How these numbers were made — and audited

Methodology

The value of a backtest is set by the process that was allowed to produce it. This page documents that process, including what it killed.

The five gates

1 · Pre-registration. Each candidate rule — entry, stop, exit — is written down and time-stamped before the deciding data is examined. Rules noticed in hindsight are labelled post-hoc and sent to gate 3 for an independent verdict.

2 · Split-sample. Parameters may be tuned only on the first 17 months. The frozen rule then faces the final 14 months — which contain the 2026 correction — untouched.

3 · Replication. Survivors re-run, parameters frozen, on two markets never used in development: QQQ and SPX. A rule that only works on its home market is treated as fitted, not discovered.

4 · Look-ahead audit. Every signal is checked to use only information knowable at fire time. Two look-ahead defects were caught in our own harness during this research — one had inflated a result six-fold before the audit removed it. We publish this because it is the defect class that invalidates most retail backtests.

5 · Full costs, itemized. Every trade on every strategy page shows Gross, Costs and Net as separate columns. The cost build-up and a slippage stress test follow below.

Margin formulas

Defined-risk structures (Meridian, Afternoon Pin). Margin = wing width − credit received, per contract — the position's exact maximum loss, identical on every account type. Shown per trade in the blotters.

Single short options, standard account (Reg-T). The published CBOE formula: (max(20% × underlying − out-of-the-money amount, 10% × underlying) + premium) × $100. Computed per trade from the SPX level at entry.

Single short options, portfolio margin (TIMS). The OCC index stress: worst-case intrinsic value across a −8% / +6% underlying move, plus premium, × $100. Computed per trade. Broker house rules may add a floor; both columns appear in the blotters so any broker's requirement can be verified against them.

Cost accounting

Fills. Every fill is a last-traded exchange price from that minute's bar — a price at which someone actually transacted — never a mid-quote and never an assumed level. Stops are evaluated on the minute-by-minute path.

Explicit charges, deducted on every trade:

Component	Per contract-side
Broker commission (SPX index options)	$0.50–0.65
Exchange + clearing fees	$0.70–0.85
Charged in this backtest	$1.35
Regulatory (per round trip)	$0.25

A 4-leg structure trades 8 contract-sides per round trip → $11.05 per trade. A single-leg sale trades 2 → $2.95. Across the program that is ≈2,349 contract-sides and $3,299 of explicit costs per year, already deducted from every figure on this site.

Slippage stress test

Last-trade fills already reflect real transactions, but a skeptic should ask: what if every fill were worse? The table re-prices the entire program with every contract-side degraded by a fixed amount — the strongest version of the objection.

Assumption	Program net / yr	Return on $218k
Every fill worse by $0.00 / contract-side	$124,808	57%
Every fill worse by $2.50 / contract-side	$118,936	55%
Every fill worse by $5.00 / contract-side	$113,065	52%
Every fill worse by $10.00 / contract-side	$101,322	46%
Every fill worse by $20.00 / contract-side	$77,836	36%

Even degrading every single fill by $20 per contract-side — far beyond typical SPX ATM spreads — the program remains solidly profitable. The edge is not an artifact of optimistic fills.

Replication results (t-statistics)

Strategy	SPY (development)	QQQ (unseen)	SPX (unseen)
Meridian / noon structure	3.8	4.1	3.0
Afternoon Pin	3.6	4.1	2.2
Premium Momentum	2.7	2.0	2.0
Defended High	2.2	2.6	1.6

A t-statistic of 2.0 means the odds of the result arising by chance are roughly 1 in 40; at 3.0, roughly 1 in 750; at 4.0, about 1 in 30,000. Requiring every strategy to clear the bar on markets it never saw is the strongest overfitting control we know how to apply.

The kill list

Tested under identical conditions, and rejected. A backtest without a graveyard is a marketing document.

Iron condor — the most popular retail 0DTE structure: $16/year after costs.
Everyday afternoon straddle, tight stop — earned for two calm years, then died in the 2026 regime change. Its wide-stop, winged descendant (Afternoon Pin) survived; the corpse taught us why.
First-hour theta capture — negative; US mornings are hostile to sellers.
Supertrend direction — 12 parameter/timeframe cells, all rejected.
Ichimoku — cloud and TK-cross, three timeframes: nothing.
Moving-average cross spreads — both variants negative.
Bollinger direction fade — loses; the premium-selling use of the same bands is under evaluation instead.
Volatility-risk-premium gate — no parameter survived both sample halves.
Overnight momentum carry — −$2,151 in the correction alone.
Opening-gap option buying — a genuine edge on Indian indices that inverts in US market structure; the inverted version is under evaluation.
Defended-low put selling — US selloffs break support; only the high side of the level trade survived.

Data

Underlying: exchange minute bars, SPY/QQQ 2016–2026 (10.7 years, SIP consolidated tape) for signal research. Options: exchange-traded minute prints for every strike within 4% of spot, expiries 0–2 days, Jan-2024 – Sep-2026 (SPY/QQQ) and Jun-2024 – Sep-2026 (SPX) — 1.5 GB, 1,900 trading-day files, zero missing days after audit. Next: independent-vendor extension to 2010 for long-history stress tests.

Sundial Research is a research demonstration by the V6 systematic trading project. All results are backtested on historical exchange data (Jun 2024 – Sep 2026 for SPX), net of modelled commissions, exchange and regulatory fees and half the bid–ask spread. They are not live results, not a solicitation, and not investment advice. Options involve substantial risk of loss.