What Is Taste

There is no arguing about taste. That is the position almost everyone holds, and it deserves to be stated at full strength before anything is done to it. Taste is preference. What you find beautiful is a fact about you, not about the thing. When people say a building has good proportions or a sentence has a good rhythm, they are reporting their own reaction and dressing it in the grammar of a property. And the reason educated people believe in good taste at all is that taste tracks status: the preferences of whoever had the money and the leisure became the standard, and everyone else learned to want what they wanted.

Much of that is defensible. It explains why taste varies by century and by country. It explains why fights about it so rarely resolve. It explains the ugly social fact that taste is used as a credential. Held carefully, it is not cynicism — it is a decent piece of deflationary philosophy, and I am not going to caricature it to make it easier to beat.

But it commits you to something specific, and the commitment is where it comes apart. If taste is preference, then taste is a report of what you like. And a report of what you like should be maximally satisfied by a thing you built entirely out of your own choices.

Two ordinary cases it cannot hold

Under the preference account this should be rare and strange. They chose every line. There was no committee, no client, no inherited standard forced on them. At each step they did the thing they preferred. A pure preference machine, given full authorship, should produce output it likes. Instead the near-universal report of anyone learning anything is a stable, articulate dissatisfaction with work they had complete control over.

You can try to rescue the account by saying they have internalised other people’s standards and are measuring against those. That is the strongest defence available, and it fails on the second case.

ReportedBoth cases are ordinary observation rather than measurement, and are marked as such in the foundation’s register scheme. They are not evidence. They are the phenomena a definition of taste has to account for, and the deflationary one does not.

Preferences have a distribution. They do not have a direction. Yet people who develop taste in a domain converge on the direction — including about their own past, which is why the eight-month drawing looks worse at four years and worse again at ten, to the same eyes, with no external judge involved. Internalised social standards would predict convergence within a tradition and divergence across traditions. What actually happens is divergence in what people make and convergence in what they can see. Something is being read that is not the reader.

Taste and preference both surface as I like this, I don’t like that. That shared surface is the entire reason the ordinary account survives.

Where the reading comes from

The foundation has the machinery already, in a primitive that has been sitting almost unused. P10, exemplarity: every condition has an exemplar — a complete and exact form against which any artifact can be measured. A condition, per P6, is a prerequisite in order for something to happen or exist; it does not cause anything and does not happen. An idea, per P7, is a condition that has been named, and naming does not create it — naming makes the relationship conscious.

One clarification before this is allowed to run, because it is exactly the point where a reader will import something the system has not claimed. The exemplar is not being asserted here to exist. In the controlled vocabulary exists means amplitude and belongs to the antinode, which is where marks are made and where we are. Whether conditions are abstract objects is an open problem in the foundation, item eleven, and it stays open in this piece. P10 gives a measuring relation. It does not hand over a metaphysics of forms, and I am not going to improvise one to make the argument feel grander.

What the measuring relation buys is this. An artifact — a mark, as distinct from an act — carries a complex value z = σ·e^(iφ). The magnitude σ is surprisal, how far the actual departed from the prediction. The phase φ is which condition the mark serves. And fidelity, the system’s only evaluative question by R11, is phase alignment: cos(φ_artifact − φ_idea). Not a score out of ten. An angle.

Figure 01

Fidelity is an angle, not a score

φ idea φ artifact Δφ

• φ of the conditionWhich condition the work is in relation to. Fixed by which idea has the actualizer, not by the maker and not by the mark.
• φ of the artifactWhich condition the mark, as made, actually serves. Read off the thing itself.
• Δφ — the shortfallThe angle between them. Approximation is a small Δφ with too little magnitude. Corruption is Δφ growing. At π the mark serves the antipode.

The definition

And its twin, which is what the ordinary account has been mistaking it for:

That is the whole trick, and it is why the deflationary account is so hard to dislodge. Both things surface in the same sentence. I don’t like it is the output of a preference and also the output of a fidelity reading, and from the inside the two feel similar enough that a lifetime of using one word for both is entirely forgivable.

Figure 02

One sentence, two mechanisms

Surface — what is said

“It’s not right.” A single verbal report, identical in both cases, delivered with the same confidence and the same inability to elaborate.

↓   what is running underneath   ↓

Substrate — two different operations

A preference is a term in the denominator: the host’s own expectation, which the host authored and can change. A fidelity reading is a comparison against an exemplar the host did not author and cannot edit. Only the second has a fact of the matter, and only the second improves in a direction.

• host-side · in E
• reads Δφ · not in E
• revisable by wanting
• not revisable by wanting
• varies, no direction
• converges on direction

Five things that follow

1. Taste has to outrun ability. Structurally.

   ConsequenceReading Δφ and reducing Δφ are different operations. The reading needs only the comparison. Reducing it needs an actualization, and R10 is unambiguous there: every artifact is a miss, because the denominator is not the host’s to command. So the gap between what can be perceived and what can be delivered is not a stage of development. It is the standing shape of the relation.

   Traces to R10, R11, Part III

2. Your disappointment is the instrument, not the verdict

   If you can see that your work fell short, the reading is functioning. Someone fully satisfied with what they made has not surpassed you; they have either stopped reading or stopped correcting. The second of those has a name in the locked vocabulary, and it is the only sin in the system: settling is ceasing to correct. Approximation — low fidelity under continuing correction — is not a fault at all.

   Traces to R10, R11, locked vocabulary

3. It was never yours

   By P8, where a person’s naming is itself conditioned by C, that person is not C’s origin. Ideas have people; people do not have ideas. So “my taste” is the same category error as “my idea,” and R12 says ownership is a category error while the manifestation remains accountable. What the phrase developing taste actually names is a condition becoming readable to you. You do not acquire an instrument. You come into a relation, and the relation is what does the measuring.

   Traces to P8, R12

4. The reading is accurate and prejudiced at once

   R14: every idea is biased toward itself — orientation and prejudice, always together. Taste inherits this exactly. A reading taken from a condition is accurate about fidelity to that condition and biased against its antipode by the identical act. Two people with excellent taste, in relation to different conditions, will disagree and both be reading correctly. This is not the subjectivity thesis coming back in through a side door: the reading has a fact of the matter. What has no further fact of the matter is which condition has you, and that is not itself settled by taste.

   Traces to R14, S3

5. Taste drifts, and the drift is invisible from inside

   ProvisionalCorruption in this system is phase drift — the artifact governed by the actualizer’s interest rather than the condition. Applied to the reading: what gets measured quietly stops being fidelity and starts being reception. The reason this is hard to catch is that a drifted reading still feels like a reading. There is still a perceived gap, still the conviction, still the inability to elaborate. Approximation and corruption are distinguished by direction of travel, and direction is not visible in one artifact. It is visible in a series. This is offered as provisional and expires at the next version of the foundation.

   Traces to Part III, corruption; the “selling out” trigger row

What it costs

This should be said without softening, because the softened version is a lie people can feel. Developing taste is acquiring a permanent, accurate perception of your own shortfall, with no matching increase in the ability to close it. R10 guarantees the shortfall survives every improvement you will ever make — a better artifact is a better miss. And the acquisition does not appear to run backwards; I have not found a report of anyone un-seeing what they learned to see.

The bargain is not that the gap closes. It is that the instrument which makes the shortfall visible is the same instrument that makes correction possible. Cut the reading and you cut the correcting with it.

That is the honest consolation and it is a small one. There is a lower-cost move available and it works: lower the reading. Care less, notice less, look at less good work, stop comparing. The dissatisfaction does subside. What you have done is settle, and the system has no gentler name for it.

Where this is unsettled

Two places, and neither is closed by anything above.

Open problem · The φ trajectory

The foundation’s third open problem is that the idea fixes φ and φ determines the readable idea — sound as a dynamical system, circular as a definition. This piece leans on precisely that circle when it says taste reads Δφ. To close it, someone would have to write φ as a trajectory: an evolution rule giving φ(t) from the host’s prior artifacts and its conditioning, without using the readable idea to define φ and φ to define the readable idea. Until that exists, “taste reads Δφ” is a description of the geometry and not a mechanism. It stays open.

Open problem · Are conditions abstract objects

Item eleven. P10’s exemplar is the part of this framework that most resembles a Platonic form, and the honest position is that the system has not said what kind of object it is and this piece has not either. The argument here needs only that an artifact can be measured against something the maker did not author. That is weaker than a form, and deliberately so.

Where I am probably wrong

The weakest joint is case two. Cross-tradition agreement about a bad knife cut may be agreement about a shared physical constraint — the meat separates at the joint or it does not — rather than agreement about a condition. If enough of the convergence evidence dissolves into physics and biomechanics, the argument for a reading distinct from preference gets much thinner, and I would want to hear that objection before any other. The fifth card is also the softest thing in the piece: I have no procedure for detecting drift in a series, only the claim that a series is where it would be detectable.

On an ordinary Tuesday

Three things you can do with this before lunch.

Read the disappointment as output, not as a verdict. When the thing you made falls short of what you can see, that is the instrument reporting. Under R10 the shortfall is structural, which means the second layer — being ashamed of having produced a miss — is measuring against a standard the system says nothing can meet. Take that layer off first. It is the one that does most of the damage and the least of the work.

Stop asking whether it is good. Ask which way it is pointing. Good is a score and fidelity is an angle. A single artifact cannot tell you whether you are approximating or drifting, because both feel like caring about the work. A run of five can. Line them up and ask what each successive one was actually serving.

Do not close the gap by lowering the reading. It is the cheapest available fix and it is the only move in the framework that is a fault. To be exact about what that does and does not cover: settling is the standing setting of the instrument, not a day off. Resting is not settling. Deciding that a particular thing is finished and shipping it is not settling either — that is R10 being accepted rather than argued with. Everything else — the miss, the slow correction, the years of falling short — is the ordinary condition of making anything.