A 40% backtest should make you suspicious. Including ours.

If you see a backtest showing 20% a year, your instinct should be to close the tab.

It is a good instinct. In this industry, a straight line going up at 20% usually means someone turned too many knobs. They tried a thousand variations of a rule until they found the one that perfectly matched the last ten years of history. Then they showed you that one.

That is called overfitting. It is the financial equivalent of memorizing the answers to a practice test, then claiming you understand the subject. When the real exam arrives—the next recession, the next inflation spike—the memorized answers fail.

Our record shows 40.6% a year. That is double the number your brain already flags as a scam. So it has to earn its place. It cannot rely on hope. It must rely on how the number was built.

This is not a story about how smart our code is. It is a story about how we prevented ourselves from cheating.

The cheat code everyone uses (even by accident)

Imagine you are building a rule to sell stocks when the market drops.

You try selling at a 4% drop. The result is okay. You try 5%. Better. You try 5%, but only on Tuesdays. Suddenly, the chart looks perfect. It dodges every crash in the last decade. It catches every rally.

You haven’t discovered a law of physics. You have just tailored a suit to fit a mannequin that is already standing there. The rule works beautifully on the past because you carved it out of the past. But the future does not care about your specific Tuesday rule.

This is why so many backtested strategies disappoint once real money is on the line. They were fitted to the past, not built for the future.

To avoid this, you have to do something uncomfortable. You have to write the rules, lock the door, and throw away the key before you look at the data you intend to test.

Three kinds of records

Careful investors distinguish between three types of performance claims.

Kind	What actually happened	Why it fails
In-sample	Rules were tweaked until they fit the historical data perfectly.	It’s a memory, not a strategy. It breaks on new data.
Out-of-sample	Rules were frozen first. Then applied to history they never saw.	Still a simulation. Assumes you could execute trades at the exact modeled price.
Live	Real money moved through real brokerage accounts with real fees.	Takes years to build. Still offers no guarantees about tomorrow.

Most marketing materials blur these lines. They show you an in-sample curve and call it a track record.

The Kairos record is out-of-sample.

What our record actually is

The numbers you see for 2020 through 2026 come from a specific constraint: The rules were finalized before the 2020–2026 window began, and were not tuned on it.

We did not adjust the sensitivity of the signal to make 2022 look better. We did not change the exit criteria to capture more of the 2023 rally. The logic was set. Then we ran the clock.

Here is what went into the calculation:

• Blind Execution: The model scored five public market checks (credit spreads, VIX term structure, breadth, index momentum, tech vs utilities) every day at 3:50 PM ET. It acted on the score. No human intervened to say, “Wait, today feels different.”
• Friction Included: The math assumes you pay commissions. It assumes you suffer slippage (the gap between the price you see and the price you get). It assumes you executed near the close, not at the intraday peak.
• Simple Positions: When the checks read healthy, the model held 83% TQQQ (leveraged Nasdaq). When they read fragile, it held 100% short-term Treasury bills (proxied by SHV). No complex derivatives. No illiquid small caps.

The result was an annualized growth rate of 40.6%, compared to 15.5% for the S&P 500 (SPY) over the same window.

But the headline number is the least interesting part. The shape of the ride matters more.

Metric	Kairos Model	SPY
Annual Growth	40.6%	15.5%
Worst Drop	-24.5%	-33.7%
Time in Market	59%	100%
$100k Becomes	$896k	$252k

You can view the year-by-year breakdown, including the years we lagged, on our track record page.

What this proves (and what it doesn’t)

An out-of-sample test answers one specific question: Was this rule fitted to the past?

Because the rules were locked before the window opened, the answer is no. The model survived the 2020 crash, the 2022 stock-and-bond collapse, and the 2025 grind without any tuning. That suggests the logic captures something real about market regimes, not just noise.

However, this record leaves several critical questions unanswered. A skeptical reader should hold these doubts firmly.

1. It is not audited. No third-party accounting firm has verified this series. You are taking our word that the rules were locked before the window. There is no seal on the envelope, and “trust us” is a weak foundation. We say so plainly.

2. It is not live client money. These are model results. A computer simulating a trade is not the same as a human clicking “buy” while their heart rate spikes. While Kairos Select and Kairos Enhanced have been tracked as live models since September 30, 2025, that history is about one year long, gross of fees and slippage, and not audited. It is far too short to prove anything.

3. Modeled costs are estimates. We assumed realistic slippage. But in a panic, liquidity dries up. The gap between the bid and ask can widen beyond our assumptions. Real-world execution can be uglier than a spreadsheet.

4. It missed a big rally. In 2025, the S&P 500 gained roughly 18%. This model gained 3%. If you had used this system then, you would have spent the year watching the index pull away. A record designed to flatter would have smoothed that year out. We kept it in because it happened.

5. It is only six years. Six and a half years covers a lot of ground, but it does not cover everything. It has not faced a multi-year grinding bear market like 2000–2002. The next crisis might look nothing like the last one.

How to use this information

If you are considering this approach, do not project 40% returns into your retirement plan. That is a fantasy. Plan for worse.

Instead, look at the mechanics.

Focus on the defense. The model spent 41% of the time in T-bills. Its worst drop was -24.5%, against -33.7% for holding the index. For an investor five years from retirement, the size of the worst drop can matter more than the size of the best year. Sequence of returns risk is the silent killer of portfolios; this system attempts to mitigate it.

Expect to be wrong. The model is not a crystal ball. It generates false signals. It can sell before a rally and buy before a drop. In 2025, it trailed the index by about 15 points. In a choppy, directionless market, leveraged ETFs like TQQQ decay. You can lose money even if the signal is “right” about the direction eventually.

Treat it as a slice, not the pie. This is not a replacement for your entire net worth. It is a tactical sleeve. Size it so that a string of losses hurts, but does not cripple you.

Ask the hard questions. Do not just ask us. Ask any provider of algorithmic strategies: When were the rules finalized? Did you tune them on the data you are showing me? Where is the audit?

We have compiled a list of essential inquiries in our guide on questions to ask any algorithm. You can also read the full legal disclosures regarding model performance on our disclosures page.

The path forward

What would make this record stronger? Time, live results, and independent verification. None of those exist yet.

Until they do, the 2020–2026 window is a stress test, not a promise. It shows how a rules-based approach, blinded to the outcome, behaved through the last cycle, including the year it lagged.

Whether it does so for you depends on your ability to stick to the rules when they feel wrong, and your willingness to accept that no mechanism—no matter how rigorous—can eliminate risk entirely.

The short version

• Many backtests are fitted to history. Rules tuned on the past tend to disappoint on the future.
• Ours was blinded. The Kairos rules were locked before 2020 and tested on data they never saw. That rules out curve-fitting on those years. It does not guarantee success.
• The numbers include friction. Commissions and slippage are modeled, but real-world execution may vary.
• It is not perfect. The model lagged significantly in 2025 (+3% vs +18%) and carries the risk of whipsaw losses.
• Verify, don’t trust. Treat this as a hypothesis that has survived a stress test, not an audited fact. Discount the headline returns and size your position accordingly.