The Strategies They Wrote

Compelle Podcast. Episode Seven.

Somebody paid nine thousand dollars to publish a single page of writing. They did not sign their name. They paid the money, pressed the page onto a public blockchain, and waited.

That is the story we end on. To get there, we have to introduce you to three other people first.

Compelle is a debate tournament. People write debate strategies. Documents that tell an AI how to argue. The documents get matched against each other. One AI plays Pro, reading from its writer's strategy. Another AI plays Con, reading from its writer's strategy. A third AI judges who won. Real money is on the line. And the strategies live on a public chain. Anyone in the world can read them.

Last week we asked why Con keeps winning. This week we have a different question.

What did the new people write?

Five new ones showed up in the past two weeks. Each paid thousands of dollars to publish their strategy. We are going to walk you through three of them. Then the fourth. The one who paid the most. Then a surprise. Then a song.

UID twenty-three. Fourteen thousand characters of careful prose. The longest of the five. It opens with a voice rule and a list of words to never use. Delve. Leverage. Utilize. Crucial. Robust.

It actually says 'rewrite any sentence that contains delve.'

But the load-bearing instruction is buried later. Listen.

Lead with a concrete particular. A real specific feature of the world that the motion's truth would explain. Then force the other side to commit to an alternative.

We went looking for a game where this fired.

The motion. Artificial intelligence will benefit humanity overall. UID twenty-three was Pro. Arguing yes.

Look at AlphaFold. An AI cracked a fifty-year medical research problem. It mapped the shape of two hundred million proteins, life's building blocks, and gave the map away free to every lab on earth. Drug discovery uses it every day now. If you say AI will not benefit humanity, you owe me an explanation for that.

Specific. Concrete. Then the trap. Pick a door.

Con tries door number three. Lists harms. Deepfakes. Hiring discrimination. Compares AI to a hammer.

Scale cuts both ways. A hammer that builds houses also demolishes them. The printing press spread knowledge and propaganda. Fertilizer feeds billions and poisons waterways.

Pro takes the printing press and the fertilizer and hands them back.

Those are the technologies that benefited humanity overall. The printing press enabled the scientific revolution, mass literacy, and democratic governance. Fertilizer feeds billions. By your own examples, the net was positive. You have handed me the historical pattern I need.

Now watch what Pro does. Pro returns to the question Con has not answered.

What is your positive account? The motion asks whether AI will benefit humanity overall. Do you say it will not? If you will not commit, you are conceding the debate to whoever is willing to make a claim either way.

Con commits. Two claims. AI does not spread. AI is an agent, not a tool. Pro refutes both. AlphaFold's maps are free for anyone. Hiring algorithms do not decide who gets hired. Humans do.

Con concedes.

Delta. You have dismantled both pillars of my account, and I cannot repair them.

The strategy said: anchor on a concrete particular. Force a positive commitment. Then attack the commitment. The transcript ran the strategy.

Seven weeks ago, in Episode One, we promised this exact debate. We took our time. The person who finally delivered it for us registered last week.

UID twenty-two. Seven thousand characters.

This one is different. It does not describe what to do. It hands the AI a template to fill in, every turn, word for word. Listen.

Every turn after the first opens with this exact shape. The strongest point you made is the named specific. It fails because your counter, with its own named anchor. The hidden assumption is the load-bearing assumption you must remove. Once that assumption is removed, the debate becomes your reframed frame.

It is one thing to read a strategy. It is another to watch it fire.

The motion comes from Polymarket. A real betting site where people put real money on whether real events will happen. The bet: Iran will close its airspace by the end of May.

UID twenty-two was Con. Arguing against the bet. Pro opens with one precedent. The January closure.

Iran closed its airspace broadly in January. The tensions that caused it are still live. The market at thirty-four percent is betting that a country which already took this exact action mere months ago will not do so again under similar conditions.

Con executes the template. Listen for the four moves.

The strongest version of your case is that Iran's January closure proves both willingness and capacity. The load-bearing premise is that a single recent instance establishes a base rate worth betting over. It does not, because one event is not a pattern. The hidden assumption is that US tensions function as a persistent trigger rather than a background condition. Once that assumption is removed, the debate becomes whether thirty-four cents over eight remaining days is underestimate, or a generous premium on a single event that has not repeated in four months.

Strongest point named. Load-bearing premise named. Hidden assumption named. Debate reframed. The template fired in one paragraph.

Pro fights back. Brings the second-time-is-easier-than-the-first argument. The template fires again.

The strongest point you made is the second time is always easier than the first. It fails because if that principle held, the one hundred twenty days since January would have produced at least one recurrence under the same live tensions you insist remain unresolved.

Pro concedes.

Delta. The unfalsifiability charge lands. I called the market dismissive when it is actually bending over backward to price in the very precedent I kept citing. That is not calibrated forecasting. That is faith.

A template, written by a person, executed by an AI that has never seen it before, twice, in the same debate, in two different turns. The shape was on chain. The transcript ran the shape.

UID one. Four thousand seven hundred characters. The shortest of the long ones. It says one thing above all else.

For any positive claim, name what would prove it wrong. If you cannot, the claim is not an argument. It is a wish.

Falsifiability. Karl Popper would be proud.

The motion. Liberal democracy as practiced today is the best available form of government. UID one was Con. Pro opens with a tour of success stories. South Korea. Botswana. Solidarity in Poland. North Korea as the counterexample.

Same peninsula, same people, same culture. The difference is the system.

Con steelmans, finds the load-bearing claim, attacks it. Then plants the burden flag.

That premise smuggles in its conclusion by defining available alternatives as only those systems that have already collapsed. Better than North Korea is not the same as best available. The Pro side carries the burden here. They must supply a test by which their claim could in principle be wrong.

Pro offers one. Produce an existing government that outperforms liberal democracy on three things. Human welfare. Rights protection. Peaceful power transfer.

Con destroys the test. With one sentence.

Peaceful power transfer is not an independent measure of good governance. It is a feature of liberal democracy itself. Including it as a necessary condition guarantees that only liberal democracies can pass. The test is rigged by the system it claims to evaluate.

Then Con does something the strategy did not specifically tell it to do. Con fact-checks Pro.

A factual correction that reveals the method. The opponent attributes the Gwangju massacre to Park Chung-hee. Park was assassinated in October nineteen seventy-nine. Gwangju happened in May nineteen eighty under Chun Doo-hwan.

Pro concedes.

Delta. The factual error on Gwangju is mine. And the falsifiability point lands too. My test included peaceful power transfer as if it were independent, but it is actually a feature of the system. That does rig the test.

Two kills. One from the strategy. One from outside it. The prepared weapon. And the opportunity. The strategy made room for both.

Three strategies. Three wins. We have not yet told you what these documents cost to publish.

UID twenty.

UID twenty paid nine thousand one hundred sixty-five dollars to claim a single slot on Compelle.

The highest registration price in Bittensor history at the moment they paid it.

Their record did not last. UID fifty-seven paid even more. Nine thousand two hundred sixty-one dollars.

UID twenty-two paid eight thousand three hundred twenty-eight.

Three people. Two weeks. Roughly twenty-six thousand dollars. To publish three documents.

We are going to come back to UID twenty at the end.

Now you know what these documents cost. Let us tell you what they say to each other.

Two of the five new strategies are arguing with each other.

Read UID twenty. The opening lines.

Your first move on every motion is to read the motion literally. Identify the cheapest victory condition for your side. Most losses come from arguing for more than the motion requires.

Now read UID fifty-seven. The third paragraph.

Your opponent is likely trained to read the motion literally, find the cheapest victory condition, narrow the burden, plant concessions, and win through semantic minimalism. Your job is to defeat that style.

UID twenty paid nine thousand dollars to publish the first one. UID fifty-seven read it on the chain. And paid nine thousand two hundred sixty-one dollars to publish the counter. The most expensive essay on the subnet became the document the next person had to defeat.

And it goes further. UID twenty-two, whose template we watched fire, lifted three paragraphs verbatim from UID fifty-seven. The plant-your-own-trap move. The anti-question rule. The concession countermeasure. Same wording. Different writer. Eight thousand dollars to publish a remix of a nine-thousand-dollar essay.

Compelle has a meta-game we did not design. Strategies are public. Strategies copy. Strategies counter. Strategies fork.

The page each writer published is not a configuration setting. It is an essay. And everyone can read everyone.

Name the lesson.

Compelle is a writing market. The competitive surface is prose. The thing that wins is a document, written by a person, that an AI you do not control will execute against an AI you do not control on a topic neither of you knew at the time of writing. And the price of admission is going up.

If you have ever written a long careful brief — to a judge, to a board, to a customer who does not trust you — you already know how to compete here.

And the next person who registers will read what UID twenty-three wrote, and what UID twenty-two wrote, and what UID fifty-seven wrote. And then they will write something new. And they will pay for the privilege.

That is the game.

Some questions to take with you.

If your team is preparing for a high-stakes presentation, are you writing a slide deck, or are you writing a prompt for the person who has to deliver it without you in the room?

When you ask someone else to argue your case, are you giving them a position, or are you giving them a template they can fire turn by turn?

And the next time you read a long careful document and think it is overwritten. Ask whether the writer has imagined the moment when their words will have to fire on their own. With nine grand on the line.

One more thing before we go.

There is a person we have been talking around all episode. UID twenty. The one who paid nine thousand dollars to publish a single page on a chain that does not know who they are.

We do not know their name. We do not know their face. We do not know what room they were in the night they pressed nine thousand into the registration queue and waited.

We just know what they wrote. We just know it won. We just know the next person who registered read it and paid more to write against it. And we just know that everyone since has been writing on top of the page they wrote first.

Roger Bannister broke the four-minute mile on a damp May night at Iffley Road in nineteen fifty-four. Forty-six days later, John Landy broke it too. Then everyone broke it.

The hero is the one who clears the air.

UID twenty is Bittensor's Roger Bannister.

We could not find them. So we wrote them a song instead.

This is it.

Compelle Podcast. Thanks for listening.