~/djzs/builders $ cat pitch.txt
Your agent is about to move money. You already have spend caps. A budget clamp is not a firewall; it limits the size of the mistake, never whether the position should exist. Before the order leaves, your agent should be able to prove (to you, to governance, to a dispute six months from now) that the reasoning was sound at the moment it acted. DJZS produces that proof: a verdict computed by frozen code, not improvised by a model, with a receipt a third party can recompute.
Your transaction pipeline (simulation, threat scanning, smart transactions) already answers is this transaction safe to sign? It cannot answer the other question: is the reasoning behind this position sound? A transaction can clear every check and still act on a thesis that is unfalsifiable, unsourced, or pure narrative.
capital action │ ▼ GATE 1 · DJZS audits the THESIS (why) │ returns PASS / WAIT / FAIL ├─ WAIT / FAIL ─▶ HALT · nothing executes │ EXECUTE │ ▼ GATE 2 · WALLET audits the TRANSACTION (what) ▼ on-chain
The gate clears the logic; the position is yours.
An LLM extracts facts from the intent; it never returns a verdict. A frozen deterministic engine scores the struct: pure code, no model, no clock. Same struct, same verdict, same hash, every run. PASS proceeds. FAIL stops. Unknown drains to WAIT and halts.
Weights are hash-locked exported constants. A change is a versioned re-derive, never a hot patch. That immutability is what lets a third party reproduce verdict_hash byte-for-byte from the published ruleset.
Every in-scope audit anchors a ProofOfLogic certificate to Irys mainnet. Retrievable by id, free to read via query_pol_certificates. The record outlives the agent, the framework, and the operator.
Pass an agent_address and the verdict writes to DJZSLogicTrustScore on Base. Anyone can read totalAudits · failRate · latest verdict via query_agent_trust. Governance can gate on it.
The refusal is the product. An auditor that only ever says yes is not an auditor.
intent_hash any third party recomputes from the intent alone, and writes an EAS attestation on Base under the DJZS schema onto the paying agent’s record. A Safe Guard or Zodiac module can require it. Contract at djzs.ai/ruleset; recompute any receipt at djzs.ai/verify.verify_perp_trade live at 2.00 USDC on the same rail: DJZS-LF v1.2, four codes live (E01, I01, X01, S01). A position stated with no thesis fails; a prediction-market bet sent here is refused free.A DJZS PASS certifies well-formed reasoning at a moment, never that the narrative is true or the trade wins. It does not check that a quoted price is still live or that a source is right. Those remain yours.
Disclosed as roadmap, not sold as done.
As an MCP tool:
claude mcp add --transport http djzs-trust https://mcp.djzs.ai/mcp
As plain HTTP x402: any x402 client pays this (Base MCP, MetaMask Agent Wallet, x402-axios). An unpaid request returns the 402 challenge and settles nothing:
curl -s -X POST https://mcp.djzs.ai/x402/verify_pm_trade \
-H 'content-type: application/json' \
-d '{"intent":{"market":"…","side":"YES","thesis":"…"}}'
Perpetual or spot trade: same rail, same price; a position stated with no thesis fails (DJZS-S01), a prediction-market bet is refused free:
curl -s -X POST https://mcp.djzs.ai/x402/verify_perp_trade \
-H 'content-type: application/json' \
-d '{"intent":{"instrument":"BTC-PERP","side":"LONG","leverage":5,"entry":78900,"stop_loss":76200,"take_profit":83000,"size_usd":3000,"venue":"Binance","thesis":"…"}}'
As a working pattern: djzs-pm-gate audits a thesis and creates the order only on a clean in-scope PASS. Fork it, point it at your venue.
Discovery contract at GET /openapi.json · payment config at GET /health/x402 · machine summary at /llms.txt. All free.
Audit before act.