Deception is bought

Competitor Advanced

The short version

• A lie that costs nothing is not believed, and should not be. Credibility is bought.
• So every deception has a ledger: what you gave up to tell it, and what it made them waste.
• The strongest fakes work by putting you in a genuinely weaker position, because nobody expects you to do that on purpose.
• Knowing when your opponent has stopped watching is itself a resource.
• You can probe an opponent's judgement the way a gazelle probes a lioness: send a small thing, judge the answer, and repeat only if the answer was wrong.

The situation

You fake a rush. You build the extra barracks, you make the aggressive-looking move at the front, and they do not buy it. You have spent a hundred and fifty minerals and thirty seconds of your attention to change nothing at all.

Or the other version. You were not faking anything. You did the normal thing, and they panicked and built four cannons, and you won at fourteen minutes off an expansion they let you take.

The difference between those two games is not cunning. It is whether the thing you did was expensive enough to be worth believing, and whether it pointed at something they actually feared.

Two kinds, and they are not the same job

The lesson 11 lecture puts two headings on the board and leaves them up all night.

Army deception moves their army to the wrong place. Draw the defensive army out, or lure the offensive army somewhere that costs it thirty seconds. The currency is position and time, and the payoff is usually a window rather than a kill.

Build order deception makes them build the wrong thing. The currency is minerals, and the payoff is a straight transfer: everything they spent answering a threat that was never coming is gone, and everything you spent instead is still on the map.

They fail differently, too. A failed army fake costs you a walk. A failed build fake costs you the build.

The best lies make you genuinely weaker

Here is the idea worth the whole lecture.

A Zerg puts lurkers on hold position in front of his ramp rather than up it. That is a worse position. It is worse in a way that anybody can see. Which is exactly why it works: no one would put himself in a weaker position on purpose, so the Terran reads the board as the true board, concludes the main force must be up the ramp where it belongs, and walks his marines and medics forward over the front lurkers.

The lie was not a trick. The lie was paying to be in a bad position, and then collecting.

You can run the same logic backwards to find your own fakes. Ask what you would never do, work out what it would cost you to do it anyway, and then ask what your opponent would conclude from seeing it. If the cost is small and the conclusion is large, you have found a good one.

A second thing that game showed

The Terran in that game never scanned and never saw his army die, because he had already changed screen to macro. He was satisfied. He knew how the fight ended.

Professionals stop watching a fight once the outcome is determined, because attention is the scarce resource (T1) and a decided fight is a bad place to spend it. That habit is correct and it is also exploitable, and it generalises: there are moments in every game when your opponent is provably not looking at a thing. Just after a fight he has won. Just after his own attack lands. The second he starts a big production cycle.

Knowing where someone is not looking is worth as much as knowing what they have.

Score the ledger

The lecture judged every replay the same way, and that method is the part you can take away. Two columns: what the deceiver gave up, and what the lie bought.

Two things about that table. The fourth row is the cheapest lie in it, and it is cheap for the same reason the first row is: the cost was a building that the real plan wanted anyway. The best deceptions are the ones where the bill is mostly paid by something you were buying regardless.

And the lecturer undercut his own examples, repeatedly and on purpose. The base relocation lost the game. The minefield trick was answered by an opponent who simply rebuilt his army in two minutes and won later. A model that only shows you its wins is a sales pitch, and he did not give one.

The game theory underneath

Halfway through, the lecture stopped and did the actual subject the course was named after.

A signalling game needs three things. The signal has to mean something to the receiver, or it is just noise. Each player has to be maximising his own outcome, which a one on one game gives you for free. And, the one that does the work: the signal has to be costly. If signals were free you would send all of them, and the receiver would learn to ignore every one.

The biology is where this is clearest, and it is where the lecture went. A peacock's tail makes it easier to catch. Birdsong tells a predator where you are. A nestling begging loudly is advertising itself to everything in the wood. All of them are expensive, and all of them are believed because they are expensive.

And the one the lecture spelled out letter by letter: stotting. A gazelle that sees a lioness jumps straight up, wasting energy, saying I am fast, pick someone else. Both animals gain. The gazelle does not have to run, the lioness does not have to waste a chase. It is a call and an answer, and an inadequate answer gets punished.

The working

Probing, and why it is the practical half

The last original thing in that lecture was the lecturer's own, and he flagged it as his own: an idea he called proxying, deliberately reusing a word that already meant something else, for affecting the will of your opponent's army by proxy.

The mechanic is the gazelle's, run in the other direction. You send a small force. Not to kill anything. To make him answer, and then to judge the answer.

• He does nothing. Wrong answer. Do it again, and again, until he stops being wrong.
• He brings everything home. Wrong answer, in the other direction, and now his hundred supply is in his base while you take the expansion.
• He brings exactly enough and a little more. Correct. Stop. Defending is cheaper than attacking, he has paid the smallest bill available, and he still has to weather the real army later.

The cost is real and worth naming. The probing force usually dies. It demands hard attention, because you have to watch the response and do damage at the same time, which is two things at once (T10 prices that). And the lecturer was honest that he could not find many clean professional examples of it, which is a good sign about him and a warning about the idea.

His own examples were a Protoss mirror where one zealot was answered with one zealot instead of two, and the prober escalated into the mineral line and won shortly after; a Zerg dropping four hydralisks on the same Protoss over and over because the answer never improved; and a shuttle flying a dark templar toward a Terran without ever dropping it, purely to see whether the Terran could unsiege and reposition. If he could not, the rest of the game was free.

What the lecture said

The two kinds of deception go on the board at ►2:00 and stay there. The hold position lurkers run from ►5:00, the build order case at ►14:00, the long tech switch analysis from ►19:00, and the base relocation at ►31:00.

Signalling proper, including the gazelles, runs from ►50:00, and probing from ►58:00.

The games in that lecture are shown off a projector screen and the player names are hard to read, so this page describes each game by what happened rather than by who played it. The lesson is in the move, not the name.

A drill to go with it. Take one of your own losses where you were surprised. Find the moment you believed something that was not true. Then ask two questions: what did your opponent pay to make you believe it, and what would one cheap look have cost you? Most surprises turn out to be cheap lies you could have checked for a hundred minerals.

Still true today