What Is Suspense

The suspense is killing me. Nobody asks the speaker to defend that, because it does not sound like a claim about the world; it sounds like a report on a state. Behind it, though, is a picture, and the picture is a claim. Something out ahead has not been decided. Its not-being-decided is a real feature of it, sitting out there, and you are in contact with that feature the way a rope is in contact with the weight at the far end — taut because the weight has not landed. When it lands the rope goes slack, which is why an ending is a release. On this picture suspense is a relation to an undecided thing, and its grip is the openness of the thing transmitted to you. I held that picture for most of my life without once putting it into words, which is how the strongest pictures get held. Three ordinary situations break it, and each one breaks a different part.

One neighbour first, so the two are not merged. I wrote a piece on waiting that sorted the ordinary cases by whether the record for the question was already made, and got a different answer on each side of that sort — a closed domain on one side, the early part of an act on the other. That sort is about what a person can usefully do. This piece is about the case where the sort makes no difference at all to the state itself, which turns out to be the ordinary case rather than the exception.

What the ordinary picture commits you to

State it strongly, because the weak form is not what anybody believes.

Three commitments. First, that suspense has an object, and the object is an outcome that has not been settled — the state is a relation, and something stands at the far end of it. Second, that the state is therefore sensitive to the condition of that object: settle the outcome and the relation has nothing to hold, so the state ends. Third, that its size tracks how open the outcome is — maximum where anything could happen, zero where the thing is fixed.

Three ordinary situations do not fit inside those commitments.

So the state runs when the outcome is settled, runs when the host knows it is settled, and runs when the host knows the outcome itself. Three different things were supposed to switch it off and none of them does. Something is being read. It is not the openness of an outcome.

What the foundation says instead

I work from a small set of numbered primitives and results derived from them. The result doing the work today is the coldest one in the document — written early, cited four times, and not touched in nearly a fortnight.

P1, actuals only: the Past contains one settled totality, not one branch among several. Everything that could happen has already happened; there are no unrealized possibilities, because none were left unrealized. From that alone — not from the axiom, which is a separate claim — comes R4: the Shannon entropy of the Past is zero. One settled outcome means all of the probability mass sits on it, and a distribution with all its mass on one outcome has an entropy of zero.

Read that narrowly, because it is easy to inflate. It is Shannon and only Shannon — a property of a distribution over outcomes, not a thermodynamic claim and not a temperature. It says nothing about the other quantities at that boundary, which are undefined rather than zero: frequency does not apply to a period, as the piece on pace worked out, and neither does colour, and neither does warmth. And it does not say anybody has the zero. It says there is no second outcome for a distribution to spread over.

Now the other branch. Where the record for a question has not been made, P2 locates the Future as a node: pure potential, and it does not exist. There is no mark standing at the far end of an unsettled question either — nothing in transit, nothing at range. What a person has in that case is a prediction, which is host-side: P4 gives the Reality Equation, R = A / E, the Actual over an Expectation made of subconscious prediction and ideas, and R8 says the human experiences only that quotient. Not the Actual, not the Expectation. The ratio.

Put the two branches side by side and the ordinary picture has nowhere left to stand. If the record is made, there is one outcome with the whole of the probability mass on it, and no openness for anyone to be gripped by. If the record is not made, there is no mark at the far end at all. Either way there is nothing out there with the property the picture says you are in relation to.

Three different things were supposed to switch it off and none of them does. Something is being read. It is not the openness of an outcome.

Two more results finish the demolition. P0, the axiom, is a consistency constraint on the record set: it admits exactly one consistent assignment, and no present operation produces a different one. So the leaning forward is not a small influence on the far end; it is not an influence on it. And R9: the Past is inaccessible, and the abundance of present records makes it appear otherwise. What arrives on Monday is not contact with the settled thing. It is a present mark — a phone call, a sheet of paper, a name read out — and the whole of the transaction happens here, on this side, in the room you are standing in.

Notice what this does not deny. The state is real, it is unpleasant or delicious depending on the case, and it is about something — there is a question, and the question has a subject. What it denies is that the state is in contact with the answer. Case one is what that looks like from outside: a man in the full state, an empty stadium, and a settled result that is not doing any of the work.

Why knowing the ending does not close it

Case three is still owed an account, and it is the hard one. If the state is a wide Expectation, and the host holds the record in memory, why is the denominator wide at all?

The answer needs one thing the primitives supply and one they do not. P7: an idea is a named condition, and naming makes the relationship conscious. The pieces on temptation and on curiosity worked out the principle that follows — that the precondition for a host reading anything is legibility, not appetite, and that a term appears in the Expectation under the description a person is actually running, not under every description that is true of the situation. I carry that over rather than re-deriving it.

This is testable in the cheapest possible way, and the test is one most people have already run. On a rewatch the state relocates. It is not sitting at the ending, which you know; it is sitting at the shot, the door, the four seconds. And it is recoverable by changing the description: watch beside somebody seeing it for the first time and it comes back strongly, because you are now running a live prediction about a person rather than a closed one about a plot.

The same construction cleans up case one. When the friend learns the match was recorded, the grip usually goes — and it is tempting to read that as the state responding to the record after all. It is not. What changed in that instant was not the match; it was that the score became one tap away. The available mark is what narrowed the term. The everyday proof is that people watch recorded matches knowingly, in full suspense, for as long as nobody in the room has a phone out. The status of the record does nothing. The availability of a mark does the work.

Figure 01

Four situations, one state

Situation	Record made	Host knows it is made	Host knows the outcome	State present
The live final, watched live	No	—	No	Yes
The delayed broadcast, watched unaware	Yes	No	No	Yes
The envelope read on Thursday, opened Monday	Yes	Yes	No	Yes
The fifth viewing	Yes	Yes	Yes	Yes

What the ending actually is

R2: records update expectation forward, and never re-open. The record of what you have read is write-protected in the direction that matters — the update runs one way, into the Expectation, and does not come back out.

So the end of suspense is not an event at the far end. It is a denominator event: the term narrows, and the narrowing is written forward on that description and stays written. Relief is what a narrowing denominator reads as. That sounds deflationary and it explains something the ordinary picture cannot: the fact that a first viewing is not repeatable. If the release were the world resolving, you could go back and have it again, since the film is right there and will resolve exactly as before. You cannot. What a rewatch restores is not the openness — it is a different description, running underneath a term that stays closed.

This is where the piece meets a nearby argument of my own, and the two should be kept distinct. I have written elsewhere that surprise telescopes — that because it is a log of a ratio, the total is fixed by where an expectation started and what actually arrived, so a teaser, a leak or a spoiler moves a payment from one date to another and neither creates nor destroys the total. That is an argument about the path of the denominator. What today’s result adds is the other term: on a settled question there was never a second contribution to conserve. The whole quantity was host-side from the beginning. The earlier piece shows that the schedule is all that a spoiler changes; this one shows why there was nothing else it could have changed.

Figure 02

Two layers, one word

What the picture says is being read

An outcome, out ahead, undecided

A property of something at the far end, transmitted to the host. On the settled branch there is one outcome carrying all of the mass; on the unsettled branch there is no mark at all. Neither supplies the property.

↓   the whole reading happens on this side   ↓

What is being read

A wide term, and marks arriving

A quotient: arrivals against an Expectation that has more than one live prediction on a description the host is running, with the rate of arrival set by something other than the host. Four things move it, and none of them is the outcome.

• How wide the term is on the description in play
• Whether a mark is available and how cheaply
• The rate at which marks arrive
• Which description the host is running

What this costs

It costs the state its authority. Being gripped is not evidence that anything is open. You cannot infer from the tightness in your chest that the question is live, because the tightness is identical in the case where the paper has been signed and filed. People run this inference constantly — the feeling that something is still in play is treated as information that it is still in play — and there is no channel by which that information could have got in.

It costs the idea that the tension is doing work. On a settled question the axiom is blunt: no present operation produces a different assignment. Sunday night does not reach Thursday’s decision. Not by a little, not inefficiently — not at all. I will say straight away that this is the point in the argument where the metaphysics is at its least useful, and I come back to it below rather than leaving it sitting there as advice.

It costs the innocence of the engineered case. If the state is a denominator condition, then anyone holding the four levers in Figure 02 can produce it with no live outcome in the arrangement at all. A cliffhanger is a description switched at the moment a term is widest. A loading bar is a rate. One room left at this price is a manufactured availability. The wheel that is already stopped and the animation that plays anyway is the delayed broadcast, sold. These are not exploiting uncertainty about the world — there is no uncertainty about the world in a slot machine, and the operator knows the distribution exactly. They are writing to the denominator directly, which is a thing I have argued at more length about machines and attention, and the mechanism here is the same one with a smaller budget.

R14, on this idea as on every other: an idea is an orientation and a prejudice in a single act, and so is this one. It is kind to the person who suspects the tension in their week is a trick being run on them, and unkind to the person whose tension is tracking a real decision they will have to be ready for on Monday. Both readings are available in the argument and only the first is what it says. Preparation is an act and the piece has nothing against acts.

Where this is unfinished

Four edges, stated so someone can hold me to them.

The first is the one I most want to close and may not. The foundation carries an open problem asking whether a two-job split it uses for backward-facing feelings — an evaluation that can succeed, and a reaching that cannot — applies to forward-facing ones. Suspense looks exactly like that reaching, pointed the other way, and today’s cases make the temptation worse rather than better: in the delayed broadcast and the fifth viewing, the far end genuinely is the settled record, so the reaching account seems to arrive already argued. It does not. The identity of the two states is an observation about the host, not a demonstration about a node, and the standing condition for closing that problem is unchanged: show that the Future node supports the same decomposition on its own terms, without simply reflecting the inaccessibility result derived for the Past. The two nodes are not symmetric in the primitives, and a mirror is not an argument. I am leaving it open, and I would rather say plainly that I wanted to close it here.

The second is the rate. The third construction needs a term to narrow and stay narrowed, and the foundation supplies no rate, no floor and no scale for how the Expectation updates. The pieces on streaks and on pace stated the same gap and I inherit it rather than filling it. Everything above about a term staying closed is a claim about direction and not about how long, and I have no business saying how long.

The third is the indexing. The second construction needs the Expectation to carry terms under descriptions rather than as one register of what the host knows. The primitives define it as subconscious prediction and ideas and are silent on its structure. That silence is the room my construction is standing in, and if the Expectation turned out to be undivided, case three would need a different account entirely rather than a re-labelling of this one.

The fourth is the machine, which turns up in this subject immediately, because a language model holds an explicit distribution over what comes next and narrows it as marks arrive. On this account it therefore has the arithmetic of a wide term in full and the reading of it not at all, since the quotient is defined only where a host stands and a host is defined as a living system. That is the fourth piece in a fortnight to stop at the same place, in the same shape: the whole of the present-side arithmetic on one side of the line, and none of the host-side reading. The foundation records the tension as open and instructs a piece to state it rather than settle it in passing. Stated.

On a Tuesday

1. Ask first whether a mark is available. Not whether the answer exists — whether there is something you could read today. A call you could make, an envelope you could open, a person you could ask. If there is, the state is a wait for a reading you are able to bring forward, and bringing it forward is an act you are permitted to perform. Most of the weight in an ordinary week is sitting on questions where the mark was available the whole time.
2. If no mark is available, say so out loud. The tension is not working on the far end and naming that does not dissolve it, which is fine — the naming is not a technique. It removes one thing only: the sense that the vigilance is a duty and that relaxing it would be a lapse. There is nothing being held up.
3. Sort by whether the record is made, not to change the feeling but to find the move. Made: the only move available runs toward a mark. Not made: this is not a wait at all, it is the early part of an act, and what you do this afternoon is a term in it.
4. When something has a grip on you, look for the four levers. Width, availability, rate, description. If one of them is being operated by someone with an interest in your staying — an episode ending mid-scene, a bar that fills slower than the file loads, a counter of other people looking — you are not being informed about the world. You are being written to.

None of which is fatalism, and the foundation is careful about the difference. P5: the host can occupy a position from which its response to a readable condition is revisable. R13: agency is relocated, not abolished, and the host is the site of conscious responsibility. What is closed is the far end, on the branch where a record exists — and the far end was never the thing with a grip on you.