Spectral Brand Theory · measurement layer

Read the perception
you don't control

The Brand Spectrometer reads how different audiences actually perceive a brand — cohort by cohort, across eight dimensions, from public artifacts — and tells you which of those readings are real and which sit below the noise.

Now reading: · cohorts · ·

The problem

One number hides a population that disagrees

The instrument's scale

Eight dimensions of a perception

A brand is a composite signal across these eight wavelengths. Two cohorts can hold the same overall regard while weighting the dimensions completely differently — a metamer. Toggle the cohorts below to see it.

A real read

Five cohorts, one brand

Each cohort's inferred eight-dimension specification. Toggle cohorts to overlay them; switch between bars and radar. The press cohorts read high and rich; owners and online debaters read lower and re-weighted — same brand, different perceptions.

Who agrees · who doesn't · what's real

A map of perceptual distance

Hover any cell to see the pair, the dimensions driving their distance, and whether the gap clears its own noise floor. Flip the switch to grey out every sub-resolution pair — what remains is real.

Noise-floor gate

Hover a pair

Distance is 1 − cosine similarity between two cohorts' eight-dimension vectors. The noise floor is the wobble from swapping AI operators alone. Above the floor = real; below = we can't tell.

In this worked example, the owners ↔ online-debaters split is the largest and is resolved; several press-to-press gaps sit below the floor and are honestly reported as sub-resolution, not findings.

The read, in numbers

Every cohort, every dimension

Hover a row to highlight that cohort in the chart above. These are the raw vectors the whole read is built from.

The managerial reading

What you actually do with it

It's an instrument, not a slideshow

Try it on your own brand

Bring your own atlas — the whole read re-renders for it. Nothing is uploaded: every atlas is processed entirely in your browser.

Make one in three steps with your own AI

  1. Get the kit. Download schema Download template Full guide
  2. Run it on your AI (Claude, ChatGPT, Gemini…) with the schema attached and your raw data. It returns a standard atlas — or an error listing what's missing, never an invalid one.
  3. Bring it back here — paste the JSON, drop the file, or host it and paste the link.
or choose a file · drop one below · or paste a URL
Drag a standard atlas .json here · max 2 MB

Host your atlas at the root of a public Hugging Face dataset or GitHub repo and paste its link to share a read with anyone.

Measurement instrument for how a brand is perceived — not a verdict on what it "is." Cohort metameric variance is the measurement; there is no presumed ground truth. Code open-source (MIT); theory CC BY 4.0. No brand assets reproduced (nominative use).