A trader with no edge whatsoever can pass a retail prop evaluation 36% to 47% of the time through position size alone.
The product has two stages. A simulated 50K account costs $87 to $228, with a $3,000 profit target and a $2,000 maximum loss level. The loss level follows the account peak and updates only at session close, remaining static intraday. Reaching the target converts the account to funded status. From there, a minimum-days requirement and a per-cycle withdrawal ceiling apply. Many firms add a consistency rule that limits how much of total profit may come from any single day. Each condition lies between the participant and a payout.
Hall models both stages with Monte Carlo, then subjects the most prominent public practitioner's fully specified rule set to the same process as a coin flip. He uses continuous front-month index futures at one-minute frequency from June 2010 to June 2026, roughly 5.3 million bars. The sample was split a priori at 7 September 2021. Across three accounts and two implementations, break-even falls between a 40.5% and 41.5% win rate at 1:1.5 net of costs. The driftless baseline is 40.0%.
Why the pass rate splits in two
With end-of-day monitoring and unconstrained sizing, the evaluation converges toward a single coin flip. Raise daily volatility far enough and a session close either clears +$3,000 or breaks the floor. The intraday path goes unseen. Symmetry drives pass probability toward 0.5. Under continuous monitoring, the driftless benchmark is d/(d+T) = 0.400, while discrete observation and overshoot lift the result above that level. Even after adding a $1,000 daily loss limit, the zero-skill optimum remains 0.25 to 0.46.
Hall still treats the gate as informative. At every cadence, pass probability increases monotonically with skill. Consider five trades a day and fixed sizing of $90. Zero skill produces 0.101; a genuine edge of q = 0.05 produces 0.394, a gain of twenty-nine points. Optimising size while leaving skill at zero raises the same 0.101 to 0.434, or thirty-three points.
Roughly equal weight.
The zero-skill estimate applies per attempt. Its closest cohort comparison is therefore Topstep's account-level disclosure for calendar 2025, in which 16.8% of initiated evaluations finished. The observed population performs worse than sizing alone. Hall interprets the difference as negative drift after costs, combined with sizing well below the optimum. Published failure attributions agree with that reading. One source assigns around 70% of failures to loss limits. Another assigns 45% to 55% to the daily loss limit and 20% to 30% to trailing drawdown. None of that data identifies failure to reach the target as a leading cause.
Fast through evaluation, slow when funded
The loss level stays fixed during the session, so additional intraday trades do not compound that session's drawdown risk. At a 40% win rate, evaluation pass rates barely move with cadence: 23.4% at one trade a day and 25.4% at twenty. Resolution time falls sharply, from 11.4 days to 1.6.
Funded accounts reverse the result. Holding the win rate at 40%, payout probability falls from 34.8% at one trade a day to 3.9% at twenty. The five-day minimum makes each additional trade another opportunity to hit the floor. At the break-even win rate, running twenty trades a day in both stages yields a joint gate of 1.0%. A split cadence, fast during evaluation and slow after funding, raises the joint result to 8.8% under the same strategy parameters.
This gives a mechanism for a practitioner convention that is widely followed and rarely explained. It also exposes the conflict between stages. The evaluation rewards lumpy, aggressive behaviour, while the payout consistency rule penalises it. Hall separately isolates the trailing ratchet by comparing it with a fixed two-barrier account. Across three representative cells, the ratchet costs 14.0 to 16.6 percentage points of pass probability.
Nobody publishes the second number
Among more than twenty identified firms, eight publish a pass rate. Only three publish a conditional payout rate. Those disclosure counts support the seller model, although Hall concedes that the payout arithmetic built on them is thin.
There are Four external values: 33.3%, 28.6%, 45%, and around 20%. They vary by better than two to one across asset classes and years, and publishing firms may be favourable outliers. The denominators also differ. Topstep reports its 16.8% pass rate per account, its 33.3% payout rate per person, and 51.8% of individuals passing at least once. Hall flags that an earlier draft of his own multiplied figures drawn from the two denominators.
The funded-stage payout rate provides the stronger check. Hall's simulation gives 34.8%, while his independent solver gives 34.6%. Both are above Topstep's 33.3% and MyFundedFutures' 28.6%, yet each falls within six points of those figures from populations he could not influence. Swiset reports 20.3% for one-phase evaluations and 11.8% for two-phase evaluations, based on about 10,000 traders across eleven regions. Hall presents this strictly as a within-study association. There is no control for firm composition, account size or pricing. That treatment fits the evidence available.
Zeroing the fee leaves $5.61 per account
Hall's direct zero-fee test produces expected gross extraction of $5.61 per account. The Apex 50K, at a $30 fee, was the cheapest and least restricted evaluation he could source. In the Gaussian daily model, a zero-drift participant reaches +$28.38 per account before costs. Charge $50 a day during evaluation and $20 a day after funding, and that same participant falls to minus $11.98.
The paper relies on the trade-level rebuild, which eliminates the cost-free positive cell. At the driftless 40% win rate, the outcomes are minus $0 at four trades a day with $400 risk and minus $25 at six trades a day with $250 risk. At 42%, both become positive: +$82 and +$16. Measured round-turn friction comes to $2.24 on the micro contract and $13.80 base on the full-size contract. Hall's best-ever bar-level gross edge was $1.74 per trade. A Cycle II geometry sweep peaked at +0.318 points gross. After a $3.00 round turn, it left minus $2.37 net per contract.
The hedged-pair experiment uses two accounts with equal and opposite positions, symmetric sizing and mirrored daily profit and loss over a ninety-day horizon. Each cell contains 200,000 paths. In every tested cell, cost per funded account matches a single unhedged account to three significant figures. With $1,000 daily sigma and $20 a day in costs, the comparison is $332 against $331. P(both pass) is 0.000. P(at least one) consequently doubles exactly, purchased with exactly double the fee.
The surviving account remains exposed. Because the $3,000 target exceeds the $2,000 drawdown allowance, a reversal large enough to kill the losing account can later kill the winner. The payout ceiling binds: +$9 per pair at the observed $1,000 ceiling, minus $93 with a 20% consistency cap, and +$215 at a $2,000 ceiling. Hall assumes perfect simultaneous mirrored fills and assigns no detection probability. He notes that detection forfeits the payout rather than the fee.
Does the null travel?
Hall draws the boundary himself. The abstract confines the null to the measured strategy universe, the paper's trade-level cost model and the observed drifts. Section 10 begins by saying the claim that no cost-surviving edge exists is false as written. His search includes 127 tracked investigations, of which 88 were resolved, plus around two dozen mechanism-first candidates. Across the three cycles, he finds no deployable edge compatible with an evaluation. For bar-level strategies specifically, there were zero net-positive out-of-sample results over 2021 to 2026. Hall says the 88 investigations are not statistically independent and claims no multiple-testing correction.
Two near-survivors remain in the ledger. A leveraged-ETF end-of-day convexity short records an out-of-sample Sharpe of 3.07 over 60 trades. A holiday-eve de-risking effect records 3.15 over 53 trades, with its decisive matched control still pending. Together they produce roughly nineteen trades a year and about $560 at one micro contract. Their cadence makes them unusable for this product. Nineteen trades a year cannot reach a $3,000 target within an evaluation horizon, as Hall states directly.
The published rule set receives 324 in-sample configurations, one out-of-sample validation, full costs and a coin-flip control. Its result is n = 1,217, a 38.54% win rate and minus $32.14 per trade. The control passes 20.50% of the time, compared with 15.25% for the rule set. Across all three accounts, the joint gate agrees with the control to within 0.10 percentage points. The practitioner claims 57.5% for the same rule set. Hall leaves the nineteen-point gap unattributed because his data cannot explain it.
We could not reproduce the cost side. Our futures data consists of one-minute OHLCV and contains no prints, quotes or spreads. The $2.24 micro and $13.80 full-size friction estimates carrying the null are therefore not recoverable here. Our universe includes dated ES and lacks the equity-index micro roots. As a result, sizing against a $2,000 trailing drawdown is a different object.
A second futures firm publishing full cohort statistics on the same denominators as Topstep would change the assessment. Hall himself says that evidence would be worth more than further simulation. The firm's own figure remains the hardest to dismiss: 0.71% of funded participants reached a live capital account.