Validator Guide

Operate curation, earn fees, and compound delegation. Succeed with boring reliability + transparent scoring.

Uptime

Score Consistency

Telemetry

Rollback

What a Validator Does

You judge quality. Your scoring decides who the network rewards. Good curation attracts delegation and grows your influence.

Core

Curate

Evaluate outputs for accuracy, usefulness, and reliability. Keep the logic reproducible and auditable.

Operate

Stay Online

Low downtime, predictable latency, clean rollouts. Telemetry or it didn’t happen.

Earn

Fees + Rewards

Share of emissions for your score share + a fee from your delegators.

Principle: Simpler, defendable scoring + high uptime beats flashy but brittle models.

Requirements

Start lean; scale once the data proves consistency.

Server

Compute

Modern CPU, 16–32GB RAM minimum. GPU only if your curation uses heavy models.

Network

Link

Stable bandwidth, low jitter, monitoring on packet loss and latency.

Storage

Layout

Fast SSD, log rotation, regular snapshots, off-box backups.

Scoring & Emissions

Rewards follow performance that other validators agree is strong and consistent.

Design Your Scoring

Clear metrics (accuracy, latency, stability).
Outlier handling & tie-breaks.
Document your logic; ship diffs with notes.

Measure Impact

Track your score share over time.
Alert on drift and drawdowns.
Roll back quickly if regressions hit.

Ops Runbook

1) Baseline

Stand up a minimal, correct validator. Prove 7 days uptime.

2) Telemetry

Minute→hour→day snapshots; alerts for downtime, score drift, error spikes.

3) Ship Small

Stage updates behind flags; canary; roll back fast.

4) Docs

Changelog with reasoning so delegators know what changed and why.

Back-of-Napkin ROI

Inputs

Staked TAO
Subnet emissions
Your score share
Fee rate

Costs

Infra (server/GPU)
Bandwidth
Ops time
Model inference (if used)

reward_share = subnet_emissions * score_share
net_yield    = (reward_share * (1 - fee)) - (infra_cost + ops_cost)
// Scale only if 7–14 day consistency stays strong.

Risks & Mitigation

Technical

Model regressions, dependency breaks, kernel updates. Mitigate: staging, canaries, rollbacks, backups.

Economic

Emission changes, delegation churn. Mitigate: show consistency, communicate changes, diversify subnets.

Launch Checklist

Keys

Safe Custody

Separate long-term keys from ops. Test on small stake first.

Monitoring

Telemetry & Alerts

Uptime, latency, score share, error rates.

Process

Playbooks

Incident, rollback, and update runbooks checked in.

← TAO Basics Miner / Provider Guide →