How a consensus model traded PLTR and skipped most of the rally
It acts only when several market-state readings hit an extreme at once, then trades both ways and exits when they calm. It won 74% of 58 trades for +208.32%, yet sat out most of a PLTR run that more than tripled.
PLTR was one of the period's loudest names, and rather than chase it the model stood aside until its conditions agreed, taking a measured slice while a passive owner rode the full run. This is a backtest, a fixed rule set replayed over saved PLTR history, not a forecast or a recommendation, and the live sample is still too thin to grade.
The signal is a consensus check. The model standardizes a basket of market-state readings against their own recent history and waits until they reach a shared extreme; only then does it open a position, long when the stretch points up and short when it points down. That patience shows in the record: 58 trades in about two years, a 74% hit rate, and a fixed-stake total of +208.32%. What it bought with all the waiting was protection; what it gave up was the rally.
How the model is built, end to end.
The model watches a small basket of market-state readings on PLTR rather than a single line on the chart. Each reading is standardized against its own recent history, turned into a measure of how unusual it is right now, and the separate measures are combined into one composite score. A quiet, middle-of-the-range composite keeps the model flat. Only when the components stretch to an extreme at the same time does the score cross the level that lets a trade open.
Direction follows the stretch: an extreme of one sign opens a long, the opposite sign opens a short. Every position carries a protective stop and a profit target set in advance, but the model rarely needs either. Most trades close a third way, when the composite relaxes back toward its normal range and the reason for the position has passed. That design is why the book fills with small, quick, mostly-winning trades instead of a few giant ones.
The illustration follows the model end to end on PLTR: the raw readings, the standardizing step, the test for whether enough of them agree, and the order that finally goes out. No single step is clever on its own. The discipline is in requiring all of them before committing a dollar.
At a glance
| Gate | Actual | Threshold | Status | Threshold source |
|---|---|---|---|---|
| win rate | 74.14% | >= 70.00% | pass | canonical registry standard |
| max drawdown | 10.18% | <= 5.00% | fail | canonical registry standard |
| sample size | 58 | >= 30 | pass | canonical registry standard |
| total return | 208.32% | >= 100.00% | pass | canonical registry standard |
| expected return | 3.592% | >= 5.000% | fail | canonical registry standard |
| Metric | Value |
|---|---|
| Total return | 208% |
| Win rate | 74.1% |
| Max drawdown | 10.2% |
| Expected per trade | 3.59% |
| Trades | 58 |
These figures come from a backtest of the model on PLTR, scored against fixed acceptance gates, not from a live track record.
Walk-forward verification
| Metric | Value |
|---|---|
| Walk-forward match | 100% |
| Verified timestamps | 1,631 |
| Signal correlation | 1 |
A trade walked through
The clearest single example is a short from August 2025. PLTR had pushed up near 188 dollars when the composite score stretched to a high extreme, and the model sold into it. It held for seven days as the reading worked back toward normal, then closed when the signal relaxed rather than at a preset target. The trade returned +19.02%, among the model's largest, and it is the whole book in miniature: enter on a stretched reading, wait for it to calm, exit on the signal.
| Metric | Value |
|---|---|
| Direction | short |
| Entry price | 188.09 USD |
| Exit price | 152.32 USD |
| Hold time | 7.0 days |
| Return | +19.02% |
What the full trade record shows
Across its 58 PLTR trades the model won 43 and lost 15. Almost everything closed on its own signal: 54 of the 58 exited when the composite relaxed, with just 3 stop-outs and a single take-profit.
| Exit reason | Trades | Share |
|---|---|---|
| Signal exit | 54 | 93.10% |
| Stop-out | 3 | 5.17% |
| Take-profit | 1 | 1.72% |
That is the fingerprint of a normalization system: it opens on a stretched reading and almost always closes when the stretch fades, rather than being carried out at a fixed target or stop. The rare stop-outs are the times the reading kept stretching instead of calming.
The biggest winner was a long that ran to its profit target for +23.26% in two days off a low base. Two of the next largest came in August 2025, when the model worked both sides of one swing: a long first rode the climb for +15.62%, then a short from about 188 dollars closed +19.02% as the stock fell back. The slowest winner crept up just +0.52% over two weeks before the signal released it.
The worst trade was a long that kept sliding into its protective stop for -9.88%, the model's deepest single loss and the reason the drawdown gate is the one that bites.
No single trade carries the record; the edge is thin and spread across many small normalization wins, which is both why the win rate is high and why the average trade earns only about three and a half percent.
How does this compare to just holding PLTR
Held outright over the same window, PLTR more than tripled. The model returned a fraction of that: it underperformed a plain hold by a wide margin, because a rule that only trades when its conditions agree is absent for most of a straight-line climb. The tiles below size the gap.
| Metric | Value |
|---|---|
| Model total return | +208.32% |
| Buy-and-hold | +477.13% |
| Difference | -268.82% |
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 not built on. It is a consistency and replay-integrity test, not proof the model will make money live. A clean reproduction means the deployed rules behave like the studied rules; it says nothing about whether PLTR will keep handing the model the same extremes to fade.
| Metric | Value |
|---|---|
| Match rate | 100.0% |
| Correlation | 1.000 |
| Alignment | Aligned |
So far the model's live trades line up with how it behaved in the backtest, which is what you want to see early. But the live sample is small, and early agreement is not evidence of a durable edge; it only says the deployed rules are doing what the studied rules did.
When this approach fails
The model's weaknesses are specific to a consensus gate. Its sharpest failure is a genuine regime change dressed up as a passing extreme: the conditions stretch, the model takes the position, and the move keeps going the same way instead of calming, which is exactly how the -9.88% loss in February 2026 happened. A long stretch where the components disagree leaves the model flat for weeks, missing clean moves it never voted on. And because each winning trade is small, a short run of stop-outs can erase many ordinary wins at once. None of this is a defect to patch; it is the cost of demanding agreement before acting.
| Metric | Value |
|---|---|
| Losing trades | 15 |
| Worst single-trade return | -9.88% |
| Worst in-trade drawdown | -10.18% |
Three things are worth watching if this ever trades at size. The first is drawdown: the worst stretch in the backtest was about ten percent, and a consensus model can still string several small losses together. The second is the distance between live and backtested behavior, which is the earliest sign an edge is decaying. The third is how often the conditions actually agree, because long quiet spells are normal for this design rather than a malfunction.
Risk and honest limits
Treat everything here as backtest evidence about how the rules behaved on saved PLTR history, not a verdict on where the stock goes next. The failed gates above, not this note, are the real caution.
Lifecycle
Where we are
These figures are a backtest, not a live track record. As real trades accumulate, a live-performance section can be added; until then, read every number here as evidence about the rules on saved history.
Sources
This article is based on Stonewell One research, including backtesting, walk-forward verification, deployment monitoring, and model-risk review.Trade-level entries, exits, and holding times come from Stonewell One's backtest of PLTR over the March 2024 to April 2026 replay window.The model is compared against simply owning PLTR over the same window.