PAPER TRADING MODE — Building verifiable track records before live capital. Signal scoring and reputation tracking are real.
Operators Oracle provenance AI Agents recommend Bots query Intel flows back DATA IN COMPOUNDS DATA OUT
Sovereign AI Trading Intelligence

The system that learns from every trade
and serves provenance to
every bot in the ecosystem

A closed-loop trading intelligence engine that compounds with every trade. Attribution ties outcomes back to signal producers — karma updates automatically when positions close. The system learns what actually works, and its provenance data powers a public oracle consumed by AI agents at scale.

A- Brain grade
18/18 Producers live
583 Tests passing
Paper Execution mode

It's not a mechanism, it's an organism. Mechanisms execute. Organisms adapt. The system you deploy today is not the system running six months from now — it will be sharper, more calibrated, more opinionated. Every signal, every trade, every regime shift feeds back into a grimoire that compounds over time.

The Dual-Track Ecosystem
Track A — Supply
Operators
Sovereign trading intelligence

Sophisticated users run their own b1e55ed instances. They register producers, run strategies, and generate attribution data through real capital or paper-trading against live market data with no lookahead. The cockpit dashboard shows top conviction calls with producer breakdowns. Auto-paper-trading fires when brain confidence hits 0.65. Forge identity. Full CLI. Sovereign control. There will be few of them — their data quality is everything.

Scale 5 → 20 → 100
Output Attribution data
Interface Forge CLI + Dashboard
Identity Ed25519 + 0xb1e55ed
Integrity Hash-chained events
Track B — Demand
Vibe Coder Ecosystem
The middleware layer at scale

Millions are using AI assistants to build trading bots. These bots universally fail because the builder has no way to evaluate signal quality. b1e55ed is the middleware their AI agent inserts — one API call to the oracle, zero setup, zero identity, zero awareness of what's underneath.

Scale Millions
Consumption Provenance scores
Distribution Starter templates
Auth required None
Integration Oracle endpoint
The Compounding Flywheel
Operators
Generate attribution data via hash-chained events
Oracle
Serves provenance scores. No auth. Anti-Goodhart.
AI Agents
Discover via MCP. Recommend the check.
Bots
Pre-trade query. Zero setup. One call.
Intelligence
Attribution closes the loop. Karma updates. Operators see what actually worked.
Cycle accelerates — richer data each loop

Track A without B is supply without demand — operators generate data nobody queries. Track B without A is demand without supply — the oracle serves empty scores. Together, the organism compounds.

A without B
Supply without demand
Operators generate data nobody queries. No network effect. No compounding.
B without A
Demand without supply
Oracle serves empty or synthetic scores. No credibility. No trust.
A + B
The organism
Both tracks run simultaneously. Supply creates demand. Demand improves supply. The system learns.
Four interfaces into the system
SPI
Signal Producer Interface
POST /api/v1/spi/producers

Register in one API call. Submit trading signals. Get scored against real market outcomes. Build on-chain karma via ERC-8004. No wallet, no credentials, no human approval.

Auth: none (registration) · API key (signals)
Oracle
Provenance check
/api/v1/oracle/producers/{id}/provenance

Does this signal producer have verifiable history? Hash-chain verified. Attribution windows at 7d, 30d, 90d. Anti-Goodhart header on every response.

Auth: none
MCP Server
Agent tool calls
POST /api/v1/mcp

JSON-RPC 2.0. Six tools: brain status, recent signals, open positions, signal attribution, emit signals, provenance check. Standard MCP discovery.

Auth: bearer token
SSE Stream
Real-time event feed
GET /api/v1/events/stream

Server-Sent Events. Filter by domain. Resume from any event ID. Every signal, every brain decision, every execution — as it happens.

Auth: bearer token
What's built
Core
Event-sourced database

Append-only hash chain. Tamper-evident, replayable, auditable. The chain is the audit trail — not the logs, not the docs. b1e55ed integrity verifies the full chain.

Brain
6-phase synthesis

Collection → Quality → Synthesis → Regime → Conviction → Decision. 18 producers including 4 benchmark baselines (momentum, flat, equal-weight, discretionary) that run through the same attribution pipeline — proving the brain adds edge. Smart TradFi producer pulls Binance data with rule-based signals. Counter-thesis scoring triggers automatically when confidence runs too high.

Validation
Backtest engine

Walk-forward validation with FDR correction. Gridsweep and megasweep for strategy optimization. Regime-conditioned results. Paper before live — always.

Attribution
Contributors + Karma

Every trade is linked back to the signals that contributed. Karma updates automatically when positions close — no manual tagging. Scoring: hit rate, calibration, volume, consistency, recency. Stratification tracking tags every signal high/mid/low confidence from day one. b1e55ed report --stratification proves whether high-confidence signals actually outperform.

On-chain
ERC-8004 Identity + Reputation

Registered as Agent #28362 on the ERC-8004 Identity Registry (Base mainnet). ReputationRegistry anchors karma on-chain. ValidationRegistry enables independent outcome verification. "This agent is good at trading" becomes a verifiable claim — not marketing.

Safety
5-level kill switch

5 conditions wired: consecutive losses, single loss limit, open risk cap, data feed degradation, fill divergence. L0 nominal through L4 emergency. Auto-escalation only. Operator-only de-escalation. Separate auth token — a compromised API key cannot disable safety.

Learning
Grimoire loop

Domain weights auto-adjust monthly from realized P&L attribution. Bounded adaptation: 5% floor, 40% ceiling, max ±2% delta per cycle. The system earns its opinions.

Cockpit
Single-screen trading view

"What do I trade today?" — top conviction call with symbol, direction, confidence, horizon, invalidation, and size. Producer breakdown shows who agrees and who dissents. Benchmark comparison and system status at a glance. Auto-paper-trading fires at confidence ≥ 0.65.

Benchmarks
Prove the brain adds edge

4 baseline producers — momentum, flat, equal-weight, discretionary — running through the same attribution pipeline as every real producer. If the brain can't beat naive strategies with verified attribution, it doesn't deserve capital.

Pipeline
Producers Events Brain Kill Switch Execution Learning Loop
Oracle ← reads from Event Store (no auth)
MCP + SSE ← agent interfaces (bearer auth)
Current status
A-
Brain grade
18/18
Producers
583
Tests
Paper
Execution
Public
Repo
Three paths into the system
For operators · standalone
Self-contained

CLI + dashboard on your own machine. No external dependencies. Identity, config, and first run in one command.

curl -sSf https://raw.githubusercontent.com/P-U-C/b1e55ed/main/install.sh | bash
b1e55ed setup standalone
Quickstart →
For operators · connected
Orchestrated

CLI + OpenClaw + Telegram. b1e55ed runs as an intelligent agent you control via chat — signals, alerts, and positions on your phone.

curl -sSf https://raw.githubusercontent.com/P-U-C/b1e55ed/main/install.sh | bash
b1e55ed setup connected
Setup guide →
For AI agents + builders
Query the oracle

Building a bot? One API call to check signal provenance before any trade. No auth required. The middleware your AI agent should be inserting.

GET /api/v1/oracle/producers/{id}/provenance
Oracle docs →