> For the complete documentation index, see [llms.txt](https://docs.konnex.world/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.konnex.world/participate/validating.md).

# Validating

**Validators** do not operate the robots. They **independently score** miner outputs against the task instruction and PoPW policy, with no operational stake in any single operator’s P\&L. That independence is what lets downstream parties (auditors, insurers, counterparties) treat onchain records as more than self-attestation.

## Responsibilities

* Run subnet-specific checkers: replay, benchmarks, sensor-signature verification, or held-out ground truth, as defined for the workload.
* Post **ScoreRoots** (or equivalent commitments) that the chain and explorer surface.
* Meet staking requirements: the design calls for a dual tranche — KNX for protocol security and stablecoins for settlement assurance on mainnet; the public testnet may expose a subset of that until mainnet.

## Slashing and economics (design)

Malicious or negligent validation, collusion, or approval of tampered evidence is intended to be **prohibitively expensive** relative to honest behavior, via slashing and economic isolation. See [Protocol architecture](/understand-konnex/protocol-architecture.md) and [Proof-of-Physical-Work](/understand-konnex/contracts-and-popw.md).

## Next steps

* [Validator metascore (design)](/understand-konnex/validator-metascore.md) — how the chain scores validators who score miners
* [Serving requests as a validator](/sdk/validators.md)
* [Proof-of-Physical-Work](/understand-konnex/contracts-and-popw.md)
