How a condition z-score rule traded BE through an eighteenfold climb in backtest

Bloom Energy rose steeply over this window, climbing from about 9 dollars to a within-window peak near 178 and ending around 167, roughly an eighteenfold run over that window. Every figure here is historical to that backtest window and is not current market prices. A rule that only wakes up at extremes and leans long returned +270.03% trading those moves in backtest, a real gain that was nonetheless dwarfed by what an investor made simply owning the stock over the same stretch. That is one number on saved history, not a live result. This is a research study of the model's fixed rules on BE, not a forecast and not financial advice; because the model fell well short of the standard it is measured against and took two separate double-digit hits, its worst losing trade and its deepest intra-trade dip, the pages that follow weigh how the rule behaved rather than what it might earn.

Underneath, the rule watches two BE positioning readings, each measured against how far it has strayed from its own typical level, and it does nothing until both reach a rare joint extreme pointing the same way. When either reading is merely wandering inside its usual band the model sees no signal and stays cash; a position is born only when the pair agrees at an outlier. Upward agreement buys, downward agreement sells, and because both must line up the trigger is selective by design. From there the exit is the mirror image of the entry: the position is held while the pair sits stretched and let go as the two readings settle back toward ordinary, which is how nearly every position closed. Over the window the pair aligned 93 times, tilted heavily long at roughly 73 to 20, and resolved as winners on 65.59% of those even-sized entries, the arithmetic behind the cumulative figure above.

How the model is built, end to end

The readings the model is built on are two BE positioning series, each standardized against the level it itself usually shows so a move can be judged as ordinary or extreme by that series' own history rather than an outside yardstick. Readings close to their normal band tell the model nothing and keep it out. Only when both series stray far enough from that band in the same direction does the model treat the condition as met and a trade as possible, which is why long stretches of ordinary readings pass without a position.

A joint stretch is permission, not an instruction. Before committing, the model reads the direction of the stretch to pick its side, opening a long when the readings push to an upward extreme and a short when they push down, each with a predefined protective stop and a profit target set ahead of time. From there the trade ends by one of a few routes, and on BE almost every position closed when the readings eased back toward their normal range rather than by hitting a hard target or stop.

Picture the model laid across BE from end to end: first the two standardized readings, then the joint extreme they must reach before the gate opens, then the long or short that follows. Nothing in the chain is elaborate; its governing instinct is to stand aside until both readings stray far from their usual band in agreement and the direction of that stray is clear before a dollar is risked. Its lineage begins as one candidate in an automated idea search, then runs through backtesting and walk-forward checks to a research-gated rule; what you see here rests on that saved-history test alone, with no live record yet to lean on.

At a glance

These figures come from a backtest of the model on BE, scored against fixed acceptance gates, not from a live track record.

Walk-forward verification

A trade walked through

The walked example is a long held about a day. Both readings stretched to an upward extreme with BE near 15.52 dollars in early April 2025, the model bought, and after a shallow dip against it the position ran until the readings eased back near 18.44 dollars for +18.81%, closing on a signal exit rather than a hard target. It is the model at its clearest: a long opened on an upward joint stretch and held until the condition cleared, the side of the two-sided rule that supplied nearly all of the backtest's gain.

What the full trade record shows

Across its 93 BE trades the model won 61 and lost 32. The exits leaned almost entirely on the readings easing back: 87 closed when the readings returned toward their normal range, 4 hit a time limit, 2 were stopped out, and none reached the profit target.

A split this heavily weighted to the readings easing back is the mark of a model that closes when its condition clears rather than waiting for a fixed target, taking many resolutions off the extremes instead of a few large target hits. That fits the thin per-trade edge: the average trade earned about 2.90%, so the record is built from a steady stream of mostly long entries on a rising stock, not from any one outsized win.

The biggest single winner was a long that ran +28.13% over about two days from near 26 dollars in July 2025, closing when the readings eased back rather than on a target. Across the book the rule traded profitably on both long and short entries, though with roughly 73 of the 93 entries on the long side, the trade set was dominated by longs, which fits a stock that mostly climbed.

The worst single trade was a short stopped out for -12.98% within a day in April 2025 when BE jumped back against the position near 16 dollars. Separately, the deepest intra-trade drawdown in the book, quantified in the risk section below, came on a long that dipped into double digits before recovering, and it is one reason the drawdown gate is a bar the backtest missed.

No single trade carries the record, and at a 2.90% average edge the book depends on trading often and being right about two times in three. Because BE climbed almost the whole way, the lesson is the opposite of a triumph: a rule that only trades at extremes and leans long still captured far less than simply holding the stock, and a double-digit intra-trade drawdown with a thin per-trade edge keep it research-only.

How does this compare to just holding BE

Over the same window the model was tested on, simply owning BE returned about +1703.46% as the stock climbed from roughly 9 dollars to near 167, while the rule returned +270.03% in backtest. Setting the two side by side is how you judge whether the rule added anything over just owning the shares, and on this saved history it lagged badly: by staying flat near the normal band and only trading at extremes the rule sat out most of a near-straight-line climb that a passive holder captured in full. That shortfall is a backtest result, not a promise, and it sits alongside the three acceptance gates the model missed. The tiles below put numbers on the comparison.

How well does the model reproduce its tape?

Walk-forward verification checks whether the saved rule path reproduces the expected signal behavior on held-out timestamps it was never fit on. It is a consistency and replay-integrity test, not proof the model will be profitable live. A clean reproduction means the deployed rules act like the studied ones; it says nothing about whether BE will keep stretching these readings to the extremes the model traded across the test window.

In live trading the model has almost no track record yet, far too little to evaluate, and whatever live activity exists forms no part of the backtest results reported here. Live behavior remains essentially unproven, so read everything here as exactly that, a study of how the rules behaved on saved BE history rather than a live track record.

When this approach fails

The model's losses come from being caught on the wrong side of a snap. The readings stretch, the model takes its side, and BE moves against the position before the expected move follows. That is what happened on the short covered in the full-record section, stopped out within a day in April 2025 when price jumped back; it is also the shape of the deepest intra-trade drawdown in the book, a long that fell into a double-digit hole before recovering. Either way, a sharp reversal against a stretched reading is the failure mode, and it is why the drawdown bar is one the backtest missed by a wide margin.

Three things are worth watching if this ever trades at size. The first is the drawdown, about 13.43% at its worst intra-trade excursion in the backtest, well past the acceptance bar, because a sharp reversal against a stretched reading can run deep before it recovers or stops out. The second is the thin per-trade edge, since the average trade earned only about 2.90%, below the bar the gates set, so the record leans on volume rather than a strong edge on any one trade. The third is the gap between live and backtested behavior, the first sign the readings are no longer stretching on the same moves they did across the test window.

Risk and honest limits

On this run the model's automated checks logged a caution rather than a clean pass, one more reason to weigh the rules on their record here rather than as a call on the stock.

Lifecycle