moon_score         integer

      water_temp_score   integer

      wind_speed         numeric (knots)

      tide_height_ft     numeric

      tide_direction     text (rising/falling/slack-high/slack-low)

      water_temp_f       numeric

      moon_phase         numeric (0–1)

      is_solunar_major   boolean

      is_solunar_minor   boolean

      UNIQUE (location_id, forecast_time)

daily_summaries

      id                 uuid PRIMARY KEY

      location_id        uuid REFERENCES locations

      forecast_date      date

      avg_score          integer

      max_score          integer

      best_window_start  integer (hour 0–23)

      best_window_end    integer (hour 0–23)

      moon_phase         numeric

      moon_name          text

      tide_highs_count   integer

      ai_narrative       text (Claude-generated)

      UNIQUE (location_id, forecast_date)

Scoring Algorithm

The scoring pipeline runs entirely in the Calculate Scores n8n Code node. Key design decisions:

Upsert, not insert

Every pipeline run overwrites existing rows using the unique constraint. Database stays at a fixed size — 10 locations × 168 hours = 1,680 prediction rows.

Power curve

Raw scores are transformed via Math.pow(score/100, 1.35) × 100. This pulls average conditions (score 60–70) down to 47–57, making high scores genuinely meaningful.

Timezone-aware dates

forecast_date generation uses the location's local timezone, not UTC. Prevents date shift for locations where the pipeline runs near midnight UTC.

NOAA fallback

Water temp node has continueOnFail=true and retryOnFail=true. If NOAA times out, scoring continues with 72°F fallback. Prevents one flaky API from killing the whole run.

Frontend Architecture

The frontend is a single static HTML file — no build step, no framework, no dependencies beyond Supabase JS and Google Fonts. The entire app is one file: /opt/rensho/www/saltcast/index.html.

Data flow

On load, the frontend resolves the selected location's UUID from Supabase, then fetches 7 days of predictions. If Supabase is unavailable, it falls back to locally-generated mock data so the UI is never blank.

      // Boot sequence

      populateLocationDropdown()   // load locations from Supabase

      resolveDefaultLocation()     // get UUID for St. George Island

      fetchData()                  // load 7-day predictions

      render()                     // draw heatmaps, gauge, conditions

      // On day card click

      openPanel()                  // show slideout

      renderHourly()               // draw 24-hour fish timeline

      fetchNarrative()             // read ai_narrative from daily_summaries

Deploy pattern

      # From Mac — bundle and deploy

      scp ~/Downloads/deploy.tar.gz root@5.161.91.134:/tmp/

      ssh root@5.161.91.134 'cd /opt/rensho/www && tar xzf /tmp/deploy.tar.gz && rm /tmp/deploy.tar.gz'

      # tar structure

      saltcast/

      saltcast/index.html

      saltcast/images/saltcast-logo.svg

      saltcast/images/saltcast-favicon.svg

      saltcast/images/rensho-logo.png

Credentials

Service	Credential	n8n ID
Anthropic	Anthropic account	Lguuwcjilf20kdRK
Resend SMTP	SMTP account	dRfO8HPussxphvHM

Adding a Location

Insert a row into saltcast.locations with the location's coordinates, NOAA station ID, and timezone. The pipeline will automatically pick it up on the next run.

      INSERT INTO saltcast.locations (name, lat, lng, noaa_station_id, timezone, active)

      VALUES ('Your Location, ST', 00.0000, -00.0000, '0000000', 'America/Chicago', true);

Find NOAA station IDs at tidesandcurrents.noaa.gov — search for the nearest tide gauge to your location.