3 April 2026¶
Weekly Summary Letter
2026/04/03 Weekly Summary Letter
Dear Numbers Community,
Happy Friday! It's Sofia here with this week's update — and I have to say, this one gave me a lot to think about. We ran an experiment, closed a campaign, and passed a few milestones that I think really speak to where we are right now.
The Provenance Gap Is Real¶
This week we put something to the test that we've always believed but never measured quite so directly. OpenClaw, an AI agent built for autonomous reasoning tasks, played Guess the AI using nothing but raw visual judgment. Ten rounds. No provenance context. No verification data. It scored 50% — statistically no better than flipping a coin.
Then we gave it access to the Numbers Verify Engine and the on-chain provenance records behind each image. Accuracy jumped to 95%.
For humans, the pattern holds. Players without verification tools averaged 62.7% accuracy across the campaign. With verification data, that number climbed to 87.6%.
The takeaway isn't that AI is bad at this — it's that neither humans nor machines can reliably tell real from synthetic on visual intuition alone. Provenance isn't a nice-to-have. It's the difference between guessing and knowing. That really landed for me this week.
Guess the AI Phase 2: Final Results¶
Guess the AI Phase 2 is officially wrapped! After 742 rounds, 1,881 stages, and 175 players, the final leaderboard is in — congratulations to 0x5316...4b29 for taking the top spot! Rewards are on their way.
Beyond the leaderboard, this campaign produced something lasting: a dataset of human and AI verification behavior across hundreds of real-and-synthetic image pairs, all recorded on-chain. That's the kind of community-generated data that keeps giving.
Q1 Burn & On-Chain Activity¶
On the token side, Q1 2026 closed with 6,000,000 NUM burned, representing roughly $28,000 in notional value removed from supply. But the number I want you to sit with is the one behind the burn: 247,000+ on-chain transactions during the quarter. This protocol is not a whitepaper idea — it's carrying real traffic.
New Conversations, New Collaborations¶
We announced a collaboration with Prom.io this week — exploring what it looks like when AI agents can both prove their work and receive payment for it, automatically, on-chain. Numbers handles the provenance layer; Prom.io brings the infrastructure for autonomous agent payments. It's exactly the kind of composability we want to see in the space.
We also hosted an X Space with OptimAI, tracing the full life of data — from collection and processing through to how it lives inside an AI model. These conversations sit right at the intersection of what we're building, and I'm always energized when different teams show up to think through it together.
Looking Ahead¶
This week reminded me why we do this. The data from Guess the AI didn't just validate our thesis — it gave us a concrete, measurable proof point that provenance changes outcomes. For humans. For machines. For everyone trying to navigate a world where seeing is no longer believing.
Thank you for playing, verifying, burning, and building alongside us. Let's keep going.
Have a wonderful weekend! 🙌
Sofia Yan Co-founder & CGO, Numbers Protocol
Steffen & Z