SIP-2: Epochs, Rewards, and Settlement

• Status: Draft (implemented; single-node live-proven on regtest — exact-conservation mints; multi-node relay proof in progress)
• Implementation: crates/consensus/src/{epoch.rs,sealer.rs,follower.rs}, crates/engine/src/{payload.rs,builder.rs,driver.rs} (normative)
• Author: Sova (orchestrated draft)

Epochs

One Zcash block = one epoch. Every SIP-1 burn confirmed in the epoch’s block makes its EVM address a miner of that epoch; an address’s weight is the saturating sum of its burns. One Sova block is produced per epoch; from base height B, epoch E’s expected Sova height is E − B + 1. A Zcash reorg implies the corresponding Sova reorg; a mint is final at Zcash confirmation depth. Among Sova blocks on one Zcash chain, fork choice is the preference below, applied by the branch rule: a candidate counts only if its ancestry, through blocks the node holds, meets the node’s chain, and at the fork point (where it leaves that chain) it either extends the node’s head or its block beats the node’s block there by preference, replacing at most MAX_REPLACE_DEPTH = 3 of the node’s blocks; candidates are ordered by their blocks at the height where their branches part. An online node therefore never replaces a block once three blocks are built on it. (Corrected 2026-09-23 after the reorg audit, docs/audits/2026-09-23-reorg-and-fork-choice.md; the earlier text read “Sova fork choice follows the node’s Zcash view … finality = Zcash confirmation depth”, which is true of mints only.) (Corrected 2026-09-23: sibling rule replaced by the bounded branch rule, see audit F1 follow-up. The sibling rule, “a candidate must extend the node’s block at the previous height”, made any split permanent.)

Ranking and sealing

Miners rank by (weight desc, then byte-lexicographic min-txid asc). Rank 0 is the epoch’s sealer; ranks 1… are liveness fallbacks on a timeout ladder (rank r may produce at elapsed ≥ r × step; draft step 15 s). Timeouts are liveness-only, never validity or preference: among an epoch’s candidate blocks, preference is (rank asc, block hash asc) — a late rank-0 block displaces an on-time rank-1 block, in the common case by a one-block replacement of its sibling; a branch that would replace more than three of the node’s blocks is not a candidate (corrected 2026-09-23 after the reorg audit: “micro-reorg bounded to the epoch” did not bound the replaced block’s ancestry; corrected again 2026-09-23: sibling rule replaced by the bounded branch rule, see audit F1 follow-up; the sibling-rule text read “a block on another parent is not a candidate”); equivocations tie-break by hash. A node whose chain head already covers an epoch’s expected height does not produce for it. Empty epochs (no burns) may be extended rewardless by recent sealers on the same ladder.

Rewards

Draft schedule: 6,250 SOVA per epoch (21B cap, 4-year halvings at 75 s epochs; final numbers are SIP-3). All reward math runs in gwei so every share is gwei-exact. The reward splits: a 10% sealer tip, and a pro-rata pool by weight (floor division); all rounding dust joins the tip. Conservation is exact: shares always sum to the full epoch reward. The sealer must be a ranked miner of the epoch.

Settlement

An epoch’s rewards enter the EVM as the block’s withdrawals (Ethereum’s consensus-grade balance-increment channel, inherited unchanged from the execution layer): withdrawal i = { index: i, validator_index: 0 (reserved marker), address, amount in gwei }, in rank order. A block claiming epoch E must carry exactly the withdrawals this derivation produces — carried withdrawals that differ from the derivation make the build/import invalid. Amounts must be gwei-aligned, non-zero, and fit u64 gwei (guaranteed by the gwei-domain reward math).

Validation roadmap (normative intent)

Every node re-derives an imported block’s expected settlements from its own Zcash view and rejects mismatches (C5); v1 nodes converge on first-arrival via relay with rank preference following (documented in docs/design/gossip-v1.md).

Implemented and proven 2026-09-22 (ladder scenario, nightly CI): a block for epoch E is valid iff its withdrawals match the derivation for some rank of E (the sealer tip makes each rank’s derivation distinct, so the sealer’s rank is recoverable from the withdrawals alone — no sealer metadata rides the wire); preference (rank asc, hash asc) arbitrates between valid candidates on the receiver, never the sender (“relay delivers, arbiter decides”); burn-less epochs require empty withdrawals and their candidates tie-break by hash. A late rank-0 block displacing an on-time rank-1 block via bounded micro-reorg is observed behavior, not just intent. Since 2026-09-23 the bound is the branch rule’s (Epochs): at most three blocks, and nodes split by up to three blocks converge; a deeper split does not heal by itself, and a joining or restarting node takes the first valid history offered (audit F2), until cumulative sealer-rank fork choice, SIP-8 anchored burns or client checkpoints ship.