Lab · Data sonification · Web Audio API + NYISO live data · Nov 2026
Listening to the grid
New York’s grid publishes its fuel mix every five minutes. I turned it into one minute of music a day: each power source is an instrument, the clean share sets the key and demand sets the tempo. Then I built the system that keeps it playing: a live view, a calendar of every day, and an archive that grows on its own.
The 30-second version
7 min read
Role
Designer and engineer, end to end
Prototype
grid-symphony.vercel.app
Built with
Web Audio API, Vercel, GitHub Actions, Python
Data
NYISO fuel mix and zonal load, every 5 minutes
Grid dashboards answer “how much” and lose people at “so what”. I treated the grid as a score instead: each fuel is an instrument, the clean share picks the key and demand sets the tempo, so a strained evening sounds different before you read a number. Every sound is backed by the same data drawn on screen, and the whole thing runs on static files and a browser synth.
5 min
between readings, from NYISO’s live feed
373 days
in the archive, saved every morning by a scheduled job
0
audio files stored; every minute is synthesized in the browser
Live prototype: grid-symphony.vercel.app · Source and raw data on GitHub
The question
A grid you can feel, not just read
New York publishes its fuel mix every five minutes. The data is public, precise and almost entirely ignored, because a stacked chart of megawatts asks a lot of someone who doesn’t already know what a good hour looks like. I wanted the opposite entry point: something you notice first and read second.
So the brief was: make the state of the grid legible in one minute, to someone with no energy background, without bending the numbers.
The mapping
Each power source plays an instrument
Sonification lives or dies on its mapping. I picked sounds whose character matches how each source behaves on the grid, so the metaphor teaches something true instead of just decorating the data.
Input | Sound | Why it fits |
|---|---|---|
Nuclear | Low drone | Steady baseload that barely moves, like the plants |
Hydro | Flowing pad | Slow swells; louder when Niagara and St. Lawrence run high |
Wind | Breath and bells | Airy and variable; bells ring more often as output rises |
Solar and other renewables | Plucked melody | Intermittent; more output, more notes |
Gas and dual fuel | Buzzing saws, brass stabs | Harsher, with a clashing note, as the clean share drops |
Clean share | Key | Major above ~54%, Dorian in the middle, Phrygian below ~46% |
Statewide demand | Tempo | A soft heartbeat from about 60 to 135 bpm |
Two variables carry the story
The instruments add texture, but key and tempo do the explaining. A listener doesn’t have to pick out seven voices to hear a day move from open and bright to tense and fast. It’s the same rule I’d apply to a dashboard: one or two primary encodings, and everything else is detail on demand.
The system
Four zoom levels, one colour scale
The views step out in time: Today (live, through the latest reading), Trends (the last 30 days, plus the month compressed into one minute), Year (every day and every hour) and All time. Each answers a different question, and every one lets you click into a day and hear it.
The Year view is where the piece became more than a toy. Laid out as a calendar, the seasons appear on their own: a clean spring and autumn when hydro is strong and demand is low, and a red summer when air conditioning pushes the grid onto gas. 2026 so far: 46% clean across 312 saved days. The cleanest day was 17 October at 64%; peak demand hit 31,221 MW at 7 PM on 2 July.
The colour scale is the key signature
The heatmap runs from strained red (under 40%) through a neutral sand (45–50%) to clean blue (55% or more). I set those bins against the same thresholds the music uses, so what you see on the calendar and what you hear when you click it agree. Blue days sound bright. Red days sound tense.
Hard calls
Four trade-offs I made on purpose
Synthesize, don’t record
Static files over a database
Never sound alone
Show the gaps
Design engineering
The pipeline is part of the design
A live data piece is only as good as its freshness. A scheduled GitHub Actions job runs every morning: it saves yesterday, re-fetches the day before, rebuilds the rollups and commits. Vercel redeploys on each commit. The scripts use nothing beyond the Python standard library, so the archive keeps growing without anyone tending it.
Layer | Choice | Why |
|---|---|---|
Data | NYISO real-time fuel mix and zonal load | Public, every 5 minutes, across 11 zones |
Storage | JSON files in the repo | Versioned, diffable, free to serve |
Updates | Daily scheduled job | Fresh by morning and self-correcting |
Live view | One serverless function | Gets around the browser’s cross-origin limit |
Sound | Web Audio API | No stored audio; a new mapping re-scores everything |
Limits
What it can’t tell you
In-state only. Clean share counts New York’s own generation. Imports show as the gap between generation and demand, not as clean or dirty.
“Clean” is a definition. Nuclear, hydro, wind and solar count as clean. That’s a choice, and a different definition would change the colours.
The music is interpretive. The key and tempo thresholds are mine. They’re fixed and documented, but another mapping would tell the same story with different emphasis.
2026 is a partial year. Figures run through early November.
Reflection
Make the data noticeable before you make it readable
Most data products start from the chart and hope people care. This one starts from a feeling, a day that sounds tense, and lets the chart explain it.
The principle I’m taking into product work: pick the one or two encodings that carry the story, make every other channel agree with them, and never let the delightful layer drift from the data.
Try it
Hear two days
Open the Year view and click a day in late July, then one in mid-October. Same state, same instruments, a different grid.