Subfork Examples

Tidal Resonance

NOAA tide predictions from San Francisco (9414290), Seattle (9447130), and The Battery, New York (8518750) become three flowing SVG ribbons. A shared time scrubber reveals how their predicted heights differ throughout one returned UTC day. This is an educational visualization, not an observation, weather forecast, storm-surge forecast, or navigation/safety tool.

Try It

  1. Import graphs/html/tidal-resonance.subfork.json and run it.
  2. Bind HTML Preview to view.html if the recommended layout is not applied.
  3. Scrub through the day, isolate a station, switch meters/feet, or change palette.
  4. Inspect the exact selected six-minute sample in the station cards. The slider supports keyboard interaction; narrow screens scroll the chart horizontally while the cards stack vertically.

No API key, Three.js, or AI service is required. No application rebuild is needed if HTML-safe JSON Stringify is already available. The single JSON contains a formatted, editable HTML template. Custom inputs are explicitly exposed and the unused HTML context input is hidden. No source folder or generator is committed.

Sources And Data Contract

Uses the NOAA CO-OPS Data API. Each of three branches makes one request with product=predictions, date=today, interval=6, time_zone=gmt, datum=MLLW, units=metric, and JSON format. Requests have a 30-second timeout and 500 KB body limit. There is no polling, pagination, or browser-side request. Rerun the graph to refresh.

today is resolved by the provider; the page labels the actual returned UTC date, not the browser’s date or timezone. To request a fixed date, replace date=today in each URL with matching begin_date=YYYYMMDD&end_date=YYYYMMDD. Keep units, datum, interval, and timezone unchanged unless also updating the renderer’s contract. To change a station, update its URL and its name/ID in the dataset node.

The tidal-resonance/1 envelope contains three labeled station payloads, HTTP statuses, source URLs, and headers, with declared meters/MLLW/UTC semantics. JSON parsing happens in the graph. Validation, time alignment, unit conversion, and ribbon construction happen in the renderer. The dataset retains raw NOAA prediction records; view validation does not silently rewrite its source output.

Visual Meaning And Limits

Each ribbon connects valid six-minute samples in time order. Height uses a shared linear scale computed across all valid station samples, including zero. The scale stays fixed when a station is isolated. Filled areas extend to each station’s local MLLW zero; these are different local reference datums, not one global sea-level surface. Negative predictions are retained and mean below local MLLW, not negative water depth. Color only distinguishes stations.

The time scrubber selects exact samples, not an interpolated measurement. Straight connecting segments are visual interpolation. Sampled min/max values are not exact high/low tide times. The browser never invents extra predictions or synthesizes fallback tides. There is no animation or hidden autoplay.

The renderer expects three station groups and at most 240 records per group. It validates UTC date strings, six-minute alignment, numeric meter heights within an intentionally generous +/-100 m bound, and duplicate timestamps. It reports invalid/off-day, duplicate, and over-cap records. The first usable station sets the displayed UTC day; other-day samples are excluded and reported, preventing silent alignment across midnight. Missing intervals split the paths instead of connecting across gaps. Missing selected-time samples say Unavailable.

HTTP/API errors with parseable JSON can produce partial station output with a visible warning; if all stations are unusable the page shows an error instead of a fabricated chart. An execution can complete with an error page because status codes are data. Transport or JSON-parse failures fail the graph and do not produce a new combined artifact. There is no cached-source recovery policy in this seed.

Source text is inserted with textContent, and dataset embedding uses html_safe=true. The page has no external scripts, fonts, or images. Validated NOAA station URLs are shown as copy fields because new tabs are blocked by the current preview sandbox. Hosting and form-triggered execution remain separate planning work; this prototype does not change sandbox permissions.

Outputs And Validation

Outputs: html, HTML response, dataset, and JSON data_response. Fourteen runtime nodes replayed three freshly fetched NOAA responses, each with 240 predictions, through dataset and response generation. Renderer checks covered unit conversion, time endpoints, palettes, station isolation, gaps, duplicate and invalid samples, literal hostile text, and partial/all-source failures. Desktop and narrow-screen output were inspected with headless Chrome; a dashboard import/run remains the final manual check. No application code or permanent prototype-specific runtime tests were added.