How to test a trading claim: the evidence register comes first

A proposed evaluation method with an explicit withdrawal of unsupported cohort performance claims.

Withdrawal of the claimed track record

Editorial correction · 9 September 2026. We are withdrawing the numerical findings previously presented as an audit of 36 influencers and 1,482 calls. This page did not provide the underlying call register and reproducible portfolio calculations. Its purported cohort results, median alpha and allocation percentages are removed.

The remaining article describes a proposed evaluation method. It establishes no creator ranking, validated allocation rule or record of repeatable alpha. The original publication date remains 13 February 2026; this correction is a separate event.

Start with a claim that could fail

“What did this person get right?” is too flexible to measure fairly after an outcome. A useful study starts with an eligibility rule: which statements count as a directional call, what instrument is identified, when the call becomes available, and what horizon it implies.

A post saying a company has an attractive future is not automatically a trade with a one-day exit. Vague, edited or deleted statements need a defined treatment. Keep exclusions visible rather than silently dropping difficult observations.

The minimum evidence register

Field	Why it matters
Original post URL and preserved public reference	Trace the statement to a source
UTC publication time and inspection time	Separate when the call existed from when it was collected
Instrument and eligible direction	Avoid retroactively choosing the asset or side
Entry rule and first executable price	Prevent using prices that were unavailable after publication
Exit rule and holding period	Prevent selecting the most flattering outcome
Costs, spread and slippage assumptions	Distinguish paper returns from implementable returns
Benchmark and position-sizing rule	Define the comparison and aggregate exposure
Exclusion reason or missing value	Make selection decisions inspectable

A register makes a study auditable; it does not automatically make its design unbiased. Sampling only surviving popular accounts, searching for successful calls, or changing the holding period after seeing returns can still distort the result.

Excess return is not automatically alpha

Subtracting a benchmark's return from a strategy's return gives a form of excess return under the chosen comparison. Calling that difference risk-adjusted alpha requires a specified risk model and defensible estimation assumptions. Benchmark choice, leverage, sector concentration and exposure timing can change the interpretation.

Report the actual return path, drawdowns, turnover and uncertainty alongside any summary. A win rate without loss sizes is incomplete; a high return without risk and implementation assumptions is incomplete too.

Precommit the test, then preserve the failures

Write down the sampling universe, execution rules, cost model, benchmark and evaluation period before calculating the headline. Where possible, reserve a later period to test the fixed procedure. Explain what new evidence would make the result fail.

A reproducible notebook should load the source register, identify exclusions, transform observations into a return path, calculate the defined statistics and expose checks. It should not turn an undocumented score into an authoritative-looking chart.

No influencer dataset or empirical result accompanies this proposed method. The new BTC/ETH downside case study demonstrates a narrower example with actual public inputs and reproducible calculations; it is not evidence of influencer skill.

Change record

Original publication: 13 February 2026. Quantitative findings withdrawn and method clarified: 9 September 2026. The unsupported numerical results remain only in version history. AI-assisted editorial review; no independent human expert review.