[WIP] Committee aggregation #282

kamilsa · 2026-01-14T12:38:27Z

🗒️ Description

Introduces aggregator role and subnet aggregation.

Every validator is assigned to one of the attestation subnets
Aggregators collect signatures that correspond to validators from their subnet
If aggregators collected 90% of signatures from their subnet by the beginning of slot 2, they produce aggregated attestation and propagate it into aggregation topic
Validators subscribe to aggregation topic, so that next block proposer may include committee aggregations into the block (without recursive aggregation for now)

Remaining work

Process the case when aggregators did not observe enough signatures by the beginning of interval 2
Add predicates for gossipsub propagations of aggregated signatures (e.g. do not propagate aggregation if we already observed the one for the same committee, but proving more validators signatures)
Update to latest master

🔗 Related Issues or PRs

leanEthereum/pm#56
leanEthereum/pm#58

✅ Checklist

Ran tox checks to avoid unnecessary CI fails:
```
uvx tox
```
Considered adding appropriate tests for the changes.
Considered updating the online docs in the ./docs/ directory.

use committee aggregated signature proofs

also rename gossip_signatures to gossip_committee_signatures

jihoonsong

Great work! Please excuse leaving some comments while it's still in draft. Just wanted to help iterate faster :)

docs/client/validator.md

jihoonsong · 2026-01-14T15:25:40Z

src/lean_spec/subspecs/containers/state/state.py

        # Configure the genesis state.
        genesis_config = Config(
            genesis_time=genesis_time,
+            attestation_subnet_count=AGGREGATION_COMMITTEE_COUNT,


I admit that I was the one who advocated for attestation committee, but based on the fact that validators only push their attestations to aggregators in their subnet without subscribing to it, I now think aggregation committee gives us slightly better description.

I don't mind whichever we choose—either attestation committee or aggregation committee—but I do think we need to stick to one thing consistently in the Lean spec and pq-devnet-3.md in the pm repo.

I don't have a strong preference too, however I think the rationale for topic names in beacon chain spec is based on the type of messages that are being propagated to this topic. For consistency we should probably stick to the same logic and keep using attestation subnets and attestation committees

Sounds great! In the same vein, what do you think about this one?

looks good to me, applied your suggestion

src/lean_spec/subspecs/forkchoice/store.py

…ted references

# Conflicts: # src/lean_spec/subspecs/forkchoice/store.py # src/lean_spec/subspecs/networking/__init__.py

src/lean_spec/subspecs/networking/gossipsub/topic.py

jihoonsong · 2026-01-15T11:36:40Z

docs/client/validator.md

-Validators participate in consensus by proposing blocks and producing attestations. This
-document describes what honest validators do.
+Validators participate in consensus by proposing blocks and producing attestations. 
+Optionally validators can opt-in to behave as aggregators in a single or multiple 


A validator is assigned to a certain subnet but can they perform an aggregator role in multiple subnets?

Good question. I think we may add this possibility in future. But lets start with one subnet per aggregator.

I think we may consider similar logic to how validators self-assign themselves for sampling of single or multiple columns data in PeerDAS

Sounds great! We can update this description to a single committee and modify it later once we introduce multiple subnets aggregator.

Good question. I think we may add this possibility in future. But lets start with one subnet per aggregator.

I think we may consider similar logic to how validators self-assign themselves for sampling of single or multiple columns data in PeerDAS

I think we need to define this properly, just to clarify, nodes self-assign themselves (validators attached or not), a distinction required here because a signedaggregatedattestation can't be produced here without a validator

docs/client/validator.md

jihoonsong · 2026-01-15T11:42:53Z

docs/client/validator.md

-
-When aggregation is added, aggregators will collect attestations and combine them.
-Aggregated attestations will be broadcast separately.
+Devnet-2 introduces signatures aggregation. Aggregators will collect attestations and combine them. Aggregated attestations will be broadcast separately.


Suggested change

Devnet-2 introduces signatures aggregation. Aggregators will collect attestations and combine them. Aggregated attestations will be broadcast separately.

Devnet-3 introduces signatures aggregation. Aggregators will collect attestations and combine them. Aggregated attestations will be broadcast separately.

…nt in configuration

Aggregate during interval 2 if more threshold signatures were received

g11tech · 2026-01-21T09:45:54Z

docs/client/validator.md

+
+In the devnet-3 design, however, there is one global subnet for signed 
+attestations propagation, in addition to publishing into per committee subnets.
+This is due to 3SF-mini consensus design, that requires 2/3+ of all 


this global bit is not required, once the aggregtors publish signed attestations in the 2nd interval, they can be imported by all validators in the 3rd interval

g11tech · 2026-01-21T10:18:18Z

src/lean_spec/subspecs/chain/config.py

+ATTESTATION_COMMITTEE_COUNT: Final = Uint64(1)
+"""The number of attestation committees per slot."""
+
+COMMITTEE_SIGNATURE_THRESHOLD_RATIO: Final = 0.9


seems too high, anyway the aggregation should be deadline based and not threshold based, this is something a client implementations can deal with themselves when to do optimistic aggregation

g11tech · 2026-01-21T10:20:53Z

src/lean_spec/subspecs/containers/attestation/attestation.py

            for data, validator_ids in data_to_validator_ids.items()
        ]
+
+class SignedAggregatedAttestation(Container):


@anshalshukla / @GrapeBaBa do we already have this type?

also better to use message, signature terminlogy

I think we also need aggregated bit vector here as well,

kamilsa added 14 commits January 13, 2026 09:47

networking: update committee docs

b5b6d8a

networking: add committee size configuration

4867d7d

store committee attestations

7bcedca

Add aggregation in 2nd interval

980b5e8

Committee aggregation

60468af

Rename aggregation committee size to count for clarity

213504a

Remove committee signatures

4fac983

Refactor build_block:

f2651d8

use committee aggregated signature proofs

Clarify attestation broadcasting and update Devnet reference

cc7548c

remove adding proposer signatures to gossip_signatures

cb1a21b

also rename gossip_signatures to gossip_committee_signatures

Refactor subnet ID computation and rename committee signatures variable

e398823

Store proposer signature if same subnet

90fc114

Update build block with selecting aggregations

cdae6a4

Uncomment on_attestation during on_gossip_aggregation

b24d3ed

kamilsa mentioned this pull request Jan 14, 2026

PQ Interop #22 | Jan 14, 2026 ethereum/pm#1866

Closed

Update gossipsub topic names to reflect devnet3

5c952ff

jihoonsong reviewed Jan 14, 2026

View reviewed changes

kamilsa added 2 commits January 15, 2026 16:13

Rename aggregation committee to attestation committee and update rela…

8a0c121

…ted references

Merge remote-tracking branch 'origin/main' into committee-aggregation

cb952f8

# Conflicts: # src/lean_spec/subspecs/forkchoice/store.py # src/lean_spec/subspecs/networking/__init__.py

jihoonsong reviewed Jan 15, 2026

View reviewed changes

kamilsa added 6 commits January 15, 2026 19:59

refactor: rename committee aggregation topic to aggregated attestation

9d721bd

update validator.md to clarify subnet usage in attestation committees

baddbeb

feat: add threshold ratio for committee signature aggregation

6556e81

feat: replace attestation_subnet_count with attestation_committee_cou…

3477d6e

…nt in configuration

feat: add committee signature threshold ratio chain config

9174f5b

feat: aggregate on gossip

3115ef5

Aggregate during interval 2 if more threshold signatures were received

g11tech reviewed Jan 21, 2026

View reviewed changes

	Devnet-2 introduces signatures aggregation. Aggregators will collect attestations and combine them. Aggregated attestations will be broadcast separately.
	Devnet-3 introduces signatures aggregation. Aggregators will collect attestations and combine them. Aggregated attestations will be broadcast separately.

[WIP] Committee aggregation #282

Are you sure you want to change the base?

[WIP] Committee aggregation #282

Uh oh!

Conversation

kamilsa commented Jan 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🗒️ Description

Remaining work

🔗 Related Issues or PRs

✅ Checklist

Uh oh!

jihoonsong left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

g11tech Jan 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

g11tech Jan 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

kamilsa commented Jan 14, 2026 •

edited

Loading

g11tech Jan 21, 2026 •

edited

Loading

g11tech Jan 21, 2026 •

edited

Loading