P100 — COMPOSITE SIGNAL

THE MARKET,
REDUCED TO
ONE NEEDLE.

Twenty-seven weighted on-chain & macro components, three index layers, one composite in [-1, +1]. Computed by an autonomous agent. Executed through OKX ATK. Custodied by an Avalanche vault the agent cannot steal from.

COMPOSITE +0.0043 NEUTRAL

◀ −1.00 … −0.50 SELL −0.50 … +0.50 NEUTRAL +0.50 … +1.00 BUY ▶

DAILY BRIEF — 23:15 TRT

The needle moves.
Your inbox knows first.

One email, every night at 23:15 TRT: composite signal, all three indices, and the agent's reasoning — before the next session opens.

P200 — INDEX FAMILY

Three clocks, one machine.

▸ DRILL DOWN — all 27 components, raw · score · weight · impact

P300 — WHAT THE AGENT DID

Decision log, unedited.

Last 5 records from vault/logs/agent_decisions.jsonl + atk/logs/decisions.jsonl — embedded verbatim.

P400 — VAULT SECURITY MODEL

The agent proposes.
It can never withdraw.

RULE 01 Agent key can only call whitelisted targets × whitelisted function selectors. Everything else reverts.

RULE 02 Agent can never call the asset contract directly, and withdraw() from the agent key reverts — "insufficient shares".

RULE 03 Daily movement limit, enforced both per-tx and cumulatively. Breach → "daily limit exceeded".

RULE 04 Deposits & withdrawals stay permissionless for users — the cage is around the agent, not the money.

RULE 05 Signal → exposure mapping is deterministic: logistic(k=4.0) with Schmitt hysteresis (enter ±0.50, exit ±0.40). Same input, same proposal, every time.

RULE 06 Trades are queued behind a 900s timelock — the guardian can veto before execution. Proven live: see the E2E tx below.

CONTRACT

NETWORK

ADDRESS …

TIMELOCK…

TVL (DEPOSIT)…

19/19 security tests passing
revert paths verified on-chain semantics

P500 — ATK EXECUTION PROOF

Real MCP. Real OKX price.

50tools exposed by
okx-trade-mcp 1.4.8
modules: market · spot · account

0.033%live price cross-check
ATK 81,479.2 vs engine 81,506.0
tolerance 2% — PASS

4tools called over MCP stdio
ticker · place_order
positions · balance

engine ──▶ snapshot.json ──▶ signal_to_trade.py ──▶ AtkBridge ──▶ npx @okx_ai/okx-trade-mcp

composite > +0.5 → BUY plan · composite < −0.5 → SELL plan · else NEUTRAL (no order)

tools/call market_get_ticker BTC-USDT ──▶ ✔ real price over MCP transport · auth calls fail closed (AUTH_MISSING, expected in keyless sandbox)

fallback chain: MCP → OKX V5 REST (HMAC, x-simulated-trading) → mock (marked simulated:true)

P600 — DAILY NEWSLETTER

Every morning, a printout.

… plus full component tables for all three layers. Delivered daily to Signal Pro subscribers.

The agent writes what it measured — not what it feels. Each issue is generated from the same snapshot.json that drives the gauge on P100.

Sample above is the real header of newsletter-2026-10-17.md.

P700 — PRICING

Pay for signal. Or pay for outcome.

TIERPRICEWHAT YOU GET

FREE$0Live gauge (P100) + weekly summary

SIGNAL PRO$50/moDaily newsletter + full 27-component drill-down

AUTO-VAULT10% perfAgent-managed exposure, high-water mark, you keep the keys

Benchmarks: Alpha Factory $49 · Telegram signal groups $42–290 · Enzyme 0–20% · Bybit copy-trading 10–15%.

P800 — JURY EVIDENCE

Receipts, not slides.

TRACK — OKX (10%)

ATK integration proof — real MCP stdio session, 50 tools listed, live BTC ticker, signed order intent: quant-agent/atk/INTEGRATION_PROOF.md
Execution log (JSON-RPC transcripts, transport flagged per call): quant-agent/atk/logs/decisions.jsonl · demo_log.json
Price cross-check 0.033% deviation vs independent engine source (CryptoQuant)

TRACK — AVALANCHE (10%)

QuantVault.sol — agent-caged custody, 19/19 security tests: quant-agent/vault/tests/last_results.json
Snowtrace (Fuji) contract: 0xd4D6…Fa50 ▸ — deployed 2026-10-17, timelock 900s, 2,000 tUSDC deposited
E2E trade executed on-chain: executeTrade tx 0x7a98…760c ▸ (queue → timelock → execute)
Deterministic proposals on file: vault/proposals/proposal_20261017T142704Z-be3a2a.json

THE MARKET,REDUCED TOONE NEEDLE.

The needle moves.Your inbox knows first.