SnapStream

Restart coordination for Solana validator operators

None of this is load-bearing. If SnapStream goes dark tonight, every validator keeps voting and the cluster still restarts without it.

Open the Drill Deck ↗ Open the AG Dashboard ↗

01 - The problem

The plan lives in a Discord thread and a shared doc. It still does.

That's how we've coordinated every real restart we've been through: read the same message, watch the same thread, and wait for everyone else to say they're ready.

Solana tried to replace that once. SIMD-0046, known as wen_restart, took roughly two years of engineering work across Anza and Firedancer, shipped behind CLI flags, and sat unused. Anza removed it from Agave in February 2026 (PR #10493). The removal PR was blunt about why: never used or fully tested, largely broken under Alpenglow, and Firedancer no longer willing to support its own implementation.

SnapStream isn't a second attempt at that same bet. It sits outside consensus and coordinates the people, not the protocol. Rehearsal is how it proves itself, not a live outage.

Three pieces, one job

Muster

A live view of which operators are back and how much stake that represents, so the group can watch supermajority arrive instead of guessing at it.

Writ

One signed, versioned restart plan, verified by more than one party, in place of the thread everyone re-reads to check they got the same instructions.

Preflight

Checks that run before anyone acts, plus a rehearsal mode that plays out the whole process against a simulated cluster without touching a live validator.

02 - Proof, not promises

Pass/fail criteria are set before a drill runs, not chosen afterward to fit the result.

Every tier writes its criteria down first. Evidence from each run goes into a manifest sealed with a SHA-256 hash, so nothing in it can be edited after the fact. A failed drill gets published too, on purpose.

Tier 1 is complete. Its full replay is public right now.

Watch the Tier 1 replay ↗