The assumption nobody measured

Proof card

Experiment: ALPHA-1A-EXP6-TREND-BASE · manifest 323828d4887c Sealed research conclusion: ALPHA-1A INCONCLUSIVE · investment conclusion: NO PROMOTION

Hypothesis

Under a flat ticks-per-side slippage model, a market whose tick is large relative to its typical daily move will have its backtest result dominated by the slippage assumption rather than by its signal.

Evidence

• Same trend signal, with the experiment's 100-day lookback as defined in its sealed manifest, applied to 8 futures markets, 2020–2026.
• ZN gross result before costs: +$122,656.
• Slippage is charged per side, meaning on every buy and every sell. A roll is charged as a full close and reopen, each leg at the full per-side cost, where per-side cost is commission plus slippage as the experiment's cost model defines it.
• At 1 tick per side: ZN nets +$58,077.
• At 3 ticks per side: ZN nets −$53,267.
• The swing between the two assumptions is $111,344, or 91% of the gross result.
• ES, GC and NQ were profitable at the low assumption and changed by less than 10%.
• YM and ZN change sign between the two assumptions. No market's result is statistically distinguishable from zero at either assumption (see Statistical weight).

In-sample vs holdout (ZN, net)

The positive low-slippage total comes entirely from the in-sample period. In the holdout, ZN lost money under both assumptions.

Statistical weight

• ZN: t ≈ 0.31 at low slippage, −0.29 at high. Neither result is statistically distinguishable from zero (|t| < 2).
• No market's result reaches |t| ≥ 2 at either level; the largest is GC at 1.78.

Method

Sealed experiment outputs, read by pinned blob hash; daily P&L reconciled to the sealed scorecards for every market at both slippage levels, and period sums reconciled to totals.

Diagnostics (post-hoc, not preregistered):

• Typical daily move: ZN moves 26 ticks on a typical weekday session, against 182–983 for the other markets. Counting all rows including Sundays gives 24 for ZN; the ranking does not change.
• Significance: t = mean over standard error of daily net P&L from the first position onward. It ignores autocorrelation, so it is approximate.

Evidence classes of the inputs

• Slippage: ESTIMATED (flat placeholder, not measured)
• Bid/ask spread: MISSING
• Commission: ESTIMATED

Failures and limits

• Slippage was never measured. A flat-tick model penalizes coarse-tick markets by construction, so part of this result may reflect the cost model's shape.
• Neither ZN result is statistically meaningful, so the swing shows sensitivity to an assumption, not a change in a real edge.
• The explanatory diagnostics were computed after the fact.
• The daily rows include Sundays, and a post-closure finding logged for this experiment confirmed that it treats them as trading days: its lookback and volatility windows count rows, and sizing and annualized figures scale per row. Positions ran larger than designed for most markets, and not by the same amount for every market.
• What that changes: the absolute dollar figures on this card come from the sealed results as recorded and are affected by the sizing error.
• What it mostly doesn't: the error mostly rescales positions, so ratios such as the swing's share of the gross result, and the t-statistics, are expected to be largely unchanged. The experiment was not rerun, so this is an expectation, not a measurement.
• The research conclusion is expected to stand.
• Backtest on $5M of hypothetical research capital; no live execution.

Out-of-sample / forward test

Holdout split shown above. No forward test.

Conclusion

SUPPORT WITH CONDITIONS, for a methodological claim only: in this experiment the unmeasured slippage assumption moves ZN's result by an amount comparable to its entire gross result. No claim is made that ZN is or is not tradeable with this signal.

Confidence

Not scored. The deciding input is classified ESTIMATED.

Not investment advice.