Lactone Labs

Every whisky has a sound.

And two that taste alike sound alike. A character is a note, how much of it there is decides how hard that note is struck, and where it came from decides where it sits — so a lot of something hits you and a hint blooms in behind it, and the pitch still names exactly one character. Two bottles with the same reading sound identical, the transform is the same for every bottle, and nothing is tuned to sound nicer.

The instrument

Tap characters, set how much of each, hear the chord. It runs entirely on your phone. A lot of a character hits immediately; a hint blooms in behind it. Each origin sits in its own place across the stereo field, and the face below draws what you are hearing.

field

cask

still

shore

One row per character, lowest note at the bottom. A row ripples at its own pitch — finer the higher it sits.
A steady row is agreement. A row that swells and dies is a disagreement, pulsing at its size. A row that stops short is a character only one of the two has.

8 locked (identical), 7 beating (they differ — faster is further apart), and 0 standing alone (only one bottle has a reading).

The comparison

Two bottles played together: every character both know becomes two voices a few hertz apart, beating at exactly the difference between them. Agreement is silence. A character only one bottle has stands alone — so how much of a comparison the evidence actually covers is audible, with nothing added to say so. That is the same line the finder draws between a full match and a partial one, heard instead of read.

The shelf above is the demo venue’s — the same bottles served at the demo finder. Bottles with more confirmed characters come first, because a chord from twelve says more than one from three.

What it is for

A second rendering of the same numbers. The compass draws a reading as a shape; this sounds it as a chord — same vector, same transform, nothing added and nothing tuned. It exists to find out what a non-visual rendering carries that a visual one does not, and who it carries it to.

It is a research instrument. It is not part of DramFinder, it is not on a roadmap, and nothing here is offered for sale.

What we want to find out

Four questions, in the order they would decide anything. The first three share one experimental design and differ only in the stimulus, so they are one piece of work rather than four.

  1. Does the sound carry the distance? Two bottles beat at exactly their difference — that is arithmetic. Whether a listener can RANK pairs by similarity from sound alone, and whether that ranking tracks the RMS distance, is not. It needs no catalogue data: synthetic vectors are enough. It decides whether this is an instrument or a demonstration with an honest mechanism.
  2. Which comparison does each rendering win? A shape is good at “how much of everything” and poor at “are these two the same”. A chord is the inverse — beating is literally a difference detector, and sixteen simultaneous voices are hard to hold. The useful question is not which is better overall.
  3. Can a listener hear the origin, and the degree? Each origin now arrives with its own envelope, and each degree is a different note. If neither is distinguishable inside a chord of four voices, both are decoration implying information. The degree half is a second, independent line of evidence on a question the catalogue has already raised: 85% of every reading ever taken is a 2 or a 3.
  4. Who does a sixteen-point radar fail? This is a coverage question rather than a perception one, and it is the one with a decision attached.

Not on the list, and deliberately: whether people prefer it, and whether it lifts conversion. The first has no answerable form and the second cannot be measured at one live venue.

What we have found

Measured 5 October 2026, on the catalogue as it stood. These answer the half of each question that does not need a listener — whether the information is present in the signal. Whether a person extracts it is the other half, and it is open.

1 · The sound does carry the distance — if you listen to the RATE

Over 6,022 real catalogue pairs sharing at least three characters, the mean beat rate across a chord tracks the matcher’s own RMS distance at r = 0.942(Spearman 0.946). So the acoustic quantity a listener receives is very nearly a monotone function of the number the finder computes.

But counting is a much worse proxy than hearing. The number of beating voices tracks the same distance at only r = 0.613 (Spearman 0.657). Two bottles can disagree a little about many characters or a lot about few, and the count cannot tell those apart while the rate can.

That changes the listening experiment rather than merely informing it: ask for a judgement of how fast, never how many. 5.0% of pairs produce no beating at all — they agree exactly on every character both know — and the fastest beat any disagreement can produce is 7 Hz.

2 · ⚠️ A pitch did not identify what was sounding — found, and fixed

For one day the degree climbed the character’s own pentatonic run. The sixteen sit on consecutive steps of one scale, so a character at degree k+1 landed exactly where the next character sits at degree k:

BEFORE   64 (character, degree) pairs  ->  19 distinct pitches
         62 of 64 (96.9%) shared a pitch with another pair

         164.8 Hz = Cereal & Biscuit at 4
                  = Green & Grassy at 3
                  = Vanilla & Coconut at 2
                  = Baking Spice at 1

AFTER    16 characters  ->  16 distinct pitches, one each

The defect was ours and it was live for a day. The problem it was fixing was real — four steps of gain on a sine is about 12 dB, and against three other voices a 2 and a 4 cannot be told apart, so the degree was in the data and inaudible. Moving it to pitch made it audible and ambiguous, which is worse: a chord stopped saying what was in it.

And it could not be retuned. Sixty-four distinct values cannot be carried by pitch here at all — 64 pentatonic steps span about thirteen octaves, and the scale was chosen precisely so no two voices can ever be rough against each other, which is the property that makes any beating a listener hears the data rather than the tuning. Widening the spacing breaks the thing the instrument is for.

So the degree moved off pitch entirely. It was duration alone for an hour, and duration is a slow cue — a 1 and a 4 were identical at the moment of entry and diverged only when the 1 stopped, so a listener had to notice an absence. It now has an instant cue as well: a 4 hits and a 1 blooms.

degree      1          2          3          4

attack    300ms      117ms       46ms       18ms    instant
sustain   0.40s      0.70s      1.24s      2.20s    settles it
pitch     ———————  the character  ———————           untouched
pan       ———————  the origin  —————————            untouched

Four axes on one tone that cannot collide, because they are different properties of it. None of them adds a partial, so unlike a richer tone none can manufacture roughness, and unlike gain none is swamped by the voices around it. Loudness now carries nothing at all — it is equal-loudness trimming alone, so two degrees of one character are the same volume.

3 · The origin is where the voice sits, and it is said twice

field      cask       still      shore
pan       -0.75      -0.25      +0.25      +0.75     stereo position
register  110-123Hz  138-220Hz  247-440Hz  494-880Hz  pitch band

The second row is not a design decision — it is how the ladder was already built. The sixteen are laid out in origin order, so field holds the lowest pitches and shore the highest by construction. Stereo reinforces something the register already carries, which is why the two are spread the same way round: low and left through to high and right, rather than two cues pulling against each other.

The honest cost, stated: pan collapses on a mono phone speaker, and this is a thing people will open on a phone. A mono listener loses the reinforcement, not the origin. Whether the register alone is enough to name an origin is open, and it is part of question 3 rather than something settled here.

4 · The two we cannot answer from here

Which rendering wins which comparison, and who a sixteen-point radar fails, are questions about people. Nothing in the catalogue can stand in for asking them.

Where it goes next

In order, because each one decides whether the next is worth doing.

  1. Settle the ambiguity above. Nothing else is worth measuring on an instrument whose chord does not say what is in it. Measure the sustain candidate against the pitch one for degree recovery, then adopt or reject.
  2. The rate experiment. Synthetic vector pairs, no catalogue data, no venue: present two chords, ask which pair is more alike, compare the ranking to RMS. The 0.942 above says the information is there, so this measures the listener and nothing else. It is the one that decides whether this is an instrument.
  3. Origin and degree identification, on single voices first and then in chords — the same apparatus as 2 with a different stimulus.
  4. Then, and only then, the comparison against the radar, and the coverage question.

None of this is funded, scheduled or underway, and no listener has yet heard a chord under any controlled condition. Everything above the plan is arithmetic over signals and over the catalogue — real, dated, and not a substitute for asking a person.

Status

Built. A research instrument, not a product — it runs, the transform is published and pinned by test, and the questions above are open. The mapping from character to pitch is in the method.