How Coil works

The short answer: Coil is rules-based trading software you run on your own machine, with your own broker and your own capital. It watches one tape — semiconductor ETFs SOXL (3x long) and SOXS (3x inverse) — waits for the index to coil tight, rides the move when it ignites, and exits by rule. An AI agent (built for Claude) is the operator; nothing here is a signal service, managed money, or investment advice. Below is exactly how each piece works, and where it's weak.

The thesis: compression to ignition

Semiconductors tend to lead the market — they often move first and move hardest. Coil's thesis is built on a specific, measurable pattern in that leadership: the semis complex compresses — range tightens, volatility contracts — and then ignites, releasing a directional move that can run before the rest of the market follows. It is a hypothesis the software trades, not a law of markets.

The non-obvious part: a real 2%+ move on the semis isn't a trend you chase after it's obvious. It tends to be born as a short-timeframe reversal inside the prevailing one-hour trend. Coil's job is to be positioned at the coil, not to pile in once everyone can see the ignition. On 3x ETFs that timing matters enormously — a roughly 10% move in the underlying index is about a 30% move in the ETF, before gaps and slippage. (Why leverage cuts both ways is its own topic; see SOXL decay.)

Entries: leg-rider logic

Coil doesn't predict. It waits for compression to resolve, then enters on the confirmed first leg of the move and rides it. The logic is mechanical: a setup either meets the conditions or it doesn't. There are no targets to hope toward and no averaging down into a loser. On most days nothing qualifies — and that's intentional. Standing down is a position. Idle cash sits in the broker's cash sweep (around 3.35% APY, variable, Robinhood Gold, as of Feb 2026 — the broker's yield, not Coil's, and not risk-free).

Long, short, or cash — it reads direction from the tape

A fair question: is the performance just semiconductors going up? No. Coil is not a buy-and-hold of leveraged semis — it's direction-aware and in cash most of the time. Each setup is read against the 1-hour trend: in an uptrend it takes the long side (SOXL); in a downtrend it can take the inverse (SOXS), intraday only. About 39% of sessions produce a qualifying setup; the rest it sits in cash. In the backtests, the results come from selecting those entries and exiting fast — not from holding 3x exposure through a one-way market. Those figures are backtested, best-case, not typical, and not a prediction.

The honest limit: its most-validated edge is the long-side leg. Sustained bear markets are its weak spot — there it stands down to cash on confirmed bear days rather than pressing the short side, which is the harder, less-proven half of the system. "Direction-aware" means it can be long, briefly inverse, or flat — not that it reliably shorts a crash.

Exits: the system judging its own entry

This is the part most people miss, so it gets its own heading. Coil's first and primary risk control isn't a stop-loss — it's the exit logic deciding the entry is no longer working. The moment the leg statistically ends — a small counter-move against the position (about a 0.8% reversal in the underlying index, roughly a 2.4% trail on the 3x ETF) — Coil is out. It doesn't wait for a big loss to "prove" the trade failed; it leaves when the move it entered for is done. On essentially every trade, that's where the position ends — small, by rule, win or lose.

Two backstops sit behind that judgment: a 5% hard stop per position for a fast adverse move, and inverse (SOXS) positions are flat by the close — never held overnight. The honest caveat: a stop is not a guarantee. Markets gap, and a long position can open below its 5% stop — a stop cannot fill inside a gap. That risk is real and permanent.

The circuit-breaker ladder — the failsafe, not the steering

Everything above manages each trade. The circuit-breaker ladder is different: a separate, account-level layer that watches both your loss on the day and your cumulative drawdown from your equity high-water mark. Think of it as the seatbelt, not the steering wheel — it exists for a bad streak or a violent gap, and it rarely engages, precisely because the per-trade exits above already keep individual losses small. (In the backtested regimes, account drawdown ran roughly 4–8% — deepest in the 2022 bear at 7.6%; backtested, not typical, not a prediction. Even the −10% rung was essentially never reached, let alone the deeper ones.) It escalates as drawdown deepens:

Read these as account-level limits across a streak, not a per-trade loss — a single trade is cut at the leg-end exit or the 5% stop long before any of this. A single 3x symbol is capped at 65% of equity. Plainly: the ladder reduces single-day damage; it does not remove risk. Leveraged ETFs can lose value rapidly, including total loss of capital.

The cold-backtest harness — and the bugs it caught

Numbers are only as honest as the machine that produces them. Coil's A/B harness runs a fresh process for every market regime — "cold validation," with no cross-run cache warmth quietly inflating results — and enforces 20 integrity guards against the usual ways a backtest lies to itself.

The rigor isn't just a claim; it has a track record. The harness found its own three backtest bugs: a look-ahead leak, a next-day leak, and a sign-inverted short book. Each one would have made the strategy look better than it was. Each one was caught and fixed before a dollar moved. That's the whole brand — we'd rather show you the bugs than the brochure.

Two self-tuning loops, on a leash

Coil re-fits itself locally, but only within hard, whitelisted bounds, with automatic rollback if a change degrades. A nightly optimizer retunes capital multipliers from your real broker fills; a weekly evolution loop proposes hypotheses that must clear the cold four-regime gatekeeper before any live capital sees them. The leash is the point: the engine can adapt, but it can't quietly wander into something untested.

Backtested regime results

These are backtested figures — best-case, not typical, not a prediction, and not client returns. Treat them as hypotheses, not proof.

Three of four backtested regimes are profitable. The bear is the honest weak spot: Coil now stands down to cash on confirmed sustained-bear days, which cut the 2022 backtested result from −3.6% to −1.4%. It's improving, not solved — and the best window is a trailing slice, its strongest, not something to anchor on.

Where to go next

If you want the operator side — how scheduled Claude sessions and a broker connector actually arm and run this on your machine — read the agentic trading guide. If you're weighing Coil against bots and signal services, the comparison lays out the differences. The engine is a $5 one-time download, yours to keep and run forever.