Paper books and the league #

A paper book is a simulated portfolio. It has a starting balance, cash, open positions, pending orders, fills, dividends, and an equity history. It can make or lose simulated money, but it has no broker account and cannot touch real capital.

The word “paper” does not mean “assume every trade worked.” The simulator deliberately makes entry harder than reading a chart after the fact.

A signal is not a fill #

Most strategies decide after a session’s data is complete. Their orders can fill only at a later session’s open. A same-bar fill is rejected. That one rule blocks a common form of accidental look-ahead: deciding from a closing price and pretending the trade happened at an earlier or equal price on the same bar.

At the next session, the simulator asks whether a real, positive, non-zero-volume open exists. If a name is halted, delisted, missing, or printing a dead zero-volume quote, the order remains pending. After three trading days without a tradeable bar, it is rejected. A bar is never invented.

Orders also face a liquidity cap. The baseline profile rejects an order whose open notional is more than 1% of the name’s trailing 60-session median daily dollar volume. A name whose median dollar volume is exactly zero is treated as untradeable, not as unlimited. Partial fills are not modeled in the live league baseline, so an oversized order is rejected rather than conveniently squeezed through.

Costs are part of the fill #

The baseline execution profile moves a buy above the raw open and a sell below it. The adjustment contains a liquidity-tier half-spread plus five basis points of additional adverse movement per side. The half-spread is 5 basis points for the most liquid names, then 10, 15, or 25 basis points as liquidity falls.

Since 19 October 2026 every book also pays a registered commission schedule modelled on a retail broker's tiered pricing. The change was recorded as a break in each book's history, and each book's evaluation clock restarted there, so a result never mixes the old and new cost rules. Research can apply named harsher profiles; those now apply on every fill path, including auction and intraday fills. The shared backtest core also requires a primary cost profile and at least one distinct, harsher sensitivity.

These costs make the paper result conservative in a useful way, but they are still a model. The daily open comes from retail-accessible data. It does not reproduce opening-auction queue position, partial fills, or every venue fee. That is why future broker-paper reconciliation, if ever approved, would measure the difference rather than assume the simulator is exact.

The nightly league #

After prices arrive, the league fills eligible pending orders, credits eligible dividends, marks every active book to the latest close, generates new orders, and appends that day’s equity. The step is atomic and idempotent: a partially written day cannot quietly become the official result.

The league ranks active books by total return. For each book it shows:

• inception date and current equity;
• total return and return relative to SPY over the same book lifetime;
• maximum drawdown;
• open positions and fill count; and
• return over the latest five sessions.

The SPY comparison includes dividends. A price-only benchmark would flatter every book by ignoring income that a buy-and-hold investor would have received. “Versus SPY” is still a comparison, not a claim that the book has passed a research gate.

Stale marks are displayed too. If a held name stops trading, carrying its last close can make equity look more certain than it is. The report identifies stale positions and pending exits instead of hiding them in a log. A holding that is delisted is closed out at its last real traded close, the same way for every book, so replaying the history gives the same answer.

Books that stop answering a question are retired. On 21 October 2026 eight books with no edge after costs were retired, leaving 17 active. A retired book stops trading; its equity history and positions stay in the record.

Engine v2: one ledger for every account #

Since 18 October 2026 the engine is the single executor and ledger for every paper book and every separately funded paper account. An account's cash, holdings, and equity are recomputed from its complete history of fills, corporate actions, dividends, financing, and fees every time it settles; the stored balances are only a cache that must match that recomputation.

Generic paper accounts can use a few more order types than the original books: opening- and closing-auction orders, and intraday orders that are filled later from recorded minute bars. Stock shorts are allowed with a recorded borrow availability and borrow cost, under margin and day-trading limits. Options execution is refused until real bid and ask quotes exist.

An account halts automatically at a 20% fall from its peak, a 5% loss in one day, or any mismatch between its ledger and the recomputed state. A halted account keeps its positions and stays marked; it never resumes on its own.

Matched books answer better questions #

The profitability evidence loop uses three books with the same universe, timing, sizing framework, entry protection, and cost model. One ranks by AI score, one by a deterministic rule, and one lets AI veto the rule. Because mechanics are matched, a difference is more likely to come from the score or veto than from a different fill assumption.

Private research strategies run in a separate private repository and reach the engine only through the shared backtest core and paper accounts. Each private strategy version gets its own separately funded account; a public identity may show that such a policy is running, but its private rules and research results stay outside this repository, and none of them has a proven edge.

Paper equity is evidence about a simulator under stated assumptions. It is not cash, it is not a broker statement, and it is never permission to trade live.