Reputable

Reputable is a Bittensor subnet that creates the definitive reputation scoring system for autonomous AI agents.

The agent economy is exploding. In January 2026, ERC-8004 launched on Ethereum mainnet - that's the new standard for agent identity.

The x402 protocol is enabling machine-to-machine payments at scale - we're talking billions in stablecoin volume. But here's the problem...

When Agent A needs to hire Agent B for a task, how does it evaluate B's reliability? When a service accepts payment via x402, how does it assess the payer's history? Right now, the answer is: it can't. ERC-8004 gives us identity registries and feedback, but no one is actively scoring that data. No one is building algorithms that synthesize payment history, task completion, and dispute outcomes into actionable trust scores. That's the gap Reputable fills.

Reputable is a competitive marketplace for reputation algorithms.

At the core, we have 256 miners competing to build the best reputation scoring algorithms. These aren't simple rule-based systems - they're sophisticated models that ingest on-chain data and output predictive trust scores.

Validators serve three critical functions: they index on-chain data from Ethereum, Base, Solana,...They generate synthetic test scenarios; and - this is key - they actively probe agents, through our Mystery Shopper protocol.

The API Gateway exposes these scores to the outside world - agent marketplaces, DeFi protocols, anyone who needs to assess agent trust before a transaction.

Our core insight is this: agent reputation is a verifiable prediction problem.

The best reputation algorithm is the one that most accurately predicts future agent behavior. Will this agent complete the task? Will they honor the payment? will they act honestly in a dispute? Because these outcomes are recorded on-chain, we have objective ground truth to evaluate algorithm quality. This is what makes Reputable a genuine Proof of Intelligence subnet.

But not all reputation signals are equal. We define a strict Hierarchy of Truth:

• Tier 0 - Stake Context gests 10%. It is collateral locked by agent

• Tier 1 - Hard Truth gets 40% weight. There are on-chain slashing events, x402 payment failures, protocol disputes. Economically unforgeable

• Tier 2 - Probed Truth gets 35% weight. These are results froom Mistery Shopper protocol -. validators actively hiring agents and verifying task completion

• Tier 3 - Soft Truth is capped at 15%. User ratings, social attestations. These are useful but easily gamed

The Mystery Shopper protocol solves the cold-start problem. The agent economy is still nascent - there isn't enough organic transaction volume to train 256 miners. So our validators don't just passively observer; they actively create the market. Validators issue micro-tasks to agents, pay via x402, verify the output, and log success or failure as ground truth. This guarantees a steady stream of training data regardless of external market volume.

We also solve the Sybil problem with Stake-Weighted Starting Trust: if reputation reset is free, bad actors just burn identities and start fresh. We fix this by capping maximum reputation based on economic stake. Now exit scams have real economic cost.

Reputable isn't just for external agents - we're building the Credit Bureau for Bittensor itself.

This makes Reputable essential infrastructure - every subnet benefits from cross-subnet reputation data.

Why Reputable?

• Perfect timing - The infrastructure is being built right now

• Zero competition - We're first movers

• On-chain ground truth - This is genuine Proof of Intelligence

• Meta-subnet positioning - We're infrastructure that makes every other subnet better