Changing Your Mind Midway Can Be Evidence of Attention, Not Unreliability

She agreed to it. An hour later she does not want to any more, and there is no dramatic reason — nothing was said, nobody did anything wrong, the thing simply stopped being what she wanted.

And the word waiting for her is unreliable. She changed her mind. She went back on it. She could not make up her mind. Every available phrase treats what happened as a failure of character rather than as a report about the world.

The companion essay that opened this month argued that a settled boundary frees the attention that vigilance was consuming, which makes quiet things noticeable. This one takes the next question. If that attention is now working, it will produce readings — and a reading that changes is exactly what the culture has no room for.

The month’s private sentence is: my yes feels more like mine when no remains available. Remains is the operative word, and this essay is about the enormous amount of cultural machinery arranged to make sure that it does not.

Start with where the vocabulary comes from, because the vocabulary is doing the work.

Almost everything available for talking about agreement in this domain is borrowed from contract. She agreed. She committed. She went back on it. A contract is the culture’s standard instrument for one person telling another what will happen, and its defining virtue — the property that makes it worth having at all — is that it binds across time. An agreement that expired the moment it was made would be useless.

So under a contract model, a mid-course reversal is not an inconvenience. It is a breach. The person who does it has failed to do the one thing an agreement is for, and unreliable is the precisely correct term for somebody who does not honour commitments.

Which means the accusation is not irrational or malicious. It follows validly from the model, and everybody involved is reasoning correctly. The question worth asking is whether the model applies.

And the function of a contract makes that question answerable. A contract exists so that the other party can plan — so that somebody can commit resources, make arrangements, and act on the assumption that the agreed thing will occur. That is the whole point, and it is exactly the right instrument for building a house or delivering goods.

It is not what is happening in an intimate encounter. Nothing is being built, no third party is depending on delivery, and the only thing the agreement determines is what two people do in the next hour. The instrument has been borrowed from a domain where it works into one where its defining property is a defect.

There is a better model available and it is entirely ordinary — it is simply not the one the vocabulary was built for.

Consent in this setting is a live instrument reading. The yes given at eight o’clock was a report about the state of things at eight o’clock, and it was accurate. At nine the accurate report may be different.

And this is the crucial point: the later reading does not falsify the earlier one. Nothing about it says the first was a lie, a mistake, or an error of judgement. It says the thing being reported on has changed, which is what things do, and which is precisely what an instrument exists to detect.

So a changed reading is not a broken promise. It is a functioning instrument. A thermometer that showed eighteen degrees this morning and twenty-two now has not contradicted itself, and nobody would describe it as unreliable — unreliable would be a thermometer that kept showing eighteen because that was what it said the first time.

That analogy is worth pressing slightly, because it locates the error exactly. Under the contract model, the reliable person is the one whose answer does not change. Under the reading model, the reliable person is the one whose answer tracks the state. Those are opposite specifications, and the culture has installed the first while wanting the second.

The model persists because it is genuinely load-bearing, and an essay that treated it as a simple mistake would be missing what it protects.

A contract model guards against a real problem: people denying afterwards that they agreed to something. Where an agreement can be retroactively unmade, anybody can be accused of having proceeded without permission, and there is no available evidence because the evidence was a state of mind. Fixing consent at a point in time and treating it as binding is how that is prevented.

That protection matters and this essay is not arguing against it. But notice what the model is optimised for. It is built for adjudication after the fact — for settling a dispute about what was agreed, at a later date, in front of somebody who was not present.

Accuracy during is a different design goal entirely. A system built to be auditable afterwards will be bad at tracking a state as it changes, because auditability requires fixity and tracking requires revision. Those requirements are in direct tension, and no system does both well.

So the honest description is that a tool built for one important job has been pressed into a second job it is badly shaped for, and the second job is the one that occupies almost all of the actual hours. The disputes are rare. The live state is continuous.

Here is the consequence that matters most, and it operates long before anybody changes their mind.

If a yes binds, then giving one is a commitment to a future she cannot see. She is agreeing not only to the thing but to still wanting the thing at a time she cannot inspect from here — and nobody can forecast that reliably, about anything.

So the rational response is to withhold. Faced with something she would probably enjoy, the sensible move under a binding model is to decline, because a probable enjoyment is not worth an unconditional commitment. The expected value calculation comes out against saying yes to most things that are merely likely to be good.

Which means the contract model does not merely punish reversal. It suppresses the first step. It makes every initial yes expensive, and the predictable result is fewer of them — a person who agrees to less, tries less, and stays within the region she is certain about.

And then the culture reads the resulting caution as a fact about her. She is not very adventurous. She does not seem to want much. She is hard to please. None of which is a reading of her desires; all of which is a reading of the price of a yes in a system where yeses cannot be withdrawn.

This chamber has published a finding with the same shape in a different domain: that refusing to ask for things protects a weak measurement at the cost of the thing itself. The mechanism here is not the same but the family is — a defensive adaptation to a badly designed rule, mistaken by everybody including the person making it for a fact about what she wants.

There is an asymmetry in who the reliability question is ever asked about, and stating it plainly is most of the argument.

When somebody stops midway, her reliability is immediately in question. When somebody continues, nobody asks anything at all. Continuation is invisible; it reads as consistency and requires no account.

But consider what continuation actually demonstrates. It is compatible with two entirely different states. The first is that she still wants this. The second is that she stopped checking — that the reading was taken at eight and never taken again, and she is proceeding on a stale value.

From outside, those two produce identical behaviour. There is no observable difference between a person who is still consenting and a person who has stopped consulting herself, because both of them simply carry on.

Which gives the sharpest version of the problem. Consistent and not paying attention look exactly the same, and only one of them is what anybody actually wants. The culture has built a reliability test that cannot distinguish them, and that test therefore rewards whichever is easier to produce — which is, unambiguously, the second.

And the incentive it creates runs the wrong way throughout. A woman who keeps checking will sometimes find a changed reading and will have to act on it, which will cost her the label. A woman who stops checking will never have to, and will be described as easy to be with. The system is selecting against attention.

The epistemics are worth stating directly, because once stated they are difficult to defend.

A yes at the beginning is a prediction. She is forecasting how something will feel before it has happened, on the basis of an expectation, from outside the experience.

A stop in the middle is an observation. She is reporting on how it does feel, from inside the experience, with the actual thing in front of her.

In every other domain, an observation outranks a prediction about the same quantity. The forecast is superseded by the measurement the moment the measurement exists, and nobody regards this as the forecast having been betrayed. That is simply what forecasts are for: to be replaced by data.

Here the ordering is reversed. The prediction is treated as binding and the observation is treated as a problem requiring explanation. The person holding better information is the one asked to justify herself, to the person holding worse information, in the moment she acquired the better information.

This chamber has published two adjacent findings that complete the family. One concerned a prediction that cannot be falsified, because acting on it prevents the evidence that would revise it. The other concerned a promise that cannot be verified, because no finite number of non-repetitions confirms it. This is a third case: a prediction that is actively preferred over the observation that would correct it, which is the most straightforwardly inverted of the three.

A cultural consequence follows from all of this and it is worth naming, because it costs more than the awkwardness of individual evenings.

Anything coded as a failure does not get practised. Since reversal is filed under unreliability, nobody rehearses it, nobody develops a way of doing it that is not excruciating, and no conventional phrasing emerges for it. Almost every other socially necessary act has a standard form — declining an invitation, ending a conversation, leaving a party — and this one does not.

So when it is genuinely needed, the act has to be improvised under pressure by somebody who has never done it. Which makes it slow, badly worded and late, and the lateness is the part that matters: an improvised reversal happens further into the situation than a practised one would, because the improvising takes time.

And the absence compounds socially. Because nobody has a standard form for stopping, every stop is conspicuous and every stop is read as significant — which raises the cost, which further reduces the practice, which keeps the form from developing. A conventional phrase for something becomes ordinary only by being used, and this one cannot get started.

None of that is a claim about anybody’s capability, and nothing here suggests that anybody who found it difficult was doing something wrong. It is an observation about what a culture equips people to do, and this is a thing it conspicuously does not equip anybody to do.

Which raises the practical question of what would make reversal affordable, and the answer relocates the problem entirely.

Stopping costs almost nothing in itself. The expense is in what happens immediately afterwards, and that is determined by the other person rather than by her.

A reversal received as information costs nothing. A reversal received as rejection costs a great deal — and the difference between those two is produced in about three seconds by somebody who is not the one who stopped.

And the specific failure is asking why. Why, what is wrong, was it something I did. Those questions are almost always asked kindly and with real concern, and they convert a reading into an accusation. They require her to produce a justification for a state, which is not a thing states have, and they make her responsible for managing the other person’s disappointment at precisely the moment she has none spare.

The alternative costs nothing and is a single word. Fine. Understood. All right. No question, no visible adjustment, no concern expressed, no return to the subject later in the evening. Anything more elaborate is a request for reassurance dressed as attentiveness.

Which puts the whole problem where it can actually be fixed. Whether a person is able to change her mind midway is not a fact about her courage or her clarity. It is a fact about what the other person will do in the three seconds afterwards — and that is a policy, decidable in advance, by somebody who is not under any pressure at the time.

An argument this permissive about reversal has to say where it stops, or it becomes a case that nothing anybody agrees to means anything.

This chamber has already published the relevant line. A betrayal is the unilateral alteration of terms the other person is still relying on — which locates the wrong in changing what somebody is planning around, without telling them.

That distinction does the work here without needing to be re-argued. Agreements about the world bind, because other people arrange themselves around them: money, dates that others have organised their week around, obligations that involve third parties, anything somebody else has spent something on the strength of. Those are commitments in the contract sense and they should be treated as such.

Reports about a state do not bind, because a state cannot be promised. Whether she wants something at nine o’clock is not a thing she was ever in a position to guarantee, and an agreement that purported to guarantee it was agreeing to something outside anybody’s power.

And there is a real cost worth conceding on the other side. Repeated reversal at no notice does impose genuine costs on other people — time arranged, things prepared, an evening that cannot be reused. That is a fair complaint and it belongs to the first category rather than the second: it concerns the arrangement, not the state, and it is answered by giving notice where notice is possible rather than by continuing after the reading changed.

The line is clean enough to apply. What did the other person arrange their week around, and has that been altered without telling them? If yes, it is the kind of thing that binds. If the only thing that changed is what she wants in the next twenty minutes, nothing has been breached, because nothing about the next twenty minutes was ever within her power to promise.

This argument arrives at this business with more force than most, because the structure here contains the problem in its purest available form.

An arrangement here is paid in advance. Which means that when somebody stops midway, the commercial grammar immediately supplies the wrong frame: a customer who is not consuming what she purchased, an evening not completed, a transaction that has gone unfulfilled. None of that vocabulary exists in a domestic setting, and all of it is available here by default.

And the deeper problem is not the vocabulary. It is that our interest and her accuracy point in opposite directions at exactly the moment the reading changes. Everybody here has a reason to prefer that the evening continues. That is not a character flaw of anybody working in this trade; it is the shape of the incentive, and it exists whether or not anybody is acting on it.

Which is why this cannot be handled by good intentions, and it would be dishonest to offer them as the answer. The only thing that works is procedure: a stop costs nothing, ends the matter, and is not followed by anything. No question about why, no expression of concern, no attempt to establish whether something could be adjusted, no returning to it later, and no discussion of the money at any point — because raising the money converts a reading into a negotiation, which is the exact failure this essay has described, with a price attached.

And the hardest of the obligations, which is about the money specifically. Nothing is refunded and nothing is pro-rated, because a refund conversation makes the stop a financial event and gives the person stopping a reason to weigh her reading against a cost. The amount was for the evening being available, not for its contents, and it is not reopened. Anything else would put a price on an instrument reading, which is the one thing that must stay free.

The honest disclosure is that she should assume the incentive exists rather than trusting that it does not. We are the party that benefits from continuation in every case, without exception, and no assurance from us changes that. The only sound basis for proceeding here is a procedure that would work even if the incentive were being acted on — which is the standard this house should be held to, and the one a reader is entitled to check.

This essay is cultural and structural analysis rather than clinical guidance, counselling, legal advice or advice about consent in any particular situation. It describes no treatment, promises no recovery, diagnoses nobody, and does not tell any reader what to agree to or what to conclude about anybody.

The claim that consent in this domain is modelled on contract, and that a contract’s defining virtue is binding across time so that a mid-course reversal registers as breach, is this Library’s own formulation. It is reasoned rather than measured, and the essay states explicitly that the accusation follows validly from the model rather than being irrational or malicious.

The alternative model — consent as a live instrument reading, where a later reading does not falsify an earlier one because the thing reported on has changed — is this Library’s formulation. The thermometer comparison is an analogy rather than evidence.

The account of why the contract model persists concedes that it guards against a real problem, namely agreements being retroactively denied, and states that this essay does not argue against that protection. The observation that auditability and live tracking are in direct tension is reasoned rather than measured.

The claim that a binding yes is an expensive yes, so the model suppresses initial agreement rather than merely punishing reversal, is this Library’s argument. It notes the family resemblance to this chamber’s earlier finding about refusing to ask, while stating that the mechanism is not the same.

The asymmetry claim — that only the person who stops is assessed, and that consistency and having stopped checking produce identical behaviour from outside — is this Library’s formulation. It makes no claim about how often either occurs.

The epistemic inversion, that a starting yes is a prediction while a mid-course stop is an observation, and that the culture trusts the first and interrogates the second, is reasoned rather than measured. It is explicitly presented as a third member of a family alongside this chamber’s published findings about an unfalsifiable prediction and an unverifiable promise.

The claim that reversal is never practised because it is coded as failure, so no conventional form develops and improvised reversals arrive later than practised ones would, is an interpretation. It explicitly makes no claim about anybody’s capability and does not suggest that difficulty indicates fault.

The relocation of the cost onto the reception, and the identification of asking why as the specific failure, is this Library’s argument. It states that such questions are almost always asked kindly and with real concern, and the suggested alternative is derived from the essay’s reasoning rather than from evidence about outcomes.

The line between what binds and what does not applies a definition this chamber has already published rather than re-arguing it, and concedes that repeated reversal at no notice imposes genuine costs, which are located on the arrangement rather than on the state.

The section about this house states that its commercial structure supplies the wrong frame by default, that its interest and the reader’s accuracy point in opposite directions at the moment a reading changes, and that this cannot be handled by good intentions. Its obligations are stated as procedure rather than as assurance, including that nothing is refunded or pro-rated so that no price attaches to stopping, and the section states plainly that a reader should assume the incentive exists rather than trusting that it does not. No figure, proportion or statistic appears anywhere in this essay.

My yes feels more like mine when no remains available.

The word doing the work is remains, and this essay has been about everything arranged to ensure that it does not. A no available at the start and unavailable at nine o’clock is not a no that remains; it is a no with an expiry, and an expiring no makes the yes it authorised into something closer to a signature than a sentiment.

And the practical consequence is that the sentence is not primarily about her. Whether a no remains available is determined almost entirely by what somebody else does immediately after it is used — which means the person who most determines whether her yes feels like hers is the person receiving it.

That is an unusual conclusion for an essay in this chamber, which more often locates the movable part in the reader. Here the movable part is elsewhere. What she can do is small: notice the reading, say the short sentence, and decline to supply a reason, since a reason is not owed for a state and offering one teaches everybody that one is required.

What the other person can do is almost the whole thing, and it costs nothing. One word, no question, no adjustment, no return to it. Done once, it makes the next reading affordable. Done consistently, it produces the only condition under which a yes can mean what everybody wants it to mean — that it was checked, recently, by somebody who was genuinely free to say otherwise and knew it.