FALCON

Home · Updates · v2.2.1

Falcon v2.2.1 — the numbers, re-run on tick data

We rebuilt the backtest to be hard to pass instead of easy. The headline figure came down. The shape of the edge did not move.

30 August 2026 · these figures replace every earlier backtest number on this site

What the run assumes, and why it is set up to hurt

The old figures came from TradingView's tester on bar data. These come from CME tick data (Databento), replayed trade by trade, with every assumption pointed against us:

• Stop entries fill at the worse of the two prices, plus a tick. Never the price we asked for. Always a tick past it.
• Market exits pay a full tick too. On NQ that is $5 a side, every side.
• $4 round-turn commission per contract.
• Everything flat by 16:59 New York. No position is allowed to sleep through a close and be marked at a convenient price.
• No winsorising, no outlier trimming, no removed trades. The old page capped oversized winners because bar data occasionally filled far past the target. Tick data does not need the crutch, so the crutch is gone.

The window is 30 August 2023 → 27 August 2026: three years, 274 active trading days. Shorter than before, because this is the span we have clean tick data for.

Verified results — NQ, 1 contract, three years

Metric	v2.2.1 — tick data
Closed trades	492
Win rate	84.15%
Net P&L	$25,075
Profit factor	1.495
Expectancy per trade	$51
Average win / average loss	$183 / −$649
Max drawdown	−$3,134
Longest drawdown	40 trading days
Longest losing streak	3 trades
Winning days	77.4%
Positive months	80%
Daily Sharpe (annualised)	3.05

Year by year: 2023 +$1,637 (22 trades) · 2024 +$1,106 (131) · 2025 +$9,755 (184) · 2026 to 27 Aug +$12,577 (155). Every year positive, but 2024 was very nearly flat — $1,106 across 131 trades is $8 a trade, and a year like that is a real possible outcome, not an anomaly to be explained away.

Exits, and how fast this thing resolves

• 414 targets · 75 stops · 3 closed flat at 16:59.
• Median winner is held 25 seconds. Median loser, 2m 29s.
• 61.6% of all trades are over inside a minute. 30.5% take one to five minutes. 7.9% run longer than five.

Losers are held six times longer than winners. That is the system working as designed and it is also the uncomfortable half of it: when Falcon is right you know in seconds, and when it is wrong you sit in it.

★ PRIME strikes — and the concentration problem they create

A minority of setups clear a higher bar and take two contracts instead of one. Roughly one a week. In this run:

• 58 PRIME trades out of 492 — 11.8% of the book.
• Net $10,856, expectancy $187 a trade.
• Win rate 77.6% — lower than the 84.15% overall.
• They carry 43.3% of total profit.

We are publishing that last number rather than burying it, because it cuts both ways. Eleven per cent of the trades producing forty-three per cent of the profit is a concentration risk: miss a run of PRIME strikes and your year looks nothing like this table. Anyone evaluating Falcon should assume they will miss some.

Long and short are not the same business

Long: 186 trades, 86.0% win rate, $16,548. Short: 306 trades, 83.0% win rate, $8,527. Two thirds of the trades are short and they produce a third of the money. Over a three-year climb in the index that is what you would expect, and it is a reason to be careful about reading this record as regime-neutral.

Why the earlier years show fewer trades

Falcon's quality bar is defined in NQ points and never changes. The engine fired at a steady ~25 raw setups a month across all three years, but NQ's typical intraday range roughly doubled as the index climbed from ~15,000 in 2023 to ~29,600 in 2026, so far more setups cleared a fixed points bar recently: about 5 qualifying trades a month in late 2023, 11 in 2024, 15 in 2025, 19 in 2026.

In quieter, lower-priced regimes the system trades less rather than lowering its standard. That is the intended behaviour, and it is also why 2023 and 2024 show modest positive results on few trades rather than losses. If NQ's ranges compress again, expect the trade count to fall with them.

2026 so far, on its own

The current regime, 1 January → 27 August 2026: 155 trades, +$12,577, profit factor 1.703, 83.23% win rate, expectancy $81, worst drawdown within the year $1,805, best day $1,354, worst day −$1,138, 87.5% of months positive.

Better than the three-year average on every measure. Eight months is not a sample that settles anything, and a single year of a high-win-rate system is exactly the length of record that flatters itself.

What we are still not claiming

• This is a high win-rate, low payoff system. Average win $183 against average loss $649. The break-even win rate is 78.0% and the run came in at 84.15%. That margin is 6.2 points — real, but not large. Worse fills or a higher commission tier eat into it directly.
• The commission modelled here is $4 a round turn. Our verified live rate is $5.76. Across 550 contracts that is about $968 the backtest did not pay, which would put net closer to $24,100 at the rate we actually pay. It is the one line on this page where the test is kinder than reality, and we would rather point at it than have you find it.
• Backtested is not live. Tick data with punitive fills is the closest honest approximation we can build, and it is still a simulation. The live record is published session by session in the forward test log, red days included.
• Three years is three years. It contains one broad regime. It does not contain 2008, or 2020, or a year where Nasdaq ranges halve.

What changed in the software

• Everything flat by 16:59 New York, with working orders cancelled and an alert for anything still open. The old build could hold through the close; a live account should not.
• The performance table on the chart now counts today only, resetting on the New York date. It previously accumulated across every bar your chart had loaded, which on a fast timeframe is several days — so a panel that looked like a session tally was not one. The header now reads TODAY · NY and changes if you switch it.