Phase 4 — Coil Integration

A — Coil's fees become TARE's yield

CoilFeeRouter is installed as Coil's fee_recipient. The permissionless harvest(token) pulls the accrued (≤1%) fee: TARE passes straight through (donated as-is); USDC is swapped USDC→TARE through the canonical Curve StableSwap-NG pool, then donated to the engine's permissionless donate_to_surplus() — no engine change.

• Min-output derived from the pool's manipulation-resistant EMA price_oracle, capped by an owner-set, hard-capped slippage.
• The router measures the actual TARE delta received, not the pool's return value. A sandwich below the EMA floor just reverts.
• The owner can only set capped slippage and recover non-core tokens — recover rejects TARE and USDC outright.

B — lock TARE, earn Coil's fees in USDC

A capped slice of each USDC fee routes — still in USDC — to gauge voters, with zero veForge contract changes. A second EmissionRouter instance whose reward token is USDC feeds the existing token-agnostic BribeDistributor; CoilFeeRouter peels off a bounded slice before it swaps.

• gauge_bps hard-capped at 50% (set to 20%); surplus always keeps the majority of every harvest.
• Sink unset or rate at 0 → 100% flows to surplus and B is a no-op.
• Weights activate at the next weekly epoch; until then distribute reads zero weight and no-ops.

C — the vault makes markets on Coil, and still can't be drained

CoilMakerStrategy is a Keep strategy (asset = USDC) that is itself a Coil trader: it rests limit intents ("sell USDC, buy TARE, at or above this price") and authorizes them on-chain via EIP-1271 — a keeper-managed registry of approved order digests, answered from isValidSignature.

• The whole safety story is one line: approve_intent floors every order's min_buy at the pool's EMA oracle — and the keeper cannot widen it. A resting order can never fill below oracle value.
• Valuation is EMA-based (idle USDC + banked TARE·EMA), counted exactly once; a resting order moves no funds until filled.
• Exit doesn't wait on a solver — harvest() swaps banked TARE back to USDC directly through the pool, EMA-floored.
• The vault is the only counterparty; the admin can't sweep USDC or TARE.

Invariants, fuzzed

For the maker strategy a stateful machine plays all three roles — vault, keeper, solver — interleaved with harvests, time travel, and EMA moves, asserting after every step:

• No drift — totalAssets == idle_USDC + TARE·EMA / scale.
• No value creation — convertToAssets(totalSupply) ≤ totalAssets (+dust).
• 1271 soundness — magic value returned iff the digest is keeper-approved, uncancelled, and unexpired.
• Price-per-share never falls across surplus-only flow; the vault is the sole shareholder.

Totals: CoilFeeRouter 27+1 · CoilMakerStrategy 12+1 · plus a veForge integration test proving the 6-decimal USDC emission path end-to-end. Full Keep suite: 160 green, no regression.

Proven on a Sepolia fork vs real contracts

Mocks prove logic; they don't prove you got a real counterparty's ABI, decimals, and EIP-712 digest right. Both risky integrations were run on a Sepolia fork against the real contracts.

• A settled a real Coil batch so Coil accrued a genuine USDC fee, then harvested through a real Curve NG pool: 1 USDC fee → 0.99989 TARE (EMA-priced); engine surplus rose by exactly that.
• C · negative — a solver tried to settle an intent the keeper never approved → Coil's EIP-1271 check rejected it (invalid signature).
• C · positive — keeper approved a sell-USDC/buy-TARE intent; a solver settled above the floor → the strategy sold 200 USDC and banked 201.96 TARE. totalAssets rose 1000.00 → 1001.96; harvest realized TARE back to 201.94 USDC; the vault redeemed 1001.94 USDC on its 1000 deposit — market-making yield captured by a contract that never held a key to anyone's money.

Sepolia addresses (chain 11155111)