Every confidential-compute machine Flare has on record, and whether it is really there.

Read straight from the FlareTeeManager registry on Coston2 and probed live. Nothing here needs anyone's permission or cooperation: the register is public, the proxies answer or they do not, and every row is re-derivable from public RPC by anyone.

Fig. 1 — Register summary at block 33682349. “Simulated” means the machine attested to a simulator: a single shared code hash that binds to no source code at all. It is a legitimate development mode, not a fault — but it is not evidence of anything.

“Check the code hash” has no answer yet.

Confidential compute rests on one primitive: a machine registers a hash of the code it runs, and you check it. That instruction is only as strong as the hash is distinctive — so this asks the registry how much a hash actually identifies. If a value is carried by many independent owners, learning that a machine carries it tells you almost nothing, because almost every machine would have given you the same answer.

bits = −log₂( machines carrying this hash ÷ machines in the registry )

Rebuilt from source, here, twice each

Flare's own recipe, run by a third party with no relationship to them: buildx on the docker-container driver, --no-cache, SOURCE_DATE_EPOCH from the commit, rewrite-timestamp=true. Each target is built twice and the digests must agree.

Fig. 3 — Every language image also required a prerequisite base image the documented recipe does not mention, built from a ref derived from go/go.mod. Without it, no extension image rebuilds at all.

The register

Sorted by how much attention each machine needs. Real confidential hardware outranks a simulator in every combination — 95% of this register is someone developing, and burying the 14 rows that carry weight beneath them would make the page useless.

Fig. 2 — Verdicts carry a glyph, a border style and a colour at once, so the register stays readable printed in greyscale or by a colourblind reader. 76 of 94 proxy URLs are on hosts whose addresses rotate by design (quick tunnels, free ngrok, Codespaces) — those registrations are expected to rot.

Measuring a hash is not the same as fixing a machine

This register answers what a code hash establishes. It does not tell an operator what to do, and most of this fleet has something to do: the diagnosis finds a blocker on the large majority of registered machines, almost all of them URLs that have rotated away from what is registered on chain.

[ Every machine, diagnosed, with the fix → ]