Talon — the live track record

When both sides stop, it costs us about half what it costs the benchmark

A double stop is the worst ordinary outcome of the day: the market moves through both short strikes and both sides close at a loss. Talon takes more of them than the benchmark — and that is by design. Tighter wings and an active defend rule mean we get stopped out more often and lose less each time.

The frequency drifts with the market. All three strategies double-stop far more often in 2026 than in 2022 — that trend is the volatility regime, not the strategy. What has not drifted is what it costs us. Talon's double-stop drag has sat between −$19 and −$24 per entry for four years, while the benchmark's has worsened by 46% and 63%.

862 trading days traded by all three, May 2022 – July 2026, at $1.20/contract all-in. Talon is 30-wide with an active defend rule; the benchmark arms are 40- and 50-wide with no defend — the higher stop frequency comes from the defend rule, the smaller loss from the tighter wings. A higher double-stop rate is not a worse strategy and a lower one is not a better one; that question is answered by net profit against drawdown, on the Talon page.

How this record works

• Every served trading day posts — no discretion, no skips. A day is scored (the number), held (our data-quality gate refused the day's market data; we attempt repair), or unscorable (the data vendor never served verifiable data for that day; the day is permanently excluded and permanently counted above). Red days render exactly like green ones.
• Results post the following day, after verification. Each night the day's tick data arrives from our vendor, passes independent quality checks on both data feeds, and only then is the day scored and published. The check mark means the day's scored values were verified against the parameters actually delivered to subscribers that morning — parameters committed to our database before the session opened.
• The benchmark runs beside us, every day, on the same tape. The standard MEIC configuration from the four-year comparison — both widths — is scored by the identical engine with identical fill rules, including the days it beats Talon and the days it cannot construct a trade at all. The benchmark's configuration is frozen; it never gets “improved.” It also trades a different part of the session — mid-morning into early afternoon, where Talon trades the afternoon — so the two run on the same days and the same tape, not in the same conditions.
• Costs are in the numbers, at a realistic rate. Every figure on this page is net of $1.20 per contract all-in — commission plus exchange and regulatory fees, the rate a retail SPX account actually pays. Entry legs are charged always; closing legs only on sides that stop; expired legs settle free. The Talon page carries the full sensitivity at $1.40 and $1.60 for higher-cost brokers.
• What MC95 is, and what it is not. We reshuffle the strategy's own daily results 20,000 times and take the 95th-percentile worst drawdown across those runs. It answers one question: given the days this strategy actually had, how badly could they have been ordered? At $6,139.48 it is 4.5× the worst single day on record (−$1,365.40, June 30, 2022) — not because it models a worse day, but because it chains bad days together into a losing stretch that runs about 82 sessions. It is bounded by our own history. It cannot price a day worse than any we have seen, and no drawdown model can. The realized worst drawdown of $3,021 came in well under it, which means the actual ordering of those four years was luckier than typical. Size for the possibility that the next stretch is not.
• Stops pay the spread. Every stop-out in every column is charged the full bid/ask spread at the moment it fires — the fill a real order gets.
• Forward days are held to a stricter bar than the certified history: the certified basis includes a handful of marginal-data days inside a disclosed, guard-validated aggregate; a forward day stands alone in public, so any day failing today's quality gates is held or excluded rather than scored.
• Tranche counts are disclosed. On low-premium afternoons a strategy may fill fewer than its scheduled entries at the credit target; the day scores what was actually constructible, and the count is printed beside the number.
• An example, from this record. On August 6, 2026 our data vendor served damaged data — on the 1-minute chains, the multi-day chains, and the tick tape, all in the same night. Trade counts and row counts looked entirely normal; the damage was missing quotes at strikes near the money, which is exactly where it matters and exactly where a casual check would not look. Four independent structural checks caught it. The day was not scored. It was scored only after the vendor repaired the session at source and all three datasets passed re-verification — the tick tape came back with 23% more data than the damaged pull. That is what the gates are for, and it is why this page can go days without a number rather than publishing one we cannot stand behind.

This is the model's record, not any subscriber's. Figures are per single-lot, including $1.20 commissions, from the published strategy at honest fills. Individual results differ with fills, contract counts, spread width, and skipped days. Backtested and model performance are hypothetical and are not indicative of future results. Nothing on this page is investment advice.