Spark Capital Allocation

How does billions of dollars flow from Sky Protocol through Spark and into DeFi lending markets? How is revenue calculated? How does it get distributed between Sky and Spark? This article provides the complete technical walkthrough of Spark's capital allocation system - everything you need to understand Sky's economics and the Prime Agent revenue model.

Why This Matters

Understanding Spark's capital allocation is essential for:

• Governance participants: Knowing how settlements are calculated and approved
• Protocol analysts: Understanding Sky's revenue model and Prime Agent economics
• Developers: Building tools that track debt, revenue, and settlements
• DeFi users: Understanding where sUSDS yield comes from

The Big Picture: How It All Works

Here is the complete lifecycle of capital through Spark:

flowchart LR
    P1["1) Governance"] --> P2["2) Deploy"] --> P3["3) Earn"] --> P4["4) Settle"] --> P5["5) Execute"]

• Capital Lifecycle Phases:

• Critical insight: Revenue for Prime Allocator Vaults is calculated off-chain through the Simplified Profit and Loss methodology and settled through Executive Votes that mint new debt on the allocator vault equal to the total revenue earned. This minted amount goes to Sky's Surplus Buffer, then Spark's profit share is returned to them. This design enables instance-level profit calculations and transparent verification in Stage 1 ^[1][2].

Part 1: The Contract Architecture

Spark's allocation system is built on a layered contract architecture that separates governance controls, role assignments, and execution logic. The contracts implementing this system are deployed on Ethereum mainnet and governed by Sky's PauseProxy ^[9].

The Three-Layer System

Spark's capital allocation uses a three-layer architecture defined by the Allocation System Primitive ^[10]:

flowchart TB
    subgraph GOVERNANCE["GOVERNANCE"]
        PauseProxy
        Subproxy
    end

    subgraph ROLES["ROLES"]
        Relayer["ALM Relayer"]
        Freezer
    end

    subgraph CONTRACTS["CONTRACTS"]
        Controller
        RateLimits["Rate Limits"]
        Proxy["ALM Proxy"]
        Vault
        Buffer
    end

    External["External Contracts"]

    PauseProxy -->|"Exec Vote"| Subproxy
    Subproxy -->|"Grant/Revoke"| Relayer
    Subproxy -->|"Grant/Revoke"| Freezer
    Subproxy -->|"Sets"| RateLimits
    Freezer -->|"Freeze"| Controller
    Relayer -->|"1) Calls"| Controller
    Controller -->|"2) Checks"| RateLimits
    Controller -->|"3) Txn Data"| Proxy
    Proxy -->|"4) Execute"| External
    Proxy <-->|"draw/wipe"| Vault
    Vault <-->|"Mint/Burn"| Buffer

• Legend — Roles:

• Legend — Contracts:

• Execution Flow:

Why Stability Fees Are Zero

In Stage 1, stability fees for all Prime Allocator Vaults are set to zero via the SP-BEAM, so that all value transfers happen exclusively through Executive Votes ^[2]. This is the foundation of the off-chain settlement model:

Why This Design?

1. Instance-level granularity: Enables per-instance profit calculations across SparkLend, Aave, Morpho, etc.
2. Flexibility: Allows for complex revenue share arrangements and penalty adjustments
3. Transparency: Forum posts provide full calculation visibility before execution
4. Stage 1 Compatibility: Supports the Simplified P&L methodology

USDT and pyUSD Exception

Two asset classes receive full profit-and-loss treatment (not floored at zero): USDT and pyUSD held by Spark in SparkLend or Spark's ALM Proxy. For these assets, Spark pays the Agent Credit Line Borrow Rate and realizes both profits AND losses — unlike standard instances where Instance Profit is floored at zero ^[2].

Rate Limits: The Safety System

The ALM Controller enforces rate limits to prevent rapid capital movements that could destabilize either party ^[9]:

flowchart LR
    A["maxAmount"] --> B["slope"]
    B --> C["lastAmount"]
    C --> D["currentRateLimit"]

• Formula: currentRateLimit = min(slope × (now - lastUpdated) + lastAmount, maxAmount)

• Spark Example: maxLine=5B, gap=500M, ttl=24h → Can draw 500M/day, max 5B total.

Rate limits apply at multiple levels:

• Vault level: How much USDS can be minted (debt ceiling + gap/ttl)
• Instance level: How much can be deployed to each protocol
• Era system: Operators cannot incur losses exceeding cap per era interval

Part 2: The Complete Capital Flow

Capital moves through five distinct stages, from governance approval of the credit line through deployment into yield-generating protocols and back to Sky through monthly settlement. Each stage involves specific on-chain interactions between the contracts described in Part 1.

Step 1: Credit Line Establishment

Before any capital flows, governance must approve the credit line:

flowchart LR
    PROP["Proposal"] --> CEILING["Set Ceiling"] --> GAP["Set Limits"] --> RISK["Verify Risk Capital"]

• Parameters Set:

• Risk Capital Requirements ^[10]:
  • Junior Risk Capital (JRC): First-loss capital provided by Spark
  • Senior Risk Capital (SRC): Second-loss capital from Sky's balance sheet
  • Both must be in place before credit line is activated

Step 2: USDS Minting (Debt Creation)

When Spark needs capital:

sequenceDiagram
    participant OP as Operator (RELAYER)
    participant CTRL as ALM Controller
    participant VAULT as AllocatorVault
    participant VAT as VAT
    participant BUF as AllocatorBuffer

    OP->>CTRL: mintUsds(amount)
    CTRL->>CTRL: Check rate limits
    CTRL->>VAULT: draw(amount)
    VAULT->>VAT: frob(ilk, amount)
    Note over VAT: Creates debt position<br/>Art += amount<br/>(normalized debt)
    VAT->>VAULT: USDS minted
    VAULT->>BUF: Transfer USDS
    BUF-->>OP: Success

• Technical details:
  • Debt is tracked in VAT as Art (normalized debt) × rate (accumulator)
  • Since duty (stability fee) = 0 in Stage 1, the rate stays constant at 1.0
  • Actual interest is calculated off-chain based on Agent Credit Line Borrow Rate ^[5]
  • Settlement is executed monthly via Executive Vote

Step 3: Deployment to Instances

Capital flows from buffer to yield-generating protocols via the Spark Liquidity Layer ^[8]:

sequenceDiagram
    participant OP as Operator
    participant CTRL as ALM Controller
    participant BUF as AllocatorBuffer
    participant PROXY as ALM Proxy
    participant INST as Instance (Aave/Morpho)

    OP->>CTRL: deposit(instance, amount)
    CTRL->>CTRL: Check instance rate limits
    CTRL->>BUF: transferFrom(amount)
    BUF->>PROXY: USDS transferred
    PROXY->>INST: deposit(USDS)
    INST-->>PROXY: aToken/shares received
    Note over INST: Capital now earning yield

• Instance types and their characteristics:

These instances are defined in the Spark Prime Agent Artifact ^[3].

Step 4: Revenue Generation

Throughout the month, yield accrues from deployed capital:

flowchart LR
    D["Deployed Capital"] --> Y["Yield Accrues"] --> R["Monthly Revenue"]

• Revenue by Instance (Example Month):

Part 3: The Monthly Settlement Cycle (MSC)

The MSC is Sky's mechanism for reconciling Prime Agent operations with Sky Core. This is where the magic happens - where yield becomes revenue, and revenue gets distributed ^[1].

Why Off-Chain Settlement?

Unlike regular user vaults where stability fees accrue automatically on-chain, Prime settlements are calculated off-chain and executed via governance. This design enables:

• Complex revenue arrangements: Instance-by-instance profit calculations
• Transparency: All calculations posted publicly before execution
• Flexibility: Penalties and adjustments can be applied
• Verification: Independent calculations by multiple parties

The Calculation Methodology

This is the Stage 1 Simplified Profit and Loss Calculation ^[2]. For the full computational methodology including Agent Rate deductions, sUSDS adjustments, and Sky Direct Exposure treatment, see Debt-Based PnL Methodology.

flowchart TB
    S1["Total Revenue"]
    S2["Total Profit"]
    S3["Adjusted Profit"]
    S4["Amount Due to Sky"]

    S1 --> S2 --> S3 --> S4

• Formulas:
  • Instance Expense = Invested × Base Rate × (days / 365)
  • Instance Profit = max(Instance Revenue - Instance Expense, 0)
  • Adjusted Profit = Total Profit - Distortion Penalty - Low Yield ASC Penalty

Detailed Example Calculation

Let's work through a real example for a month:

• Given:

  • Spark's average deployed capital: $2,000,000,000
  • Agent Credit Line Borrow Rate (Base Rate): 5.0% APY
  • Month: 31 days

• Instance-by-Instance Calculation:

*Note: Instance Profit cannot be negative. When Revenue < Expense, Profit = 0.

• Expense Calculation Formula:

Instance Expense = Deployed × (Base Rate / 365) × Days
Example (SparkLend): $800M × (0.05 / 365) × 31 = $3.40M

• Penalties (if applicable):

  • Distortion Penalty: 0-100% of Total Profit (discretionary)
  • Low Yield ASC Penalty: MIN(10 × Excess Low Yield ASC %, 100%)

• Final Settlement:

Adjusted Profit = $2.88M - $0 (penalties) = $2.88M
Amount Due to Sky = $11.25M - $2.88M = $8.37M

The Verification Process

MSC calculations go through a three-party verification process to ensure accuracy ^[14]:

flowchart LR
    INITIAL["Initial Calc<br/>Amatsu<br/>(7 days)"]
    INDEPENDENT["Independent Calc<br/>BA Labs<br/>(7 days)"]
    FINAL["Final Calc<br/>Core GovOps<br/>(12 days)"]

    INITIAL --> FINAL
    INDEPENDENT --> FINAL

• Allowed Deviation: If the Initial and Independent calculations differ by ≤1%, the amount is considered "Agreed" and no resolution is needed ^[15]. The Independent Calculation is performed by the Core Council Risk Advisor (BA Labs) ^[16], while the Final Calculation is produced by Core GovOps ^[17].

• Disputed Amounts: If deviation exceeds 1%, Core GovOps works with all parties to resolve. Prime Agents can also submit a Dispute Notice within 5 days if they believe the calculation is incorrect.

How Settlement is Executed

The settlement mechanism uses a specific on-chain pattern. Here is the flow from the MSC #2 executive spell code, which established the _takeAllocatorPayment pattern used in all subsequent settlements ^[12]:

flowchart LR
    A["1) takePayment"] --> B["2) suck + grab"] --> C["3) transferUsds"]

• Settlement Steps:

• Real Example - MSC #4 (November-December 2025) ^[13]:

• Why Grove received $0 profit: Grove's yields likely matched or fell below the Base Rate. Since Instance Profit = max(Revenue - Expense, 0), and Expense is based on the Base Rate, earning at or below the Base Rate results in zero profit.

The On-Chain Mechanism in Detail

function _takeAllocatorPayment(address vault, uint256 wad) internal {
    bytes32 ilk = AllocatorVaultLike(vault).ilk();
    uint256 rate = JugAbstract(MCD_JUG).drip(ilk);
    uint256 dart = ((wad * RAY - 1) / rate) + 1;

    // Step 1: Create sin[vow] AND dai[vow] simultaneously
    VatAbstract(MCD_VAT).suck(MCD_VOW, MCD_VOW, dart * rate);

    // Step 2: Increase vault debt, cancel out the sin
    VatAbstract(MCD_VAT).grab(ilk, vault, address(0), MCD_VOW, 0, int256(dart));
}

• Key insight: Spark cannot unilaterally extract revenue. The total revenue is first minted as debt to the vault and collected by Sky, then Spark's profit share is returned via a separate transfer ^[6][7].

Part 4: Understanding the Rates

Three interconnected rates govern the economics of Spark's capital allocation: the Base Rate, the Sky Savings Rate (SSR), and the Agent Credit Line Borrow Rate. Understanding their relationship explains how Sky's revenue is generated and distributed.

The Rate Hierarchy

flowchart TB
    BASE["Base Rate"] --> SSR
    BASE --> BorrowRate["Borrow Rate"]
    BASE --> AgentRate["Agent Rate"]

• Legend — Rates:

Why This Matters for Revenue

The spread between what Spark earns and what it pays to Sky is Spark's profit margin:

flowchart LR
    Yield --> Cost["Cost (5%)"] --> Margin

• Spark's Profit Margin by Instance:

*Negative margins are floored at zero — Spark absorbs underperformance.

Part 5: Actively Stabilizing Collateral (ASC)

Prime Agents must maintain liquidity for peg stability ^[11]:

Minimum Requirement

"Prime Agents must maintain at least 5% of their Collateral Portfolio in Actively Stabilizing Collateral."

What Counts as ASC

• Resting ASC (provides buy support at ≥0.999 USDS/USD):

  • USDC in LitePSM
  • USDC in PSM3 (Base, Arbitrum, Unichain, Optimism)
  • Cash stablecoins in Curve (paired with USDS)
  • USDC in GUNI pools

• Latent ASC (convertible to Resting within 15 minutes):

  • Cash stablecoins in SparkLend, Aave, Morpho
  • Cash stablecoins in Curve/Uniswap (not USDS-paired)
  • Cash stablecoins in Prime ALM Proxy ← Important!

ASC Tracking

flowchart LR
    PORTFOLIO["Portfolio: $2B"] --> REQ["5% Required: $100M"]
    REQ --> ACTUAL["Actual ASC: $130M ✓"]

• Example breakdown: PSM3 $80M + ALM Proxy $30M + Curve $20M = $130M total ASC

• Penalty for non-compliance: Low Yield ASC Penalty reduces Adjusted Profit.

Part 6: The Surplus Buffer Role

The Surplus Buffer is central to understanding settlements ^[4]:

flowchart LR
    Inflows --> Buffer["Surplus Buffer"] --> Outflows

• Surplus Buffer Flows:

• Key insight for settlements: At settlement time, the Executive Vote executes a two-step on-chain mechanism that increases the allocator vault's debt while simultaneously crediting the Surplus Buffer. Then, the Prime's profit share is sent from the Surplus Buffer to their SubProxy. See Part 3 for the detailed technical mechanism.

Note that certain assets in Spark's portfolio are classified as Sky Direct Exposures — assets Sky Protocol owns but implements through Prime Agent Allocation Systems (e.g., PSM USDC, Curve pool positions). Prime Agents bear no borrow rate on these, and all revenue flows to the Surplus Buffer.

Part 7: Complete End-to-End Flow

Here is the complete lifecycle visualization:

flowchart LR
    A["1) Draw"] --> B["2) Deploy"] --> C["3) Earn"] --> D["4) Verify"] --> E["5) Settle"]

• End-to-End Lifecycle:

• Settlement Phase Roadmap ^[2]:

In Phase 3, the Base Rate will accrue automatically on-chain, and Executive Votes will only be needed to settle remaining amounts (demand-side primitives, penalties, Agent Rate).

Part 8: Comparison with Other Primes

Spark is one of several Prime Agents in the Sky ecosystem, each using the same Allocation System Primitive infrastructure but targeting different markets. All share the AllocatorVault → AllocatorBuffer → ALM Controller architecture and participate in the same Monthly Settlement Cycle.

While Spark focuses on DeFi lending, other Primes use the same infrastructure differently:

flowchart LR
    Core["Shared Core"] --> Spark
    Core --> Grove
    Core --> Keel

• Shared Infrastructure: All Primes use AllocatorVault → AllocatorBuffer → ALM Controller

Key Takeaways

1. Credit Line, Not Grant: Sky provides a credit line at the Agent Credit Line Borrow Rate. Spark must generate yields above this rate to profit.

2. Zero Stability Fees, But Debt IS Minted: In Stage 1, Prime Allocator Vaults have zero stability fees (set via SP-BEAM). Revenue is calculated off-chain, and at settlement time, debt IS minted on the vault via vat.suck() + vat.grab() to collect revenue.

3. Two-Step Settlement: First, total revenue is minted as debt and goes to Surplus Buffer. Then, the Prime's profit share is returned to their SubProxy.

4. Revenue Split is Formulaic: Sky gets (Total Revenue - Adjusted Profit), Spark keeps Adjusted Profit.

5. Three-Party Verification: Initial Calculation (Operational Executor Agent) + Independent Calculation (Core Council Risk Advisor) + Final Calculation (Core GovOps). Allowed deviation is 1%.

6. Surplus Buffer is Central: All settlements flow through it - revenue collected first, then profit returned.

7. Rate Limits Protect Everyone: The ALM Controller's rate limits prevent rapid capital movements that could destabilize either party.

8. ASC Requirements: 5% minimum in Actively Stabilizing Collateral ensures USDS peg stability.

9. Instance Profit Floors at Zero: If an instance earns below the Base Rate, Spark absorbs that loss — Sky never receives negative profit from an instance. Exception: USDT and pyUSD have full P&L treatment (both profits and losses).

Frequently Asked Questions

Q: Can Spark withdraw revenue without governance approval?

No. All settlements require Executive Vote approval. Spark cannot unilaterally extract funds.

Q: Why not just use on-chain stability fees like regular vaults?

In Stage 1, stability fees are set to zero via the SP-BEAM so that all value transfers happen through Executive Votes. This enables instance-level profit calculations, transparent three-party verification, and flexible penalty adjustments — none of which are possible with purely automatic on-chain accrual. In later phases, the Base Rate will accrue automatically on-chain, and Executive Votes will handle remaining settlement components.

Q: What happens if Instance Revenue < Instance Expense?

Instance Profit is floored at zero. Spark absorbs losses on underperforming instances; they can't pass negative profit to Sky.

Q: How often does settlement happen?

Monthly in Phase 1 (current). Phase 2 formalizes monthly settlement with automated monitoring. Phase 3 targets daily settlement at 16:00 UTC.

Q: What if settlement calculations disagree?

Multiple parties calculate independently. If difference exceeds 1% (Allowed Deviation), Core GovOps resolves the dispute. Primes can also submit Dispute Notices.

Related Articles

• Spark Protocol - Overview of Spark as Sky's first Prime Agent
• Debt-Based PnL Methodology - Full computational methodology for MSC calculations
• Allocation System Primitive - The underlying infrastructure
• Sky Direct Exposures - Assets Sky owns but implements through Prime Agents
• Diamond PAU Architecture - EIP-2535 upgrade to the ALM Controller
• Direct Deposit Modules - Historical predecessor (D3M)
• Sky Savings Rate - How SSR is funded from protocol revenue
• Actively Stabilizing Collateral - ALM requirements
• Grove - Alternative allocation path for RWAs
• Keel - Solana-native capital allocation
• Obex - Incubator Prime Agent for emerging allocators

Sources

1. Sky Core Monthly Settlement Cycle - A.2.4
2. Stage 1 Simplified Profit And Loss Calculation - A.2.4.1.2.2.1.1
3. Spark Prime Agent Artifact - A.6.1.1.1.2.6.1
4. Surplus Buffer - A.3.5.1
5. Agent Credit Line Borrow Rate - A.3.1.2.5
6. Payment of Amounts Due to Prime Agents - A.2.4.1.2.1.4.1
7. Collection of Amounts Due from Prime Agents - A.2.4.1.2.1.4.2
8. Spark Liquidity Layer Documentation
9. dss-allocator GitHub Repository
10. Allocation System Primitive - A.2.2.9.1
11. Minimum Actively Stabilizing Collateral - A.3.3.2.2
12. Executive Spell Code - _takeAllocatorPayment Implementation
13. MSC #4 Settlement Summary (November and December 2025) - Sky Forum
14. Initial Calculation by Operational Executor Agent - A.2.4.1.2.1.2.1
15. Allowed Deviation - A.2.4.1.2.1.3.3
16. Independent Calculation by Core Council Risk Advisor - A.2.4.1.2.1.2.2
17. Final Calculation by Core GovOps - A.2.4.1.2.1.3