# Finly

In a 2013–2026 cost-modeled historical simulation, Finly finished **$38,629 ahead of SPY**: the same $10,000 became $106,711 with Finly and $68,082 with the S&P 500 tracker. That historical result is separate from live performance, and it is the economic case behind one autonomous AI trading strategy with two coordinated sleeves: a diversified four-fund allocation and a capped-risk SPY options sleeve. AI may study and explain a trade, but it may not control the account. Fixed, tested code chooses the direction, contracts, size, maximum loss, broker fields, and whether an order may reach Alpaca paper trading.

Paper trading follows real market prices with virtual money. SPY is a fund that tracks the S&P 500. QQQ is a fund that tracks the Nasdaq-100.

## One strategy, two coordinated sleeves

Finly runs one autonomous competition strategy with two coordinated sleeves. Its four-fund allocation keeps roughly half in QQQ, divides most of the remainder across three sector funds selected for stronger longer-term price trends, and keeps 3% in cash. The competition mix was chosen and locked before paper trading began. Technical code and evidence files retain the internal research identifier G4 for reproducibility; G4 is not a product generation or a separate strategy.

The coordinated options sleeve uses fresh SPY price momentum to choose a positive or negative short-horizon view. Option-market evidence and Qwen3-32B's reading of public Alpaca news may support, weaken, or stop it; neither may choose the sign. Fixed code compiles a one-contract bullish call spread, bearish put spread, or no trade and caps any new entry at $500 of possible loss. The two sleeves share one account while keeping separate decision and revision records.

## Historical results

In a cost-adjusted historical simulation from 2013 through 2026, a modeled **$10,000 became $106,711 with Finly's underlying four-fund research rule, before the competition's 3% cash scaling**. SPY became **$68,082** over the same dates. That is a **$38,629 historical ending-wealth difference**, with total return of **+967.11%** for Finly versus **+580.82%** for SPY after modeled 0.05% one-way trading costs.

A simpler industry version was checked on **21,218 earlier market days** from 1927 through 2007. It averaged **13.37% growth per year versus 9.48% for the market**, stayed ahead in all 21 tests that changed which trading day of the month the holdings were updated, and kept a positive advantage when the trading-cost assumption rose fivefold to 0.25%.

## Interactive options demonstration

The public demonstration shows a bullish call spread, bearish put spread, and no-trade path. One illustrative SPY plan has an exact **$366 maximum loss** and **$634 maximum gain**. Source-removal and 32 perturbation results are published as confidence diagnostics; hard account, quote, size, loss, order-binding, idempotency, and broker-read-back checks retain execution authority. No broker order or fill occurred in the demonstration.

Live options eligibility was checked across **517 sampled SPY signal windows**. On a symmetric fixed modeled quote surface, **11 cleared every alpha gate: 7 bullish call spreads and 4 bearish put spreads**. Their maximum losses were **$440–$455**, conservative modeled values after costs were **$10.08–$22.26**, and reward-to-risk ratios were **2.30–2.41**. This uses historical SPY signals and a fixed modeled quote surface, not historical option quotes or options P&L.

## Live Alpaca paper account

Finly runs in the cloud against a dedicated, verified **$100,000 Alpaca paper account**. The runner uses **Alpaca's official connection**, locks the code version allowed to trade, saves its progress before each action, and checks the broker afterward. The public dashboard publishes account value, positions, risk totals, and the latest decision without exposing credentials, account numbers, or order identifiers.

At exactly 4:00 p.m. ET on September 2, Finly was **up $141.24** while a same-$100,000 SPY price benchmark was **down $284.76**. Finly therefore finished **$426.00 ahead of SPY**, after **15 ETF fill events** and with no deposits or withdrawals.

During the September 2 session, the options sleeve completed **24 live evaluation cycles**. Fourteen candidates failed deterministic certification, six lacked enough model evidence, and four arrived after the entry window closed. Every cycle ended `NO_TRADE`, so no options order or fill occurred and the sleeve added **$0 of new options risk**.

The public verification run discovered **827 automated tests: 825 passed, 0 failed, and 2 optional private-ledger checks were skipped**. They cover the portfolio calculation, causal historical timing, trading costs, bullish and bearish options paths, ordinary-market calibration, option payoff, data freshness, risk limits, order construction, bounded cancel/reprice exits, account checks, lost acknowledgements, restart recovery, encrypted state, public-data filtering, and competition scoring.

## Why it matters

Most trading demos show either an attractive chart or an AI recommendation. Finly joins a diversified allocation, an AI-assisted options decision, exact account checks, and broker-shaped orders in one inspectable strategy. A persuasive AI explanation cannot silently become a larger position or a different order.

## Where to start

Start with the live dashboard, then replay the “evidence agrees” and “evidence conflicts” paths in the interactive demonstration.

- Product and live account: https://owlsowo.github.io/finly-bot/
- Source repository: https://github.com/owlsowo/finly-bot
- One-page proposal: https://owlsowo.github.io/finly-bot/judge/Finly_Judge_Brief.pdf
- Mathematical technical note: https://owlsowo.github.io/finly-bot/judge/Finly_Technical_Proposal.pdf
- Engineering appendix: https://owlsowo.github.io/finly-bot/judge/Finly_Engineering_Appendix.pdf
- Presentation: https://owlsowo.github.io/finly-bot/judge/Finly_Consulting_Deck.pdf
- Product video: https://owlsowo.github.io/finly-bot/judge/Finly_Demo_Video.mp4

Finly is an educational paper-trading project. Historical simulations are not live account results or promises of future returns. Options calibration and public examples are not broker fills unless the live dashboard links a broker receipt.

## Reproduce the public checks

From a clean clone with Node.js 26.7.0, run `npm install && npm run verify`.