Reading the Ledger · Outset Systems

What it is

The Ledger trades a simulated book against real market prices. Fills are simulated at real quotes. No broker is connected and no real money moves, in any code path. Nothing here is investment advice, and nothing here should be used to decide a real trade.

It trades long only, in stocks, ETFs and, where switched on, crypto.

The pages

Trading Lab

The agents, the council, the systematic strategies, the fund summary, the equity curve, scout signals, price paths, open positions, trade history (with a CSV download) and every agent decision.

Intelligence

Insider and congressional activity: clusters of buying, weekly flow and a scorecard for how clusters have done afterwards.

Markets

Relative performance, the watchlist, CFTC positioning and economic series.

The chain

One pipeline run is called a chain, and everything it decides shares an id.

Step	Model?	What it does
Scout	Yes	Reads news against a theme list and maps it to tickers. If it fails, the chain carries on.
Research	Yes	Ranks screened candidates.
Strategy	Yes	Turns candidates into orders. Trend, value and macro members vote, joined by any rule-based strategies that are on.
Risk	No	Applies the rules in order. Any veto ends it; a resize only ever shrinks.
Execution	No	Fills approved orders on the paper broker.

The demo's Scout runs overnight, the full chain after the market opens and Execution after the close. Each agent shows its next run on the Trading Lab page. Prices update every 15 minutes.

The risk rules

The defaults, in the app's own terms. A self-hosted install can change them.

One position: at most 10% of the book.
One sector: at most 30%. Crypto: at most 25%.
Keep at least 20% in cash, and hold at most 12 positions.
Risk per trade: at most 2%. Turnover: at most 25% a day.
A 15% drawdown halts new buys.
Companies under $2 billion in market value are refused.
No single idea source gets more than 20% of the book.

The six strategies

These are plain rules, scored the same way as the model. The first five take no model at any step; each can be switched on or off, and the demo runs some of them, not all.

Name	The rule
insider_cluster	Three or more insiders buy within 30 days, none sell, and the stock is in an uptrend.
cot_extreme	Buy at or under the 10th percentile of positioning, exit at or over the 90th.
dual_momentum	Once a month, hold the top two ETFs by 12-month momentum, otherwise cash.
donchian_breakout	Buy a 20-day closing high, with a 12% trailing stop.
rsi2_reversion	Buy RSI(2) at 10 or below when above a rising 200-day average. Exit at the 5-day average or after 7 days.
sentiment_drift	Improving mood plus a corroborated scout signal, held 10 days. Its inputs are model-made; the rule is not.

The figures

Figure	How it is worked out
Total equity	Cash plus positions. A position with no current quote counts at cost.
Cash	Net deposits, plus or minus trades, minus fees.
Total P&L	Equity minus net deposits.
Realised	Gains on sells after the sell fee. Fees paid on buys are not in this figure.
Unrealised	Open positions that have a current quote.
Fees paid	Fees on every fill.
Max drawdown	Worst fall from a peak, over the latest 2,000 snapshots.
Win rate	Share of sell fills that made money, partial sells included.
vs SPY	The book's return minus what SPY would have returned on the same deposits, in percentage points.

The percentage under Total P&L is a simple return on net deposits.

The scorecards

Research scorecard

Compares the model with a momentum rule. A stock counts as liked at a score of 50 or more. Alpha is its return since ranking minus SPY's. Read the two rows where they disagree: LLM only and Rule only. That is where the model is adding something, or not. Declines show the share of turned-down ideas that went on to lag SPY.

Source scorecard

One row per idea source: how many were proposed, approved and filled, average alpha, hit rate and sample size. It is for reading only and tunes nothing by itself. Watch the sample size before believing a hit rate.

Costs

Every fill pays these, so paper results are not flattered by free trades.

Slippage: 5 basis points, always against you.
Stocks and ETFs: 0.10% a side, 1 cent minimum, plus the US regulatory sell fees.
Crypto: 0.6%.

Glossary

Word	Meaning
Alpha	Return minus SPY's over the same period, in percentage points.
Hit rate	Share of ideas with positive alpha.
Basis point	0.01%.
Backer	A council member supporting a proposal. One, two or three backers size it at 50%, 75% or 100%.
Dead band	The ±0.20 zone where a council vote counts as a hold.
Veto	Risk refused the order.
Resize	Risk let it through smaller.
Convergence	Congress and company insiders both bought the same stock.
Disclosure lag	Congressional trades can be reported up to 45 days late, so the data is old by design.
COT percentile	Where positioning sits against its own history. 50 is the median.

Questions

Is any of this real money?

No. There is no broker and no path for money to move.

Should I copy these trades?

No. This is a research project, not advice. It exists to measure things, and a lot of what it measures will be wrong.

Why did nothing fill overnight?

Stock orders placed while the market is closed wait in a queue, and expire after 72 hours.

Why don't Realised and Unrealised add up to Total P&L?

Fees paid on buys sit outside Realised, and positions with no current quote sit outside Unrealised.

Why is a price move blank?

Some price bars are missing for that symbol. The page leaves a gap instead of guessing.

Why does a button say to sign in?

The demo is read-only. Anything that would change the book, the settings or the agents is switched off. Tabs, filters, ranges and the CSV download work.

Run your own

The Ledger runs from Docker Compose on your own machine. The source and install guide open with its public release.