Test a pool ranked out of the Leios voting committee

[mkoura]

The Leios voting committee seats the leiosCommitteeSize pools with the most stake,
largest first, ties broken by ascending pool id. A local testnet has three pools and
room for 900, so every pool is always seated - neither the ranking nor what happens to
a pool that misses out was covered by anything.

TestLeiosCommitteeRank

Starts a cluster instance whose Dijkstra genesis gives the committee fewer seats than
the cluster has pools, which is what makes the ranking observable. It then:

• finds the pools the committee left out;
• checks they were left out on rank alone - they have a registered BLS key, and no
  seated pool ranks behind them under (-stakeSet, poolId), the key
  selectLeiosCommittee sorts on;
• searches the pool logs for the rest of the epoch the committee was read in, and
  checks that the pools without a seat answer every EB announcement with
  NotOnCommittee and cast no vote, while the seated pools vote, decline for no reason
  of their own, and still add up to a certificate.

The quorum is lowered along with the committee. A seat is weighted by the pool's share
of the active stake, not of the committee, so a committee that doesn't hold every
pool tops out at SMALL_COMMITTEE_SIZE / NUM_POOLS and could never reach the default
0.75. That ratio is the bound to respect when changing the committee size, and it is
noted next to the constant.

Evidence

From a full testrun on leios_fast, with the genesis coming out as
leiosCommitteeSize: 2, leiosQuorumStakeThreshold: 0.5:

pool	AnnouncementAccepted	Voted	Certified	NotVoted
pool1 (seated)	22	21	21	1 ChainTipDoesNotAnnounce
pool2 (seated)	22	21	21	1 ChainTipDoesNotAnnounce
pool3 (ranked out)	22	0	21	21 NotOnCommittee

Two things this settled. The reason a seatless pool reports is NotOnCommittee, not
SignerNotInCommittee - worth pinning down, because LeiosSeat carries only a weight
and a vkey with no pool id, so it is not obvious from the ledger source which one the
node would give. And the lowered quorum holds up: 21 certificates off 22 announcements,
with seat weights near 1/3 each, means both seated pools voted every time, since one
alone (0.333) is under the 0.5 threshold.

Second commit

test_expired_bls_key read bksRegisteredIn from one cluster pool and put every other
pool on the schedule that followed from it, with a comment standing in for a check. The
assumption is load bearing - the test asserts every non-rotated pool is still voting at
expire_epoch - 1 and that none is at expire_epoch, which needs
expire_epoch <= min(reg) + max_key_age and >= max(reg) + max_key_age respectively,
so it only holds when the stamps are identical. Now read from every pool and asserted,
so a cluster that registered pools across an epoch boundary reports that instead of
failing as a pool that kept voting too long. Latent today (POOLS_IN_GENESIS=true
stamps everything epoch 0), but the kind of thing that only bites on a slow machine.

Shared code

• leios.NOT_VOTED_MSGS groups the reasons a pool can give for being unable to vote.
• leios.skip_if_no_ebs_in_genesis takes the genesis instead of a cluster instance, so
  a test that starts a cluster of its own can be ruled out before paying for the
  startup.
• helpers.get_pool_id_hex replaces the inline Bech32 conversion in
  bls.get_committee_seat.

Two notes for whoever reviews:

The test can't be exercised against a dev cluster - .source.dev sets FORBID_RESTART=true, and cluster_getter rejects a custom scriptsdir under it, the same wall test_expired_bls_key hits. It needs a full testrun, which is where the table above comes from.

The plural handling (out_pool_names, cut_off_rank = min(...)) is currently unreachable, since SMALL_COMMITTEE_SIZE = NUM_POOLS - 1 always leaves exactly one pool out. It's there so the test doesn't encode an arithmetic identity that breaks the moment the committee size is chosen independently of the pool count.

[Copilot]

Copilot review overview

🟡 Changes recommended

Unresolved test-evidence validation gaps remain, and the new helper lacks unit coverage.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 1

Open (1)

• Membership check allows mixed announcement outcomes · New

What changed in this PR

Adds Leios committee-ranking and out-of-committee voting coverage, shared helpers, and stronger BLS registration-epoch validation.

Changes:

• Adds reduced-committee ranking and voting tests.
• Adds shared Leios and pool-ID helpers.
• Validates BLS registration epochs across pools.

File	Summary
cardano_node_tests/​utils/​helpers.py	Adds pool ID normalization.
cardano_node_tests/​tests/​test_leios_blocks.py	Adds committee ranking and voting checks.
cardano_node_tests/​tests/​test_bls_rotation.py	Validates consistent BLS registration epochs.
cardano_node_tests/​tests/​leios.py	Adds shared Leios constants and genesis skip logic.
cardano_node_tests/​tests/​bls.py	Uses shared pool ID normalization.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.