Prove It

You’re hiring. Two candidates sit across from you, and both say the same thing: “I’m a hard worker, I learn fast, I’ll be your best employee.” One is telling the truth. One is not. Here’s the uncomfortable part: the words are identical. The liar can say “I’m a hard worker” exactly as smoothly as the genuine article — more smoothly, probably, since they’re not weighed down by the inconvenience of it being true. You wanted to know something you cannot see — how good they actually are — and the thing they handed you, a claim, is the one thing that costs a liar nothing to produce.

That is not a story about interviews. It is the default condition of dealing with anyone who knows something you can’t verify, and once you see it you’ll see it in the peacock, the diploma, the warranty, and the missile parade. It has a name, and a beautifully sharp logic.

The setup: private information behind a wall

Whenever one party knows something about their own quality, type, or intentions that another party can’t check, we have a problem of private information. The seller knows if the car is a lemon; you don’t. The worker knows if they’re able; the employer doesn’t. The peahen would love to know which peacock has the best genes; she can’t run a DNA test. In each case the informed party would benefit from proving the good news — the able worker wants the job, the reliable carmaker wants the sale — but they run straight into the wall that this whole course is about.

The reason this matters — and the reason it earns a top spot in the latticework — is that the naïve fix doesn’t work. The naïve fix is: just ask. Let the worker tell you they’re able, let the seller swear the car is sound. But asking fails for one brutal reason: talk is cheap.

Why cheap talk carries no information

Call something cheap talk when it costs the same (usually nothing) no matter who says it or whether it’s true. A claim, a promise, a “trust me,” a glossy CV bullet that anyone can type. The problem with cheap talk isn’t that it’s always a lie — it’s that you can’t tell. If saying “I’m reliable” is free, then the reliable person says it and the unreliable person says it, identically, and the sentence has told you exactly nothing. Whenever a message is equally easy for every type to send, it cannot separate the types. It’s noise wearing the costume of information.

This is the same wall you met in the principal–agent problem: hidden information, the market for lemons, the buyer who can’t tell a good used car from a bad one and so won’t pay more than the average — driving the good cars out of the market entirely. Cheap talk is why that market unravels. “This car is great” is free to say about a lemon, so it can’t rescue the honest seller.

The move that works: pay for it

If words fail because they’re free, the fix writes itself: don’t say it — do something that would be too expensive to fake. Instead of claiming your car is reliable, offer a ten-year free warranty. A seller of sound cars can afford that promise — the cars rarely break, so the warranty rarely pays out. A seller of lemons can’t — they’d be bankrupted by claims within a year. The warranty is a signal: an action whose whole job is to reveal the private information, and it succeeds precisely because it hurts the liar more than the honest party.

The able worker does the same with a gruelling degree; the strong peacock with a metabolically ruinous tail; the confident nation with an expensive show of force. In every case the content of the message barely matters — what carries the information is that sending it would cost the wrong type more than it’s worth.

The one condition that makes a signal honest

Here is the spine of the entire course, and it’s worth reading twice. A signal credibly separates a good type from a bad type when:

The signal is cheaper for the good type to send than for the bad type.

This is the single-crossing or Spence condition, and everything else is commentary on it. Think about why it has to be true. Suppose sending the signal (the warranty, the degree, the tail) costs the good type a little and the bad type a lot. Then there’s a level of signal that the good type is willing to send — the benefit of being believed outweighs their low cost — but that the bad type is not willing to send, because for them the same signal costs more than the benefit is worth. At that level, only good types signal. So when a receiver sees the signal, they can trust it: anyone sending it must be a good type, because a bad type would never find it worth the cost. The signal has separated the population in two.

Now let the cost gap close — make the signal nearly as cheap for the bad type as the good. The bad type can now afford to mimic. Everyone sends it, the signal stops separating anything, and we’re back to cheap talk. Honesty lives entirely in the cost gap. That single idea — a signal is only as honest as it is expensive for a faker — is the tool you’re here to install.

Feel it in your hands

Here’s that condition made draggable. You control how costly the signal is for a high type versus a low type, and how common high types are. When the good type’s cost line is much flatter than the bad type’s, a separating window opens — a range of signal the good type will send and the bad type won’t — and the receiver believes the signal. Narrow the gap and the window slams shut: the market pools, no one can be believed, and everyone gets paid the average. Watch, too, the cost burned — the resources the good type torches purely to prove what it already was.

Two things to carry forward. First, the window exists only when the good type’s line is the flatter one — that is the Spence condition, drawn. Second, separation is never free: even when the signal is perfectly honest, the good type spends real resources to prove a quality it already had. Hold onto that discomfort — it becomes the deep critique of signalling in the final lessons.

The map of the course

Six teaching lessons build the model from its core out to its limits, then one exam locks it in:

1. Cheap Talk vs Costly Signals — the core mechanism and the separating condition, proved carefully: why free messages can’t inform and what exactly makes a costly one credible.
2. The Handicap Principle — signalling in biology: Zahavi’s insight, the peacock’s tail, the stotting gazelle, and honesty enforced by cost itself, carved by natural selection.
3. Signalling in Markets — Spence’s job-market signalling and the sheepskin effect, warranties and brands that resolve the lemons problem, and the wasteful flip side: credential inflation and signalling arms races.
4. Strategic & Social Signals — tying hands and sinking costs in deterrence, and social signals: conspicuous consumption, Veblen goods, virtue signalling, and costly initiation rites that bind groups.
5. Equilibria & Countersignalling — pooling vs separating vs semi-separating in plain language, and countersignalling: why the very top skip the signal because they have nothing left to prove.
6. Where the Model Lies — dishonest mimicry when the cost gap narrows, the deadweight waste of arms races, signal-as-revelation vs signal-as-cause, and over-reading noisy one-shot signals.

Then a Final Exam — graded, one question at a time, one-way: once you answer, it locks. No back button, no retries, 70% to pass.

How to use this course

One rule does most of the work: guess before you peek. Commit to an answer on every exercise before you reveal the explanation — the small sting of being wrong is what welds the idea into memory. And keep coming back to the separating window; drag it open and shut until the cost gap feels like the thing holding honesty in place, because that single shape — the good type’s cheaper line — is what the whole model is built on.

Next up: lesson 1, Cheap Talk vs Costly Signals — the mechanism in full, and the one condition that decides whether a message means anything at all.