She was the last statistician on Earth who could still hear music.
Not play it — that was a different talent, one the algorithms had mastered decades ago. Not analyze it — that was trivial, reducible to Fourier transforms and spectral decomposition. She could hear it, and she could hear something no one else had noticed: the distributions.
Every chord, she realized on a Tuesday, was a probability cloud.
Not metaphorically. The harmonics of three notes — root, third, fifth — cascaded up through the overtone series, each partial mapping back to one of twelve pitch classes with a weight that decayed as one over its harmonic number. Add them up. Normalize. Every major triad was a point on the eleven-simplex.
Her colleagues laughed. "You're putting distributions on notes? They're frequencies, not random variables."
"They're both," she said.
She computed the Fisher-Rao distance between every pair of triads. She expected noise. She expected some clustering, maybe a graph that looked vaguely musical. Instead, the twenty-four triangles of the Tonnetz — a diagram Euler had drawn three hundred years ago by ear alone — appeared in her nearest-neighbor analysis with perfect fidelity.
One hundred percent.
The eigenvalue spectrum showed degenerate pairs. A torus. The circle of fifths emerged as the angular ordering. Tritone-related chords sat at maximum distance, opposite poles on the information sphere.
And the ordering — Relative closer than Leading-tone closer than Parallel — told her something no music textbook had ever stated explicitly: the Fisher metric knew that the third was the soul of a chord. The interval that separated major from minor, bright from dark, the thing that made you cry at weddings — it was the pitch class that contributed most to the spectral signature's distinctiveness.
She tried five different decay models. Inverse. Inverse-square. Square-root. Eight harmonics. Thirty-two. Every time, the Tonnetz emerged, patient and whole, as if it had been waiting inside the math for someone to ask the right question.
Her paper was three pages. She submitted it to Journal of Music Theory and Annals of Statistics simultaneously, because she genuinely didn't know which field it belonged to.
The reviewer from music theory wrote: "This is information geometry, not music theory."
The reviewer from statistics wrote: "This is music theory, not statistics."
She framed both rejection letters. They were, she thought, the most beautiful confirmation she'd ever received. The bridge between two fields is always invisible from either shore.
She kept listening. The music kept being probability. The probability kept being beautiful.
In the evenings, when the distributions settled into silence and the Fisher metric relaxed to zero, she played piano. Badly, with wrong notes and uneven tempo. But she played the way Euler must have listened — with the full weight of knowing that every resolution was a geodesic, every modulation a coordinate change, and every chord a tiny universe of probability, vibrating at the edge of meaning.