Skip to content

Epoch 158 reward underpayment after v0.2.9: preserved inference-slot weight was reset #764

Closed @huxuxuya opened 2026-02-14 16:52 UTC 1 comment Updated 2026-02-17 22:46 UTC

After upgrade v0.2.9, part of epoch 158 rewards appears to be distributed incorrectly. Validators that had ML nodes in preserved inference slot (POC_SLOT, TimeslotAllocation[1]) were not paid.

Suspected root cause: - upgrade migration reset TimeslotAllocation[1] for effective-epoch data; - reward settlement uses preserved weight from this slot (effectiveWeight = preservedWeight + confirmationWeight); - as a result, preserved weight was reduced to zero for affected participants during settlement.

Impact: - epoch 158 reward shares are skewed; - validators with significant preserved/inference-slot weight are affected most.

Requested actions:

  1. Reconstruct historical slot allocation for epoch 158 at effective block height.
  2. Recalculate expected rewards with the same chain formula and filters.
  3. Prepare and execute compensation distribution via upgrade/governance proposal.

💬 Comments (1)

@AlexeySamosadov commented 2026-02-17 21:52 UTC

PR with fix: #771 — removes resetPocSlotsInEpochGroupData from v0.2.9 upgrade handler. This function reset TimeslotAllocation[1] in EpochGroupData which is read during reward settlement, zeroing preservedWeight for all validators in epoch 158.

Note: this is a forward-fix for chain replay correctness. Compensation for epoch 158 affected validators requires a separate governance proposal.

🔄 Auto-synced from Issue #764 every hour.