Gloas (ePBS) testnet profiles for Anchor epbs testing

[shane-moore]

What this adds

Two Gloas (EIP-7732 / ePBS) testnet profiles for ssv-mini, so Anchor's epbs branch can be
exercised end to end on a local Glamsterdam-style chain with its own genesis and SSV contracts:
block proposal through the ePBS block structure, and attestation against the new 3s Gloas deadline.
A 4-operator all-Anchor cluster does real duties; no go-SSV nodes involved.

Built for Anchor devs: point it at your epbs work and watch the cluster propose and attest on a
Gloas chain.

Quick start

ANCHOR_COMMIT=epbs make prepare-anchor   # build node/anchor from sigp/anchor epbs
make run-gloas                           # Gloas at genesis, 4-operator all-Anchor cluster
make logs SERVICE=anchor-node-0          # tail an operator

ePBS is gated on the Ethereum Gloas fork (gloas_fork_epoch); there is no SSV-side fork knob. The
epbs branch already carries this model, so a plain ANCHOR_COMMIT=epbs build is all you need.

Requires Docker and a recent Kurtosis CLI (1.19.x; 1.15.x rejects a GpuConfig builtin in the
pinned ethereum-package). Full walkthrough and edge cases are in the new "Gloas (ePBS) testnet"
section of the README.

Testing your own Anchor changes

prepare-anchor builds node/anchor from any branch/tag/commit on your ../anchor clone's origin:

ANCHOR_COMMIT=my-epbs-feature make prepare-anchor && make run-gloas

Or build the image straight from your working tree and skip prepare entirely:

cd ../anchor && docker build -f Dockerfile.devnet -t node/anchor .
cd ../ssv-mini && make reset PARAMS_FILE=params-gloas.yaml   # scoped teardown + re-run

Either way the profile consumes whatever is tagged node/anchor. There is no hot-swap for Anchor
nodes (unlike restart-ssv-nodes), so the iteration loop is: rebuild the image, then make reset.

The two profiles

Profile	Command	Exercises
params-gloas.yaml	make run-gloas	Gloas at genesis, 4 Anchor operators, 10 keyshare validators registered on-chain. Proposal duty (attestation timing needs the #1061 fix in the image).
params-gloas-builders.yaml	make run-gloas-builders	Above plus 2 genesis ePBS builders and buildoor: external bids, payload reveals, chain-level PTC (payload_attestations).

Reference behavior (tell a regression from a known gap)

Measured live on an epbs image without #1061, local devnet:

• Proposals work. Cluster publishes Gloas blocks ~+0.13s into the slot, canonical (verified on-chain: multiple cluster-proposed slots, all canonical=true, "Successfully published block" on every operator).
• Attestations: broken without #1061; working only once #1061 is also applied. Measured on an epbs image without #1061: 0 cluster attestations published, even though the attestation QBFT reaches COMMIT consensus. The sync-committee half stalls and takes submission down with it (see the #1061 note below). With #1061 also in the image, attestations publish ~+4.09s vs the 3.0s Gloas deadline (QBFT itself ~15ms; the remaining gate is the committee partial-signature batch waiting on the sync-message flow, which fires on Lighthouse's static 4s sync timer; head-event pipeline healthy at ~+0.12s). That ~+4.09s is an Anchor/LH timing-wiring gap, not a cluster failure, and is identical on every operator.
• Cluster-proposed slots are execution-empty by design. Anchor self-builds and never reveals the payload (SIP-94 envelope stub), so PTC votes payload_present=false on them and the EL does not advance that slot. Expected, not a bug.
• Gloas is gated on the Ethereum fork; SSV attestations need the #1061 fix in your anchor image. ePBS activates purely from gloas_fork_epoch; there is no SSV-side fork knob. With Gloas active the committee runs GloasBeaconVote for attestations, but the sync-committee signing path still runs BeaconVote (feat(validator_store): migrate sign_committee_sync_committee_signatures to GloasBeaconVote at CStar sigp/anchor#1061, open), so the two resolve to different QBFT instances instead of sharing one. Observed mechanism: the attestation instance reaches COMMIT, but the sync instance round-changes every slot and never completes, and because the committee's post-consensus partial-signature batch is sized to attestations + sync messages, it never fills, so the agreed attestation is never submitted (0 published, every operator). So attestation testing needs an anchor image carrying #1061; proposals, external-builder bids, and chain-level PTC work regardless. (Consensus succeeds; only submission is blocked.)

Related Anchor issues: sigp/anchor#1061 (sync GloasBeaconVote migration; required for SSV attestations), sigp/anchor#1082 / #1078 (PTC duties).

Mechanics worth knowing

• Imports a forked ethereum-package pin (shane-moore/ethereum-package, by commit), fetched automatically by Kurtosis. Needed because Gloas genesis requires ethereum-genesis-generator >= 6.0.0 (no tagged release ships it) plus a one-line fix so genesis ePBS builders do not collide with the preregistered keyshare validators (indices 64-73).
• Uses nethermind + lighthouse glamsterdam-devnet-5 images (geth has no Glamsterdam build).
• Registration scripts carry two live-verified fixes: explicit gas limits (nethermind's eth_estimateGas underestimates SSVNetwork delegatecall paths ~3x) and msg.value on the payable v2.0.0 bulkRegisterValidator.
• register_validators (default false) gates on-chain registration so the default aetheria flow is untouched; the Gloas profiles set it true so operators get duties.
• All-Anchor cluster (ssv.count: 0): no node/ssv image needed, and monitor is disabled.

Why draft

This is a testing tool for the in-flight epbs work, not a finished feature. Opening as draft for
visibility and so Anchor devs can use it now; happy to adjust scope or take out of draft per review.