Agentic trading glossary

By the TradeAgentic engineering team·Published 30 August 2026·Updated 8 September 2026

Plain definitions for terms that come up when evaluating autonomous trading software. Written for someone deciding whether to buy, not for someone building.

Glossary of agentic trading terms

A

Adverse selection

Being filled precisely when the market is about to move against you. A cost that does not appear in a backtest but does appear in a brokerage statement.

Agentic loop

The cycle an autonomous system runs continuously: observe, judge, act, then grade the result and adjust. The grading step is what separates it from ordinary automation.

Agentic trading

Software that decides and acts on its own rather than executing rules a person wrote, and that changes its behavior based on the outcomes of its own decisions.

Algorithmic trading

Any trading driven by a program. The broadest term here, covering everything from a simple scheduled order to a fully autonomous desk.

Autonomous trading

Trading that proceeds without a person approving each order. Not a claim about accuracy, only about who decides.

B

Backtest

Running a strategy against historical data to estimate how it would have behaved. Useful for elimination, unreliable for prediction, and easy to fool yourself with.

Bracket order

An entry submitted together with its protective stop and target, so no window exists in which the position is unprotected.

Broker-resident stop

A protective order held at the brokerage rather than inside the trading software, so it still fires if the software stops running. The single most important safety property to verify.

C

Cold start

The period after installation when a system has no live record of its own. Worth knowing how long it lasts and how the system behaves during it.

Concentration cap

A limit on how much of an account any single position or asset class may become, regardless of how attractive the opportunity looks.

D

Daily loss stop

A fixed threshold at which trading halts for the remainder of the day. Effective only if no part of the software can widen it.

Drawdown

The decline from a peak in account value. The number that determines whether a strategy is survivable, more than the return does.

E

Execution quality

How closely fills match the prices a strategy assumed. Poor execution can erase an edge that looks real on paper.

Expectancy

Average profit or loss per trade after costs. A high win rate with negative expectancy is a common and expensive trap.

F

Fill

The completed execution of an order. What you actually got, as opposed to what you asked for.

H

Hold clock

The maximum time a position is kept before being closed regardless of outcome. Prevents a trade becoming an accidental investment.

K

Kill switch

A control that cancels all orders and closes all positions immediately. Should be reachable in one action and impossible for the software to gate.

L

Live-fire check

Placing one small real order at startup to confirm the whole path works and protection genuinely arrives at the broker, before larger size is risked.

Long only

Unable to take short positions. Common for retail crypto, and a meaningful constraint in falling markets.

M

Market data entitlement

Permission to receive a particular quality of price data. Consolidated feeds cost more than single-venue ones and behave differently.

Martingale

Increasing position size after losses to recover them. Reliably catastrophic. Its presence in any product is a reason to walk away.

O

Order path

The full route from decision to executed order at the broker. The thing a live-fire check exists to verify.

P

Paper trading

Trading with simulated money against real market data. The correct place to evaluate autonomous software, for longer than feels necessary.

Preflight check

A startup self-test that verifies credentials, data and the order path before trading is permitted. Should block trading on failure with no override.

Profit target

A level at which trading stops for the day. Sound as a stopping rule, dangerous if it causes size to increase.

R

R multiple

Profit or loss expressed as a multiple of the amount risked on that trade. Lets outcomes be compared across position sizes.

Reconciliation

Comparing the software's records against the brokerage account and correcting the difference. Necessary because the account is the source of truth.

Regime

The prevailing market condition. Strategies that work in one regime frequently fail in another, which is the central problem fixed rules cannot solve.

S

Shadow trade

A recorded decision that was not funded, tracked as though it had been, so refusals can be scored alongside trades.

Sizing ramp

Running early trades at reduced size before committing full size. A cheap way to catch failures that only appear with real money.

Slippage

The difference between the expected price and the achieved price. Small per trade, decisive over many.

Source license

A license including the code and the right to build it, so the software remains maintainable independently of the vendor.

Spread

The gap between the buy and sell price. The cost a trade must overcome before it is worth taking at all.

Stop-limit

A protective order that becomes a limit order at a trigger price. Will not fill at any price, unlike a stop-market.

U

Universe

The set of instruments a system is permitted to trade. A smaller universe is easier to monitor and usually easier to understand.

W

White label

Licensing software to present under your own brand, normally alongside source access.