docket / v0.1 / base mainnet

A portfolio manager
that shows its work.

Three specialist agents debate every rebalance on Base. Every threshold is a number. The Risk Manager can veto. You pay only when the agent acts — via x402.

Read the source

live demo available post-hackathon

~0.003 USDC per decision · no subscription · no API key

the problem

Why most trading bots fail you.

Three reasons retail capital consistently loses the agent game — and the thing they all share.

01

The market doesn't sleep. You do.

Crypto runs 24/7. Human attention doesn't. By the time you check the chart, the move has already happened.

02

Signal in, trade out.

Most AI trading bots give you a verdict with no derivation. You can't audit reasoning you never see. You're trusting a vibe.

03

Tuning means editing code.

Want a tighter risk band? Open the repo. Change a number. Redeploy. No product should ask its user to be a committer.

Retail crypto traders have long underperformed buy-and-hold. The literature is consistent; the cause is not the market, it's the interface.

the solution

Three specialists. One decision. Full math.

Each agent runs a textbook quant strategy — not a black box. Weighted consensus must clear 60% to execute. Risk Manager vetoes override everything.

01

Market Structure Analyst

Z-score over Chainlink.

Pulls 20 rounds from the Chainlink ETH/USD feed. Computes rolling mean, standard deviation, and momentum. Flags moves more than 1.5σ from the mean.

feed
chainlink
price
$3,842.11
mean
$3,901.22
z-score
-1.82σ
momentum
+0.31
canon

Statistical mean reversion

Bollinger Bands / Thorp-era stat arb. Momentum filter guards the falling-knife case.

data: chainlink oracle

02

Liquidity Analyst

Slippage elasticity via Elsa.

Probes the book at three sizes — 10, 100, 1,000 USDC. Measures how slippage scales. Vetoes fills when the curve goes non-linear.

10 USDC
0.01%
100 USDC
0.08%
1,000 USDC
0.91%
elasticity
1.14
band
linear
canon

Kyle's Lambda (price-impact elasticity)

Kyle 1985 market microstructure. Super-linear impact signals a distressed book.

data: elsa live probes

03

Portfolio Risk Manager

Drift, cooldown, position limits. Veto power.

Holds the rulebook. Enforces minimum time between actions. Caps any single asset at a hard position ceiling. Can override the other two.

drift
4.2%
cooldown
6m 12s
reserve
1.0%
max pos
80%
veto
ARMED
canon

Threshold-band rebalancing to constant-mix

Perold–Sharpe 1988. Drift floor is justified by round-trip cost, not taste.

data: wallet state

how it works

Observe Debate Act.

  1. 01

    Observe

    Chainlink round fires. Wallet state read. Slippage probed.

    base mainnet · ~60s cadence

  2. 02

    Debate

    Three agents output structured opinions — signal, confidence, x402 receipt. Aggregator weights votes.

    weight: 0.5 / 0.3 / 0.2

  3. 03

    Act

    Consensus ≥ 60, no veto, drift > 3% → Elsa swap. Receipt hashed. Audit trail updated.

    elsa x402 · receipt on-chain

personalization

Match the agent to your temperament.

One profile, different thresholds. No config files, no redeploy — the agent compiles to your chosen risk.

Switch profiles any time. Every x402 receipt records the profile in force at decision time.

built on

  • Base logo

    Mainnet. L2 settlement, low fees, 2-second blocks.

  • Chainlink logo

    Price oracle. One source of truth across the debate.

  • x402 logo

    HTTP-native agent payments. No API keys, no subscriptions.

  • Coinbase logo

    x402 originates here — the standard for agent-native payments.

  • Elsa logo

    x402-native swap execution.

Watch the Docket debate.

Every rebalance is a case opened, argued, ruled on, closed. Every trade on the Docket.

Read the source

live demo available post-hackathon

built for Elsa Agentic Hyperthon · Apr 17, 2026 · base mainnet