[#976] Aggregate devshards stats with CurrentEpochStats

[dcastro]

In this PR: when a subnet is settled, we take the subnet's HostStats and merge them with the participant.CurrentEpochStats.

NOTE: participant.CurrentEpochStats keeps track of the total number of inferences performed (InferenceCount) and the number of successfully validated inferences (Validated). Subnets were not keeping track of these 2 stats, so we had to alter the subnet to start doing that and include those stats in the settlement.

---

Since the subnet's stats are now merged into participant.CurrentEpochStats, they will automatically be taken into account when:

• calculating punishments WorkCoins/RewardCoins (see bitcoin_rewards.go)
• participant's inactivity status (see status.go -> ComputeStatus)

[Copilot]

Pull request overview

This PR extends subnet host performance statistics with inference_count and validated, propagates those fields through protobuf/JSON boundaries (subnetctl + Kotlin client + chain types), and aggregates the resulting settlement stats into chain-side per-epoch storage and participant CurrentEpochStats.

Changes:

• Add inference_count and validated to subnet host stats data models and protobuf schemas; update hashing/serialization to include them.
• Track InferenceCount and Validated in the subnet state machine, and add/adjust tests around Validated counter transitions.
• Aggregate settlement host stats into chain SubnetHostEpochStatsMap and into Participant.CurrentEpochStats, with keeper/msg-server test coverage.

Reviewed changes

Copilot reviewed 21 out of 21 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
testermint/src/main/kotlin/data/subnet.kt	Adds inference_count and validated fields to Kotlin settlement host stats DTO.
subnet/types/state.pb.go	Regenerated protobuf bindings to include new host stats fields.
subnet/types/domain.go	Extends HostStats domain model with InferenceCount and Validated (+ doc comments).
subnet/state/machine.go	Updates state transitions to increment/decrement the new counters during inference/validation lifecycle.
subnet/state/machine_test.go	Adds tests validating HostStats.Validated transition behavior for validations.
subnet/state/hash.go	Includes new fields in host stats hash preimage (state root consistency).
subnet/proto/subnet/v1/state.proto	Adds new host stats fields to subnet state protobuf schema.
subnet/cmd/subnetctl/marshal_test.go	Updates settlement marshal/unmarshal tests to cover new fields.
subnet/cmd/subnetctl/main.go	Extends JSON marshaling output with the new host stats fields.
inference-chain/x/inference/types/subnet_settlement.pb.go	Regenerated chain settlement protobuf bindings to include new fields.
inference-chain/x/inference/types/subnet_escrow.pb.go	Regenerated chain escrow/epoch stats protobuf bindings to include new fields.
inference-chain/x/inference/module/commands.go	Extends CLI settlement JSON parsing/mapping to include new fields.
inference-chain/x/inference/keeper/subnet_settlement.go	Includes new fields in settlement host stats hash computation.
inference-chain/x/inference/keeper/subnet_settlement_test.go	Updates settlement hashing test fixtures for new fields.
inference-chain/x/inference/keeper/subnet_host_stats.go	Aggregates new fields into SubnetHostEpochStatsMap and introduces aggregation into Participant.CurrentEpochStats.
inference-chain/x/inference/keeper/msg_server_settle_subnet_escrow.go	Applies new aggregation into participant CurrentEpochStats during settlement processing.
inference-chain/x/inference/keeper/msg_server_settle_subnet_escrow_test.go	Adds tests ensuring settlement updates CurrentEpochStats and subnet epoch stats map.
inference-chain/proto/inference/inference/subnet_settlement.proto	Adds new fields to settlement protobuf schema.
inference-chain/proto/inference/inference/subnet_escrow.proto	Adds new fields to escrow/epoch stats protobuf schema.
inference-chain/api/inference/inference/subnet_settlement.pulsar.go	Regenerated Pulsar fast-reflection code for new settlement fields.
inference-chain/api/inference/inference/subnet_escrow.pulsar.go	Regenerated Pulsar fast-reflection code for new escrow/epoch stats fields.

[heitor-lassarote]

LGTM, the only thing that I think would be nice in addition would be a Testermint test, but since validations are currently not working in E2E as expected, I think this may be skipped. In addition, testing some things such as missed requests will probably require some extra infra, which sounds out of scope for this. So I think, for now all is fine.

[akup]

Here are new state fields, and on update it will lead to immediate consensus splits and data corruption on live networks.

@gmorgachev How adding fields is supposed to be handled in subnets? It is not binded to block height of the mainnet

[akup]

Threads subnet InferenceCount / Validated into L2 hashing and L1 settlement so subnet work feeds CurrentEpochStats. It has hard consensus and economics risk that is flagged: L2 hash changes without a coordinated rollout can split hosts, and incremental Validated accounting in machine.go is easy to get wrong (ordering, double application, underflow), which then flows into L1 rewards. L1 settlement may also mis-attribute stats across epochs, and storage migrations / tests need tightening before this is safe to ship.

Must fix

• Version or rollout gate for L2 hash changes — Including InferenceCount and Validated in subnet/state/hash.go without a protocol/version boundary risks an immediate split between upgraded and non-upgraded hosts; coordinate upgrade or feature gating.

This one point needs to be handled in some generic upgrade tooling (proposed here: #1005 (comment))

• Idempotent validation outcome handling — When vote thresholds fire, ensure Validated is not incremented again on late or duplicate votes after the record is already validated (subnet/state/machine.go).
• Order-dependent running tallies — The logic updates sm.state.HostStats[rec.ExecutorSlot].Validated via scattered increments and decrements based on the real-time stream of MsgValidation messages.
  If a malicious sequencer orders an invalid vote before a valid vote, the rec.VotesInvalid == 0 condition fails, and the counter is not incremented. If the inference subsequently crosses the validation threshold, it appears another increment might occur in applyValidationVote (line 828). This spread of ++ and -- logic across different handlers based on transient state (rec.VotesValid == 0) guarantees that any future changes to voting rules will introduce accounting bugs.

Should fix

• Test isolation — Stop mutating global sdk.GetConfig() in tests (parallel runs / flakiness); use TestMain or scoped setup (msg_server_settle_subnet_escrow_test.go).
• Aggregation safety and failure behavior — Aggregation of InferenceCount and Validated lacks overflow protection. Fix: Use bits.Add32 and return an error if an overflow occurs.

Nits

• Factor duplicated settlement test setup into a shared helper.
• Tests re-deriving executor slot (inferenceID % len(hosts)) — couple to the real assignment from state/records instead of duplicating internals (machine_test.go).
• Redundant participant loads vs data already in SubnetHostEpochStatsMap — simplify if one source of truth suffices.

[akup]

Should handle votes duplication, order-dependent ++/--, overflows checks

[dcastro]

Should handle votes duplication, order-dependent ++/--, overflows checks

You mean in the updates of HostStats.Validated in machine.go?
Yeah, that is already being handled, there is a test in TestApplyDiff_Validation_ValidatedCounterTransitions that ensures duplicate calls and reordering of MsgValidation remains idempotent