Glossary of automated & agentic trading terms

Automated trading has its own vocabulary, and a lot of it is used loosely — sometimes deliberately, to make a tool sound smarter or safer than it is. This glossary defines the terms the way we use them on this site, in plain language, with the caveats attached. Where a term has a deeper page, we link to it. Nothing here is investment advice; it is reference material.

Agents & connectors

AI agent

A language model wired to tools and allowed to act, not just chat. An agent can read data, make a decision, and call an external function — for example, fetching a quote and submitting an order. On its own a model cannot trade; it becomes an "agent" only when you connect it to systems and grant permission. See can Claude trade stocks? for what that actually requires.

MCP (Model Context Protocol)

An open standard for exposing tools to an AI model through a connector. An "MCP server" publishes a set of functions — get quote, list positions, place order — that an agent can call. It is the plumbing that lets a model interact with a brokerage or a data feed in a structured, permissioned way.

Agentic trading

Trading in which an AI agent runs the execution loop — read the market, decide, place orders, confirm fills — typically on a schedule. The key distinction is who decides the trade: an improvising agent reasons fresh each run, while a rules engine hands the agent a fixed, tested strategy to execute. We compare the two in self-tuning vs. static trading bots.

Broker MCP

An MCP connector published by a brokerage that lets an authorized agent call broker actions — quotes, positions, buying power, and order placement — on your behalf. Robinhood's Agentic Trading is one example; equivalent broker connectors work the same way. The broker, not the model, executes and clears every order.

Alpaca

A brokerage and market-data provider. Coil uses a free Alpaca data key for price history and quotes; the data feed and the broker that places orders can be different services. Your credentials stay on your own machine.

Instruments

Leveraged ETF

An exchange-traded fund that aims to deliver a multiple — typically 2x or 3x — of an index's daily return. Because it resets daily, its return over longer periods diverges from a simple multiple of the index, and choppy markets erode value through compounding. This is "volatility decay." See SOXL and leveraged-ETF decay for the mechanics.

Inverse ETF

A fund built to move opposite to its index — it rises when the index falls. Inverse leveraged ETFs (like a −1x or −3x product) combine the inverse exposure with daily-reset decay, making them especially unsuited to buy-and-hold. Coil is long-only and never uses inverse ETFs; when the tape turns down it raises cash and rotates into a defensive Macro book rather than shorting.

3x semiconductor ETF

A 3x leveraged ETF that seeks three times the daily return of a semiconductor index. It rises sharply when chips rally and falls just as sharply when they drop. It is one example of a long leveraged vehicle Coil can use to accelerate a semis leader — at reduced size, at a real entry — never a buy-and-hold. The mechanics, and its inverse sibling, are covered in leveraged & inverse ETFs.

Strategy mechanics

Compression-to-ignition

The market idea Coil is named for: a leader tightens into a quiet, low-volatility "coil," then expands violently. The engine expresses it by buying leaders at real entries — a name scored READY or SETUP, at support, coiled and ready — across the whole market, never chasing an extended name and never catching a falling one. Entries on evidence, not prediction.

Leader rotation

The return backbone Coil's scoring is built on: hold whatever the market's current leaders are, and rotate as leadership changes — into the names and sectors showing the strongest, most durable relative strength, out of the ones fading. In a point-in-time research backtest it is what compounds a difference over a buy-and-hold benchmark; that edge concentrates in leadership regimes and thins out otherwise.

Entry window (READY / SETUP / WAIT / CHASE / FALLING)

Coil's label for how tradeable a name's entry is right now. READY and SETUP are real entries — at support, coiled, ready to buy. WAIT means the setup isn't there yet. CHASE flags a name pinned at its highs (extended — don't chase). FALLING flags a name still dropping (a falling knife — don't catch). The engine buys only READY or SETUP.

Sector rotation / sector phase

Capital moving between sectors as the cycle turns. Coil phases every sector — leading, improving, weakening, lagging — and tilts toward sectors turning up while damping ones rolling over, so a leader is bought when its whole sector has the wind at its back.

Market posture

A top-level read of the whole tape — roughly, how offensive or defensive to be. In an agile, offensive posture the engine takes entries more freely; in a defensive posture it tightens entries, trims size, and leans toward cash and the Macro book. Posture sizes the whole book, not a single trade.

Buy score

The single ranked number Coil assigns each name by blending opportunity, entry-window, hold-conviction, leadership, growth, sector phase, and posture. The dashboard ranks candidates by buy score; the engine buys from the top down and sizes by conviction. It is the one factor that decides what gets bought.

Trailing stop

An exit that follows price in your favor and triggers if price reverses by a set amount. Coil's protective stops are ATR-scaled — sized to the market's current volatility rather than a fixed percentage — so a quiet tape gets a tight leash and a violent one gets room to breathe.

Hard stop

A fixed maximum loss on a single position, no trailing involved. If price hits the level, the position is closed. Every Coil position carries an ATR-scaled protective stop.

Circuit breaker

An account-level risk rule that throttles or halts trading after losses, regardless of any individual signal. Coil's immutable safety core includes a drawdown ladder keyed to the account's high-water mark, an account whitelist, position caps, and a kill switch, and it fails closed. It is designed so one bad stretch can't compound unchecked.

High-water mark

The highest account equity reached so far. Risk rules measured against the high-water mark — rather than against the starting balance — react to drawdown from the peak, which is the loss that actually matters to you.

Single-symbol cap

A ceiling on how much of the account any one symbol may occupy. Coil applies a single-symbol concentration cap, so the portfolio is never fully concentrated in one position even at maximum conviction.

Sweep / cash yield

On no-setup days the account sits in cash, where the broker's own cash-sweep program may pay interest. For example, Robinhood Gold advertised roughly 3.35% APY in early 2026. That is the broker's variable rate — not paid by Coil, not guaranteed, and not risk-free — but it means idle days aren't necessarily idle capital.

Testing & metrics

Backtest

Running a strategy against historical data to estimate how it would have performed. A backtest is a hypothesis about the past, not a guarantee about the future — and it is only as honest as the harness that produced it.

Cold backtest / regime

A backtest run in a fresh process for each market regime (a distinct environment such as a bull, a bear, or a choppy year), so warmed-up state or cached data from one period can't leak into another and flatter the result. A "regime" is just a label for the kind of market a window represents. Coil ships a cold-backtest harness that runs one fresh process per regime.

Look-ahead bias

A backtesting error in which the strategy is accidentally allowed to use information it could not have known at the time — for instance, a future price or a same-bar close used to make an earlier decision. It inflates results dramatically. Guarding against it is a core job of Coil's research-backtest harness — point-in-time membership, next-open fills, one fresh process per window, decisions made only on information available at the bar — alongside the safety self-test suite the buyer can run themselves.

Overfitting

Tuning a strategy so tightly to past data that it captures noise rather than a real edge — it looks brilliant on the test set and falls apart on new data. The honest defense is a small number of robust rules tested across regimes, plus humility about how few trades you actually have.

Walk-forward

A validation method that tunes on one slice of history, tests on the next unseen slice, then rolls forward — repeating so each evaluation happens on data the strategy didn't train on. It is a stronger check against overfitting than a single in-sample backtest.

Self-tuning

A rules engine that re-fits a bounded set of parameters to recent results within fixed limits, rather than running identical static settings forever. The rules stay the same; their calibration adapts. In Coil this is deliberately narrow — position sizing only — and it is different from an improvising agent, because the logic is frozen and testable. More in self-tuning vs. static trading bots.

Learning loop (Coil)

Coil's local self-tuning mechanism. At each close, a pass on your own machine (python3 -m coil_v4.learn) reads your own closed round trips and leans position sizing — only sizing — toward what has been working, split by book (S&P 500 / Nasdaq-100 / Macro) and by buy-score tier. It never touches which names are bought, the entries, exits, stops, or caps. Tilts are shrinkage-weighted and hard-clamped (each dimension, then the stacked product, to roughly 0.55×–1.35× overall); it stays neutral until you have a couple of closed round trips; it learns from live round trips only, so nothing tunes until you arm the engine and close real trades; and every tilt is written to a readable audit trail (optimization_log.json). It runs offline and nothing leaves your machine.

Profit factor (PF)

Gross profit divided by gross loss across a set of trades. A PF above 1.0 means the wins outweighed the losses over that sample; the higher the number, the wider the margin. It says nothing about how many trades produced it, so a high PF on a small sample is weak evidence.

Max drawdown

The largest peak-to-trough decline in account equity over a period, expressed as a percentage. It is the worst stretch you would have had to sit through. A strategy with a strong return but a deep drawdown can still be unbearable — and dangerous — to actually hold.

Commerce

Merchant of record (MoR)

The company that legally sells a product to you and is responsible for the transaction — collecting payment, charging the right sales tax or VAT, and handling refunds and chargebacks. When a MoR is in the loop, you are buying from them, not directly from the developer.

Gumroad

The merchant of record Coil uses for checkout. Gumroad handles payment processing, receipts, and license keys, so the purchase is a clean, standard transaction. Coil is a single $29 download you keep, so all sales are final — the only thing ever corrected is a genuine billing error, such as a duplicate charge, which is a billing correction rather than a change-of-mind refund. Coil itself never sees your card details.

How these terms show up in Coil

Coil is a rules-based, long-only engine that scans the whole market — the S&P 500, the Nasdaq-100, and a Macro book — scores every name, and buys leaders at real entries, that you install and run yourself with your own broker credentials and capital. Claude runs on a schedule as the agent; a broker MCP like Robinhood places the orders. The strategy buys names scored READY or SETUP, accelerates the strongest with long leveraged vehicles at reduced size, scales out on a ladder behind a trailing stop, and rotates to cash and the Macro book instead of shorting — all clamped by an immutable safety core and validated by a research-backtest harness across a decade of market windows. Here is how the research backbone looked, year by year, against SPY.

Read the table with its honest rider, not just the headline: +638% versus SPY's +282% over the decade is real, but through the end of 2025 Coil ran roughly even with SPY at about one-third less drawdown (−23% vs −32%, positive in 9 of 10 years, worst −1% in 2018). The gap is carried by leadership regimes, where the strongest years land (2025 +51%, 2026 H1 +86%) — quoting those two as an expectation would be exactly the kind of cherry-picking this glossary warns against. These are research figures, and the engine is newly live. To see the full mechanics behind these terms, read how Coil works, or weigh the approach against the field in Coil vs. trading bots and signal services.