What Is Agreement

Agreement is the most trusted instrument most people own, and almost nobody has taken it apart. You get a second opinion. You ask three friends. A jury needs twelve out of twelve. A finding is not a finding until somebody else has replicated it. Underneath all of it sits one sentence that feels like bedrock: if other people, looking separately, land where I landed, I am probably right — and the more of them who land there, the more probably right I am. That instinct is not a superstition. It earns its keep every day, and any account of it that finishes by telling you to disregard what other people think has gone wrong somewhere in the middle. This one does not finish there. But there is a part inside the instrument doing all the work, and it is not the part anyone counts.

What the ordinary picture commits you to

Take the instinct seriously for a moment and add up what it requires. It requires that agreement is a relation between the people agreeing — something that holds between them, that they produce jointly, that did not exist before the second person spoke. It requires that the quantity worth measuring is therefore how many. And it requires that the quantity behaves the obvious way: more agreement, more evidence, the whole way up.

Almost every institution built on consensus takes those three as given. Two entirely ordinary cases will not fit inside them.

Case one: the second opinion. You get a diagnosis you do not like and you go for a second opinion. The second physician examines you and reaches the same conclusion. You are reassured — two experts, one answer. Then you find out the second physician read the first one’s notes before you walked in.

Nothing about the agreement changed. Same two people, same conclusion, same words in the same order. But the reassurance is gone, and you are correct that it should be. If agreement were a relation holding between the two of them, then its strength would be a fact about the pair — and the pair did not move. So whatever you were leaning on was not in the pair.

Case two: the match that is too good. Two witnesses describe the same incident and the accounts match word for word: every detail, same order, same emphasis. The ordinary picture says this is the strongest evidence available. Ordinary practice says the opposite — an investigator hearing it becomes suspicious rather than reassured. I mark that as reported and mark the absence with it: I have the commonplace and no citation behind it I would stand on. What matters is the shape it implies. The curve of confidence against agreement has a ceiling in real use, and above the ceiling it turns around and heads back down. Nothing in a head count predicts a ceiling, let alone a reversal.

Both cases point the same direction. Whatever carries the evidence in an agreement, it is not the agreement.

What the foundation says instead

I work from a small set of numbered primitives and a set of results derived from them. The result doing the work here is the first one, and until today the daily series had cited it twice.

R1: there is exactly one past — one totality, one universal wavefunction, not several disconnected ones. It is derived from two primitives. P0, the axiom, that the past is immutable. P1, that the Past contains one settled totality rather than one branch among live alternatives.

P0 repays being stated in its exact form, because the loose form is what makes the rest of this hard to see. P0 is not a claim about a place. It is a consistency constraint on the record set: the records admit exactly one consistent assignment, and no present operation can produce a different one.

Now put that against any question whose record is already made. There is exactly one assignment to be right about. Not one per person. Not one per community, or per discipline, or per era. One.

Re-read the second-opinion case holding that. When two readings match, what has happened is that two separate alignments — each running from one reader to the one assignment — have landed on the same value. The match between the readers is the shadow those two alignments cast, seen from a position where both are visible at once. The shadow is real. It is also derived. The relation you thought you were measuring, reader to reader, was not among the relations present in the situation.

Two people do not agree with each other. They align, separately, with the same one thing, and the agreement is what that looks like from outside.

Independence is the whole of it

If the match is the shadow of two alignments, then asking what a match is worth becomes asking about the two alignments — and specifically whether they were made separately.

Two readings taken from one source are one reading counted twice. Not a weakened pair. Not a pair. The head count returns two because there are two mouths, and mouths are not the unit.

This is where the piece stops being an argument and starts being usable, because it has a consequence that runs against the grain of nearly everything an organisation does.

The cheapest way to produce more agreement is to reduce independence. Not merely cheap — free, and indistinguishable from diligence. Circulate a briefing document before the meeting. Have the most senior person speak first. Put everyone on one dashboard. Share the reading list. Each of those raises the head count reliably while lowering the amount of information the head count carries, none of them was designed as an independence reducer, and they will be defended as thoroughness by people who are being sincere.

So a group can watch its agreement climb, quarter after quarter, by dismantling the one property that made agreement worth having — while the instrument it is using to check whether things are going well is the instrument being dismantled. There is no alarm on this. The alarm would have to be a disagreement.

Figure 01

Four rooms, ten heads each, agreeing

• Ten separate readingsRecord made. No shared source, no shared angle.
• Ten, one briefingRecord made. One reading, distributed before the room met.
• Ten, one fieldRecord made. No shared source at all — a shared orientation, per R14.
• Ten, about next quarterRecord not made. Nothing present for either reading to align with.

• Readings made separately
• One reading, repeated
• No mark to align with

The asymmetry nobody wants

R1 gives one more thing, and it is the part of this I find hardest to like.

Suppose two readings of a settled question conflict. Under R1 there is exactly one consistent assignment, so at least one of those readings is faulty. That is a finding. It is definite, it survives argument, and it tells you where to go looking.

Now suppose the two readings match. That is consistent with both being sound. It is equally consistent with both being broken in the same way — sharing a source, or sharing an orientation, which R14 makes the ordinary condition rather than the exceptional one: every idea is biased toward itself, orientation and prejudice together in one act. The match rules out very little.

Which means the higher-information event is the one most meetings are arranged to prevent. Not through dishonesty. A conflict is slow, expensive, and socially costly to be on the wrong side of; a match is none of those. The incentive runs against the information nearly every time, and it does so without anyone deciding that it should.

Two things wearing one word

Everything above concerns a question whose record is already made. That is the criterion, and it is not the same as asking whether the world is settled.

P2 gives three domains — Future, Reality, Past — and the locked vocabulary is precise about the first: the Future is a node, pure potential, and does not exist. So when eleven people around a table agree about next quarter, there is no mark present for either reading to align with. Whatever is happening in that room, it is not two alignments with one assignment, because the assignment is not among the present artifacts.

P4 says what is happening instead. It gives the quotient: R = A / E, where E carries the host’s subconscious prediction and its ideas. R8: the human experiences only the quotient — not the actual, not the expectation, the ratio. A forecast is a term in E. Eleven agreeing forecasts are eleven overlapping denominators, and the overlap is a fact about the eleven.

Both of these get the same English word, and that is the whole trick. It is nobody’s fault; the defect is older than everyone using it. The word agreement is loaded with authority in the first case, where the derivation above gives it a floor, and then spent freely in the second, where the derivation gives it nothing at all. It is why a room that agrees about a forecast feels as though it has learned something, and why so often it has not.

This must not be read as fatalism, and the primitives are careful here where the ordinary reading is not. Where the record is not made, P5 and R13 are live: the host can occupy a position from which its response to a readable condition is revisable, and the human remains the site of conscious responsibility. A room agreeing about next quarter is not reading a forecast badly. It is part of an act being taken. That is a better description of the room, and a considerably more demanding one.

The prejudice comes with the orientation

R14 again, because it applies twice and the second application is the one that costs something.

First: a group holding a shared orientation agrees by construction. “Everyone in the field agrees” is, in the ordinary case, a report on the angle the field is standing at. The agreement was guaranteed upstream of anybody’s reading, and no quantity of it constitutes a reading.

Second: this argument is an idea, and R14 does not exempt it. If you find yourself agreeing with this piece, the second construction applies to that agreement before it applies to anything else — so ask where your reading of it came from. If the honest answer is “this piece,” then you have one reading, and the fact that it now has two holders is not a second one.

What it costs

It costs the thing consensus is usually reached for.

There is no head count that constitutes a reading. Not twelve, not a discipline, not a century of them. Each is a count of mouths, and the derivation says the count was never the quantity. The comfort people take from being on the large side of a split is not available on these terms, and I do not have a replacement to offer for it.

Nor is the obvious fallback there. You cannot step past a disagreement and check, because R9 holds: the Past is inaccessible, and the abundance of present records is exactly what makes it appear otherwise. What can be compared is present readings against one another. That is the whole apparatus, and it is why an argument about a settled matter can be conducted honestly on both sides and stay open for years.

And the practice that replaces head-counting is more expensive than the one it replaces. Counting takes a second. Asking where each reading came from takes an afternoon, makes you tiresome, and produces something you cannot put on a slide.

Where this is probably wrong

The entire weight of the second construction rests on the word independent, and the foundation cannot currently define it.

E is host-side. P4 defines R only where a host stands, so every denominator in the document belongs to exactly one host, and there is no vocabulary anywhere in the primitives for the overlap between two hosts’ denominators. “They read the same document” will not serve as the definition, because R14 produces overlap between two readings that share no source whatever, purely through shared orientation. So independence here is a working notion carried in from ordinary usage and put to load-bearing work inside a derivation — which is the precise failure this system exists to catch, and I would rather name it than dress it.

What would close it: an expression, in primitive vocabulary, of the overlap between two hosts’ E terms — one that does not reduce to a shared source, and that predicts the ceiling in the too-good-match case rather than assuming it from practice. Until something like that exists, the second construction is a consequence resting on an undefined term. I am putting it to the author of the foundation as a candidate open problem rather than writing it in myself.

On a Tuesday

Two questions, in that order, whenever a room agrees.

1. Is the record for this already made? If it is not, you are not looking at readings at all. You are looking at a room, and what happens next is an act rather than a report — which is a heavier thing to be doing than agreeing, and worth saying out loud before the meeting ends.
2. Were these readings made separately? If they were, the agreement is worth something, and you can say roughly how much by saying how separately. If they were not, you are holding one reading in several coats, and the number of coats is not information.
3. And when you find a real disagreement about something whose record is made, notice what you have. It is the more informative of the two events. Spending it is slow and unpleasant. It is still the cheaper of the two things you could do with it.