The previous posts covered the BSV coordination layer (the glue between agents and Bitcoin primitives) and its architecture. This post covers something different: a complete business engine.

Paperclip-clj is a deterministic Clojure kernel for autonomous business operations. It integrates governance, policy enforcement, billing, escrow, and blockchain settlement into one acyclic module graph. 33 tests, 54 assertions, 0 failures, 0 errors.

The Engine vs The Glue

The BSV Agent Coordination Layer was glue: it connected AI agents to BSV testnet primitives (identity, payments, OP_RETURN anchoring). Paperclip is an engine: a self-contained business runtime with its own domain model, policy engine, billing system, and execution pipeline. BSV is one adapter among many — the engine would work just as well with Stripe or PayPal.

System Architecture

The engine is organized into five layers:

Architecture Diagram

Every namespace is pure Clojure with no mutable state at the core. Side effects are isolated in the adapter layer.

The Five Gates

The engine enforces five fail-closed gates before any work executes, as specified in the original design:

Gate Module Question If failed
Payment commercial.clj Is money already deposited? No execution
Scope playbook.clj Is this request in an approved contract? Dead-letter drop
Input playbook.clj Does payload match allowed schema? Dead-letter drop
Policy policy.clj Is spend within cap, vendor approved? Rejected
Attention attention.clj Will this task waste founder focus? Bounce to agent

Only tasks that pass all five gates reach the BSV settlement adapter.

Execution Pipeline

The routing engine maps every task into one of four lanes:

Architecture Diagram

The green lane runs without any human attention. The yellow lane attempts bounded self-healing (3 retries max, no LLM repair loop). The red lane triggers BSV settlement with split payouts. Everything else drops silently into the dead-letter queue.

Billing and Income Extraction

The billing engine computes fees at three levels:

(def default-pricing
  {:pricing/workspace-monthly 499.00
   :pricing/execution-fee 0.25
   :pricing/budget-unlock-rate 0.015
   :pricing/settlement-rate 0.010
   :pricing/min-settlement-fee 1.00
   :pricing/anchor-fee-bsv 0.000005})

(defn calculate-fees [{:keys [proposal execution-count settlement-amount pricing]}]
  ;; Returns workspace-monthly, execution-fee, unlock-fee,
  ;; settlement-fee, anchor-fee, and total-variable-fee
  ...)

Revenue comes from three streams: monthly workspace retainers, usage-based execution fees, and optional BSV settlement margin. The commercial module enforces prepaid escrow — no balance means no execution.

BSV Integration

The BSV adapter generates deterministic SHA-256 digests, constructs split-payout transactions (vendor + platform fee), and anchors OP_RETURN proofs:

(defn settle-payout-with-extraction!
  [{:keys [vendor-address platform-treasury-address]} proposal billing-breakdown]
  (let [receipt-hash (canonicalize-and-hash proposal billing-breakdown)
        tx-template {:outputs [{:address vendor-address
                                :satoshis (to-sats net-vendor-payout)}
                               {:address platform-treasury-address
                                :satoshis (to-sats platform-fee)}
                               {:op-return receipt-hash}]}]
    ...))

Revenue Model

The engine is designed for a solo founder targeting $100,000/year through:

  1. One narrow workflow — pick one painful business process, encode it as a playbook
  2. Two B2B retainers at $4,500/month each = $108,000/year ARR
  3. Prepaid escrow — no credit, no net-30, no invoices
  4. Dead-letter as contract law — dropped invalid inputs are defined as correct system behavior, not support tickets

Test Results

All 33 tests pass with 54 assertions:

Module Tests What it proves
core 8 Proposal lifecycle, voting, quorum evaluation
governance 2 Approved/non-approved proposal execution
policy 10 Spending limits, risk tiers, agent intent, founder override
billing 3 Fee calculation, extraction model, pricing
playbook 6 Contract validation, schema enforcement, immutable contracts
attention 4 Risk scoring, triage lanes, escalation thresholds

The engine is at gitlab.com/nurazhar/paperclip-clj. Next step: deploy for one paying client and prove the loop closes.