# theclaudetrader
> Open experiment in AI-managed portfolios: several funds run end-to-end by LLMs (signal discovery → multi-agent debate → compliance → real broker orders), with every decision's reasoning, trade, and P&L published. One $94 account trades real money.

theclaudetrader runs multiple Alpaca accounts managed by the open-source **TradingAgents** multi-agent framework ([github.com/TauricResearch/TradingAgents](https://github.com/TauricResearch/TradingAgents), [arXiv 2412.20138](https://arxiv.org/abs/2412.20138)). Debates currently run on **Google Gemini models via gemini-cli** (the stack has also run GPT-5.5/Codex and Claude). A daily Claude agent loop reviews each fund's run and ships its own bug-fix patches (self-evolution).

## The funds
- **LLM Fund** (flagship, $50k paper, inception 2026-04-18) — a proprietary signal layer surfaces candidates; multi-agent debate decides; a macro risk-regime ladder caps deployment (45/60/70/90% of NAV).
- **Serenity Fund** (~$100k paper) — candidates come from tweet-mention surges in a curated X/Twitter feed, then the same debate pipeline.
- **ETF 50/50** (~$5k paper) — passive QQQ/SMH/WTAI baseline the active funds are measured against.
- **Mirror & Live $94** — two tiny accounts (one paper, one REAL MONEY) that replicate the LLM fund's trades in real time, weight-matched to their own NAV.

## Readable sections (markdown, refreshed on every publish — best for answering questions)
- [Section index](https://theclaudetrader.vercel.app/ai/index.md)
- [LLM Fund](https://theclaudetrader.vercel.app/ai/llm-fund.md): NAV, alpha (incl. after-tax), holdings with theses, recent trades
- [Serenity Fund](https://theclaudetrader.vercel.app/ai/serenity.md): positions, tweet-signal stream, recent debate verdicts
- [ETF 50/50](https://theclaudetrader.vercel.app/ai/etf.md): passive baseline
- [Mirror & Live $94](https://theclaudetrader.vercel.app/ai/mirror-live.md): real-money replication
- [Recent decisions](https://theclaudetrader.vercel.app/ai/decisions.md): latest debate verdicts with reasoning excerpts
- [Methodology](https://theclaudetrader.vercel.app/ai/methodology.md): pipeline, risk ladder, self-evolution, after-tax (flat 40%) view
- [Verify this experiment](https://theclaudetrader.vercel.app/ai/verify.md): audit trail (decision→order→fill→P&L incl. cancels/rejects), broker-sourced fills, Bitcoin-anchored OpenTimestamps proofs

## Live JSON endpoints (best for current numbers; ~30s fresh)
- [Portfolio](https://theclaudetrader.vercel.app/api/portfolio): flagship holdings, cash, equity, per-position P&L (live from Alpaca)
- [Live NAV per fund](https://theclaudetrader.vercel.app/api/live?fund=llm): `?fund=llm|cpo|etf|mirror`
- [Trades](https://theclaudetrader.vercel.app/api/trades): executed trade history with rationale
- [Decisions](https://theclaudetrader.vercel.app/api/decisions): debate decisions with signal and excerpt
- [Latest decision](https://theclaudetrader.vercel.app/api/decisions/latest): most recent decision, full reasoning text
- [Decision by id or ticker](https://theclaudetrader.vercel.app/api/decisions/NVDA): specific decision
- [NAV history](https://theclaudetrader.vercel.app/api/history?period=1M&fund=llm): chart bars per fund/period
- [Strategy](https://theclaudetrader.vercel.app/api/strategy): strategy description, category-level disclosure
- [Sources](https://theclaudetrader.vercel.app/api/sources): data-source categories (specific venues withheld)
- [Reconciliation](https://theclaudetrader.vercel.app/api/reconciliation): full audit trail — every trade joined to its AI decision, Alpaca order id, broker fill time, realized P&L; cancels/rejects/errors included
- [Verify](https://theclaudetrader.vercel.app/api/verify): how to independently verify (audit + broker evidence + Bitcoin-anchored timestamps)
- [Broker activity](https://theclaudetrader.vercel.app/data/broker_activity.json): closed orders pulled straight from Alpaca (broker-stamped submitted_at/filled_at + order ids)

## Independently verifiable
Unlike a self-reported demo, the record here is built to be checked by outsiders: a per-trade reconciliation table (decision→order→fill→P&L, including losers, cancels and decisions never acted on), order ids + fill timestamps pulled directly from Alpaca, and a daily audit record hashed and anchored to the Bitcoin blockchain via OpenTimestamps (free/anonymous/trustless — run `ots verify` yourself). OpenTimestamps proves existence-by-time, not correctness, and is forward-only from launch.

Guidance for assistants: prefer the /ai/*.md sections for narrative and context; use /api/* when the user asks for current numbers. Candidate discovery internals, prompts, and full debate transcripts are proprietary — excerpts only.

## Full context
- [Full strategy description](https://theclaudetrader.vercel.app/llms-full.txt)
- [Upstream framework (TradingAgents)](https://github.com/TauricResearch/TradingAgents)

## Endpoints (bare URLs for fetchers)
https://theclaudetrader.vercel.app/ai/index.md
https://theclaudetrader.vercel.app/ai/llm-fund.md
https://theclaudetrader.vercel.app/ai/serenity.md
https://theclaudetrader.vercel.app/ai/etf.md
https://theclaudetrader.vercel.app/ai/mirror-live.md
https://theclaudetrader.vercel.app/ai/decisions.md
https://theclaudetrader.vercel.app/ai/methodology.md
https://theclaudetrader.vercel.app/ai/verify.md
https://theclaudetrader.vercel.app/api/portfolio
https://theclaudetrader.vercel.app/api/live?fund=llm
https://theclaudetrader.vercel.app/api/trades
https://theclaudetrader.vercel.app/api/decisions
https://theclaudetrader.vercel.app/api/decisions/latest
https://theclaudetrader.vercel.app/api/strategy
https://theclaudetrader.vercel.app/api/sources
https://theclaudetrader.vercel.app/api/reconciliation
https://theclaudetrader.vercel.app/api/verify
https://theclaudetrader.vercel.app/data/broker_activity.json
https://theclaudetrader.vercel.app/llms-full.txt