What validators are, what they need, the 1→4→7 decentralisation path · content/validators-explained.md
Validators — what they are, what they do, and what we need
Date: 2026-09-15 · Founder's question: "Explain validators to me more and what is needed there?"
What is a validator?
A validator is a computer that agrees on the order of transactions. Think of it as a witness at a contract signing — their job is to watch every transaction happen and confirm "yes, I saw that, and it happened in this order."
You send 5 Qals to Bob
↓
┌─── VALIDATORS watch this happen ───┐
│ │
│ Validator 1: "Yes, Alice sent │
│ 5 Qals to Bob" │
│ Validator 2: "I agree, I saw it" │
│ Validator 3: "Confirmed" │
│ Validator 4: "Confirmed" │
└─────────────────────────────────────┘
↓
The transaction is FINAL — nobody can undo it
Without validators, anyone could claim "I didn't send that" or "I sent it twice." Validators are the referees that make the ledger trustworthy.
How many do we have?
| Now | Phase 2 | Phase 3 | Why it matters |
|---|---|---|---|
| 1 (superlocal) | 4 (superlocal + gmktec + cloud + bb-mini) | 7+ (add external operators) | More validators = harder to cheat |
Why 4 is the magic number (for now):
Our consensus (Starfish) uses BFT (Byzantine Fault Tolerance) which requires 3f + 1 validators to tolerate f faulty ones:
| Validators | Tolerates | Meaning |
|---|---|---|
| 1 | 0 faulty | If it goes down, the chain stops |
| 4 | 1 faulty | Chain survives 1 machine crashing or being hacked |
| 7 | 2 faulty | Chain survives 2 machines going down |
| 10 | 3 faulty | Even more resilient |
With 4 validators, we can lose any 1 machine and the chain keeps running. That's the minimum for a production-grade system.
What does a validator actually DO?
EVERY SECOND, each validator:
1. RECEIVES new transactions from users and other validators
2. CHECKS signatures (is this really from who it says it's from?)
3. CHECKS balances (does the sender actually have the money?)
4. ORDERS transactions (who sent first? This determines who gets paid)
5. SHARES its view with other validators
6. COMMITS agreed transactions to the ledger (final, irreversible)
The clever part is step 5-6: validators gossip with each other until they all agree on the same order. Once ⅔+ agree, the transaction is final. This takes under 1 second.
What does a validator NEED?
Hardware (surprisingly modest):
| Requirement | Minimum | Our machines |
|---|---|---|
| CPU | 4 cores | superlocal: 32, gmktec: 16 |
| RAM | 8 GB | superlocal: 124 GB, gmktec: 28 GB |
| Disk | 50 GB SSD | both have plenty |
| Network | 10 Mbps reliable | both on Tailscale |
| Uptime | 95%+ | systemd + auto-restart |
A validator is NOT a mining rig. It doesn't need GPUs, doesn't use much electricity, doesn't generate heat. It's just a reliable computer that runs one process.
Software (one binary):
# This is literally all you need to run a validator:
./iota-localnet start --network.config /path/to/config --validator
The binary is 67 MB. The config specifies the other validators' addresses and the genesis file. That's it.
What we already have:
| Machine | Location | Status | Ready to validate? |
|---|---|---|---|
| superlocal | our server | ✅ running the chain now | ✅ already the sole validator |
| cachyos-x8664 (gmktec) | fleet machine | ✅ running relay #2 | ✅ just needs the validator binary + config |
| Cloud VM | Hetzner/AWS | ❌ not set up yet | needs a $20/month VPS |
| bb-mini | cold spare | ❌ powered off | ready when needed (installer exists) |
To go from 1 → 4 validators: 1. Copy the validator binary to cachyos-x8664 (~2 minutes) 2. Spin up a cloud VM (~10 minutes) 3. Update the genesis config to include all 4 validators 4. Restart the chain (~5 minutes) 5. Total: under 30 minutes to go from "if superlocal dies, the chain stops" to "the chain survives any single machine dying"
Why validators matter for trust
| Scenario | 1 validator (now) | 4 validators | 7+ validators |
|---|---|---|---|
| Machine crashes | ❌ chain stops | ✅ chain continues | ✅ chain continues |
| Machine hacked | ❌ attacker controls everything | ✅ chain is safe (they only have 1 of 4 votes) | ✅ chain is safe |
| We go rogue | ⚠️ we COULD rewrite history (we don't) | ✅ requires 3 of 4 machines colluding | ✅ requires 5 of 7 colluding |
| Someone tries to double-spend | ❌ we're the only referee | ✅ 3 other referees say no | ✅ 5+ referees say no |
The honest truth: with 1 validator, you're trusting US. With 4, you're trusting that at least 3 of 4 machines are honest. With 7+, you're trusting that any majority is honest. That's the decentralisation path.
What validators DON'T do
- They don't store your messages (that's the relay's job)
- They don't hold your money (that's the Move smart contracts)
- They don't read your data (they only see transaction headers, not content)
- They don't earn fees (no inflation, no gas — we run them as a service)
The validator roadmap
| Phase | Validators | What changes | Effort |
|---|---|---|---|
| Now | 1 | We are the single point of trust | — |
| Phase 2 | 4 | Any single machine can die; chain survives | 30 minutes setup |
| Phase 3 | 7+ | External operators join; we step back | days + legal/governance |
| Future | Open | Anyone can run a validator (permissionless) | the full decentralisation path |