What Is a Reputation

My reputation. The possessive does the work so quietly that it is worth saying the sentence out loud before examining it, because almost nobody hears anything odd in it. A reputation is the thing you spend a career building. It is an asset, and the language around it is the language of property throughout: you have one, you protect it, you can trade on it, it can be damaged, ruined, stolen, restored. Law agrees — defamation is actionable as an injury, which means a court will treat the harm as done to something that was yours. Credit bureaus formalise a slice of it and sell the slice. Employers check it. Parents tell children it is the only thing they really own, and they are not being sentimental; they are describing what they have observed, which is that what people believe about you determines a great deal of what happens to you, that the belief accumulates slowly, and that it is very hard to change.

I do not think that picture is foolish. It tracks real regularities, and any account that leaves those regularities unexplained is worse than the picture it replaces. But it is wrong about what a reputation is, and it is wrong in a way that four ordinary cases show before any foundation is consulted.

What the ordinary picture commits you to

Say that a reputation is a thing you have. Four consequences follow, and none of them holds.

There is a fifth case, and I leave it out of the list because it is a fact about language rather than about reputations. The stock phrases are “your reputation precedes you” and “it travels ahead of you.” Quoted here as ordinary speech and not used: heard once, each is a harmless figure. Heard as a description, it is the error in miniature: it puts the thing in motion, coming into a room from somewhere else, when nothing about it went anywhere. I will not use that phrasing again in this piece, and by the end of it the reason should be structural rather than fussy.

The step in the foundation that does the work

The result I need is R6, and not at the clause it is usually read at. R6 says that the Past cannot perceive, cannot recognise, and cannot stand in relation, and it gets there by a chain: immutable, therefore cannot change; cannot change, therefore cannot move; cannot move, therefore cannot perceive; cannot perceive, therefore cannot relate. The load in that chain is carried by one premise, which is that perception entails a change of state. Nothing perceives without something in it having altered.

That premise is readable in two directions, and the direction that matters here is the second one. Read toward the node, it yields R6 as stated. Read toward the living side, it says something plainer and more useful: wherever there is perceiving, there is a thing that changed. I worked that reading a week ago on the word witness, and stopped at perception, which is where the word witness stops. R6 names a second capacity in the same breath, and it has not been worked: recognition.

Recognition entails a change of state for exactly the reason perception does. To recognise is to be in a state, after, that you were not in before. So when someone in a doorway knows who you are, the change is in them. Not partly in them and partly in you — in them. You may be entirely unaware of it, asleep, or dead. The event that the English word recognition names has one site, and you are not it.

Being known is not a state you are in. It is a change that happened somewhere else, to somebody else, and the ordinary word for it puts you at the centre of an event you were not present at.

That gets the single recognition. It does not yet get a reputation, because a reputation persists between recognitions — it is there in the two years when nobody thinks of you at all, and it is there for people who have never laid eyes on you. Two more things are needed, and once they are in place the whole subject comes apart cleanly.

Two components, and they must not be merged

Take the locked vocabulary literally. There is an act, which is a process, and there is an artifact, which is a mark. History is the artifacts — present, surrounding, and revisable. That is the first component: the things said about you, written down, filed, told at dinners, screenshotted, remembered. Those are marks, and they are here, in the only domain where anything exists.

The second component comes from the equation. R = A / E, and E comprises subconscious prediction on the real side and ideas on the imaginary side. A recognition alters a host’s state, and what the alteration amounts to, in the vocabulary, is a change to that host’s E. What a reputation installs is chiefly a prediction: a term on the real side of somebody’s denominator, standing for how you will behave when it next matters. This is the same primitive the series has leaned on twenty-seven times, taken at its decomposition clause rather than at the whole equation, and taken on the real side, which is the side the recent pieces have not worked.

Figure 01

Where the two components are, and where you are not

• Component one · the marksWhat was said, written, filed, remembered.Artifacts. They are in the Eternal Now, on the shelf, surrounding everyone including you. They are add-only: a mark made in 2019 is not available for deletion by anybody, yourself included.Location: the present record. Not in you.
• Component two · the termsA prediction about you, in somebody’s denominator.One per host who has read the marks. On the real side of E. Revisable — this is the part of the whole arrangement that is designed to move, and it moves on the host’s reading rather than on your conduct.Location: other people. Not in you.

This is why the four cases resolve at once. You have never inspected it because inspection would require reaching a term in someone else’s denominator, and no operation of that kind is available to anyone. There is no fact about which of the four hundred is real because there are four hundred terms and the question was asking after a single object. It moves when you do nothing because a term updates on its host’s reading, and a promotion across the hall is new material for that reading while your conduct that week supplied none. And it goes on after you because the marks are on the shelf and the terms are in the living, and neither of those two locations was ever your body.

The law already says this, in its own definition

Worth noting because it is the strongest outside confirmation I found, and it comes from the institution with the most practical experience of the problem. The Legal Information Institute’s entry defines reputation as the general esteem or regard in which a person or entity is held within the community — which locates it in the community and not in the person, exactly as above. Its note on damages is sharper still: a plaintiff recovers for harm that diminishes the esteem in which they are held by others. And the instrument the law uses to read a reputation, under the federal rule on character evidence, is reputation testimony — a witness describing what they understand a person’s standing to be. That is a host’s report of a host’s own reading, put on the record as the best available evidence, which is precisely the instrument Case 01 said was the only one there is.

So the substantive definition agrees with the derivation while the surrounding procedure treats the thing as an injurable possession, and both have been sitting in the same paragraph for a long time without the two being reconciled. I take that as corroboration rather than as grounds: the reasoning above stands on the foundation and would stand if no legal system had ever mentioned the word. Marked as report.

Why being misjudged feels like being lied about

Now the result that makes the experience intelligible, and it is the one that makes this subject uncomfortable rather than merely clarifying. The host reads only the quotient. Not the Actual, not the Expectation — R. A person who holds a term about you does not experience themselves as holding a prediction about you. They experience a fact about you. The denominator is not available to introspection as a denominator; it shows up as how the world is.

So when somebody has you badly wrong, and you try to correct it, you are not disagreeing with them about their opinion. From inside their position there is no opinion there to disagree with. You are telling a person that the world is not as it appears to them, which is why the conversation so reliably goes badly and why it feels, from your side, like being called a liar by someone who thinks they are simply describing the weather. Both of you are right about your own reading and the readings are of different things.

Add the bias result and the picture closes. Every idea is biased toward itself — orientation and prejudice arrive together, in the same act. A host reading the same marks from a different orientation produces a different term, and does so without any dishonesty at any point. Two people can hold incompatible terms about you, both having read the record carefully, both accurate about the condition that has them, and both prejudiced against its antipode by the same act of reading. There is no version of this where one of them is simply lying. That possibility exists too, but it is not needed to explain the disagreement, and reaching for it is usually the more comfortable mistake.

The permanence is on the other side

Write-protection has two clauses, and this subject turns on holding them apart. Records are add-only; and the update a record makes to expectation runs forward and does not re-open. Apply the two clauses to the two components and they come apart in a way the ordinary picture has exactly backwards.

The marks are permanent. What you said in 2019 is not available for deletion. No apology reaches it, no settlement unmakes it, no amount of later conduct edits it. A retraction is a new mark placed beside the old one, which is the only move anyone has ever had in any matter whatsoever. I wrote about that operation four days ago under the word undo, and the finding there was about the record and about what reversal cannot be. Here the record is only half the subject, and the other half behaves oppositely.

Because the terms are the revisable part. A denominator is the thing in the whole arrangement that is built to move. It updates forward on new material, and it will go on updating for as long as its host is alive and reading. So:

Almost all reputational effort is aimed at the wrong half. People try to manage the record after the fact, which is the immutable half, and they try to argue directly at the reading, which is the half they have no access to. The one operation genuinely available — add a mark and let readers read it — is the one that feels least like doing something, because its effect is not yours to schedule.

The arithmetic of what one mark is worth

There is a consequence here that I did not expect when I started, and it is the part of this piece I would most want checked. An artifact carries a magnitude and a phase, and the magnitude is surprisal — how far the actual departed from the prediction. The size of the update a mark makes to a reader’s denominator is therefore not a property of the mark on its own. It is measured against the term that was already there.

Run that out. If a host’s term about you predicts decent conduct with high confidence, then decent conduct is nearly informationless when it arrives, and indecent conduct is very far from the prediction indeed. The consequence is a sharp asymmetry, and there is no term for anybody’s harshness in it:

The better you are thought of, the less any further good act can do for you, and the more any single bad act can do to you. Not because readers are cruel. Because surprisal is measured against the prediction that is already in place, and a good name is a high prediction.

The ordering is what the foundation gives. To see the scale of it, take the standard definition of surprisal and a stated prior. Let a reader’s term assign some probability to your behaving well on this occasion. Then the confirming case carries the negative natural log of that probability and the departing case carries the negative natural log of its complement, in nats. The arithmetic is one line and reproducible; the four prior levels below are my choices, made to span the range, and are not measured from anything. I write them as percentages deliberately, because the algebraic shorthand collides with a retired phrasing in the foundation’s own vocabulary and I would rather the notation be slightly longer than have a mechanical sweep fire on it every time.

Figure 02

What one act is worth, against four priors

• Behaving well — the confirming case
• Behaving badly — the departing case

• A poor nameprior 30%1.204
• A poor nameprior 30%0.357
• A mixed nameprior 70%0.357
• A mixed nameprior 70%1.204
• A strong nameprior 95%0.051
• A strong nameprior 95%2.996
• A spotless nameprior 99%0.010
• A spotless nameprior 99%4.605

One neighbour named, so the two are not merged

I published a piece a week ago, in a different series and on a different machine, about the premium attention pays for bad news — the measured finding that a loss weighs roughly twice an equivalent gain. This is not that, and the difference matters. That premium is a claim about valence, and it lives in the reader: the same magnitude is paid differently depending on the sign. What is above has no sign preference in it anywhere. Set the valence premium to zero, give the reader perfect symmetry between good and bad news, and this asymmetry survives untouched, because it comes from the level of the prior rather than from the direction of the departure. The two mechanisms are independent and they stack, which is why the lived effect is larger than either. Anyone reading both should hold them apart.

What I will not claim, and why

The obvious next sentence is that an old reputation is harder to move than a new one, because a long history of confirmation makes each further confirmation smaller. It reads well, and I am not going to write it, because it needs something the foundation does not have. The equation supplies no rate, no floor and no scale for the update to a denominator — nothing that says how much a term moves for a given surprisal, or how a term’s history changes its responsiveness. That absence is the most load-bearing gap in the whole document, and a piece that quietly assumes a learning rate to get a satisfying paragraph is doing exactly what this series exists to avoid. The level-dependence above needs no rate: it compares two acts against one prior at one moment. The age story needs a rate, so it stays unwritten.

What the primitives do not have

A reputation is, by construction, the case where a term about one person stands in many denominators at once. That is the whole subject, and the foundation has no expression for it. The primitives describe one host with one E. There is nothing in the document that says what it is for a term to stand in more than one of them, and therefore nothing that says what is and is not shared when it does.

This is the third piece in this series to run across that hole, and I am naming it rather than writing across it. The closure condition, stated specifically enough that someone could hold me to it: an expression in the primitives for a single term standing in more than one denominator, with an account of whether “the same reputation” names one term with many instances or many terms with a common source. The ordinary word requires the first — that is what makes it sayable that a reputation was damaged, in the singular. Everything the document actually supports gives the second. Until that is settled, every sentence in this piece about handing a reading over, or about a reputation being shared across a town, is running over ground the foundation does not describe, and I would rather say so at length than let the prose paper it.

What it costs

Four things follow that are not pleasant, and the piece is worth less if I soften them.

You cannot clear your name. Not in the sense those words are usually taken. There is no operation available to any person that reaches into another host’s denominator and sets a term. Courts can compel the removal of a mark and can add a very loud new one; neither of those is the reading, and the reading is what the phrase is about. What remains available is to add marks and let time pass in hosts you do not control.

You will never learn your reputation. This is worse than uncertainty, because uncertainty has an answer that you happen not to have. Here the question has no object. Every honest attempt to find out returns somebody’s report of their own reading, which is a new mark rather than the thing.

The unfairness is structural, and the people doing it are not doing anything wrong. Someone will hold a term about you fitted to marks you made when you were a different person, and that is what a denominator is: a prediction built from the record available. There is no misconduct to appeal against. The word for what you want from them is update, and updating is not a thing anyone can be asked for and handed.

And you are doing the same thing, right now, to a few hundred people. This is the reading of the essay that took me longest to write down. You hold terms about people fitted to marks that are years stale, you experience those terms as facts about them, and you are no more aware of it than the people holding terms about you. The position is symmetric, and the discomfort of being on the receiving end is the best available evidence of what you are doing at the other end.

On an ordinary Tuesday

Conditioned freedom is the position from which a response to a readable condition is revisable, and the living system is the site of conscious responsibility. Non-ownership of the reading does not relieve you of the making. Five things that follow:

1. Stop managing the reading and make the next mark. The marks are the only component you touch. Everything else in the arrangement is somebody else’s denominator or a record that is closed.
2. When someone has you wrong, argue with the material and not with their honesty. From inside their position it is not an opinion, so treating it as one reads as an accusation. Ask which marks they are working from. That is a question they can answer.
3. Expect nothing for the thousandth decent act. Not cynicism — arithmetic. Do it for the fidelity reading, which is available to you, rather than for the update, which is not. The evaluative question was never whether it worked.
4. If you want to know how you are read, ask for marks rather than for a verdict. What did I do that gave you that? A verdict is a quotient and cannot be inspected by the person giving it. A mark can be named, and the answer is frequently a specific thing from four years ago that you had forgotten.
5. Audit one term you are holding. Pick a person you have decided about and date the marks you are using. If they are old, you are the structural unfairness in somebody else’s essay, and this is the only item on the list you can act on unilaterally today.