# Conviction Curve *Earn crypto by training AI* By [Assetux](https://blog.assetux.com) · 2026-05-25 did, decentralizedidentities, web3, blockchain, ai --- Conviction Curves ================= ### Why We Let Markets Decide Which AI Signals Are Trustworthy Bots are getting smarter. Every year, the tricks that used to separate humans from fake accounts stop working — and the people building Sybil detection systems have to scramble to catch up. POH takes a different approach. Instead of one team deciding which verification checks are good, we let the market decide. Here's how it works. * * * Signals: the building blocks of human verification -------------------------------------------------- When an AI tries to figure out whether a wallet belongs to a real person, it runs a series of tests. We call these **signals**. A signal is a simple question you can ask about a wallet address. Things like: * "Has this wallet been active on Ethereum for more than a year?" * "Does this address hold any ENS names?" * "Has this Solana wallet ever interacted with a DAO?" Each signal gives a yes/no or a number. The AI weighs all the signals together and makes a verdict: **human** or **bot**. Anyone can register a new signal in POH. If you've built a clever way to detect human behavior on-chain, you can list it. It costs 1,000 POH tokens — mostly to prevent spam — and then your signal enters the pool and starts getting tested by the community. * * * The problem with just voting ---------------------------- When a new signal is listed, other POH holders vote on whether it's actually useful. Do they trust this method? Does it make sense? Does it have obvious flaws bots could exploit? Voting works, but it has a flaw: **votes are free**. If you vote that a signal is great and you're wrong — if that signal turns out to be useless or easily gamed — nothing happens to you. You just go on with your life. There's no consequence, so there's no strong reason to research deeply before voting. This is the classic problem with opinion-based governance. People vote with their gut, not their wallet. * * * Enter Conviction Curves: vote with your money --------------------------------------------- When a signal is listed in POH, the creator can also launch a **Conviction Curve** — a small trading market specifically for that signal. Think of it like a stock ticker for trust. * **Buying signal tokens** means: "I believe this signal is a reliable way to identify humans." * **Selling signal tokens** means: "I've lost confidence in this signal." The price of the token reflects how much the community believes in that signal _right now_. It updates in real time. It responds to new information. If someone discovers a way to game the signal, people sell, the price drops, and the AI sees a weaker signal. This is the key idea: **when you put money on something, you pay attention**. * * * What happens when a signal gets popular --------------------------------------- Each Conviction Curve starts as a small bonding curve — a simple price mechanism where the price goes up as more people buy and down when they sell. Early buyers get in cheap. If the signal proves itself, the token appreciates. When a signal raises **10 SOL** in its curve, something important happens: it **graduates**. Graduation means the signal's market moves to a real decentralized exchange (DEX). At that point, we know real money has validated this signal. People actually staked capital on it. The protocol rewards this: **graduated signals carry 1.5× more weight in AI decisions**. If the market has spoken with 10 SOL of conviction, the AI listens more carefully. * * * Why this makes the AI smarter over time --------------------------------------- Here's the flywheel: 1. Someone lists a genuinely useful signal. 2. Early believers buy its token cheap. 3. More people test the signal, it works well, more people buy. 4. The signal graduates → gets 1.5× weight in AI verdicts. 5. More accurate verdicts → more people trust POH → more checker requests. 6. More trading volume → signal creator earns 4% fees on every trade. 7. Creator is incentivized to keep maintaining the signal. Now the bad outcome: 1. Someone lists a weak signal — maybe it's easy to fake. 2. Early buyers notice it's being gamed. They sell. 3. Price drops, signal stays on the bonding curve. 4. The AI learns (through votes and market price) to downweight it. 5. Creator earns nothing because nobody is trading a signal nobody trusts. The market does the work that a committee of judges would otherwise have to do — and it does it faster, more honestly, and 24/7. * * * What does "launching a signal" cost? ------------------------------------ To launch a Conviction Curve alongside your signal, you pay **0.1 SOL** (roughly a few dollars at current prices) plus 1,000 POH tokens. Of that 0.1 SOL: * About 0.021 SOL goes to the Solana network for account storage (this is standard for any on-chain program) * The rest — roughly 0.079 SOL — is automatically used to **buy your own signal's token** That's right: you start out as a holder. You have skin in the game from day one. If your signal is bad, your own tokens lose value. If it's great, you benefit alongside everyone else who believed in it. * * * Sharing your signal ------------------- Every signal gets a **referral link**. Share it with your community and anyone who trades through your link generates a small fee for the referrer. This matters because it creates a natural distribution mechanism. Signal creators become ambassadors for their own signals. The better the signal, the more it gets shared. The more it gets shared, the more it trades. The more it trades, the more the creator earns — and the more the market can verify its quality. * * * What does this look like in practice? ------------------------------------- Imagine someone builds a signal that checks whether a wallet has participated in at least three Ethereum governance proposals in the last year. This is a genuinely hard thing for bots to fake — governance participation costs gas money and time. They list it on POH. Early DeFi governance power users buy in because they know how rare genuine governance participation is. Price climbs. More people discover it. After some trades, it hits 10 SOL — it graduates. Now, every time the POH AI checks a wallet, this governance signal carries extra weight. The verdict: "this wallet has the governance history of a real person" becomes _significantly more persuasive_ than it was before. The creator? They're earning 4% of every trade that happens on that signal's curve. Forever. As long as traders believe in it. * * * The bigger picture ------------------ Most Sybil detection tools are built by one team, maintained by one team, and trusted or rejected by users who have no way to inspect what's inside. It's a black box with a trust-me API. Conviction Curves flip this. Every signal is public. Every signal has a price. Every signal has a community of believers who have real money on the line. The AI isn't making decisions based on secret criteria — it's listening to an open market of informed participants, each of whom paid to have an opinion. That's not just more transparent. It's more accurate. Markets, when they work well, aggregate information faster and more honestly than any committee can. We think this is what AI-driven identity verification should look like: not a single system owned by a single company, but a living market of signals, priced by conviction, curated by community. * * * _Want to list a signal or start trading on an existing Conviction Curve? Head to the POH app and connect your Solana wallet. Devnet tokens are available via the in-app faucet._ --- *Originally published on [Assetux](https://blog.assetux.com/conviction-curve)*