Trading-signal metrics: the register comes before the result

A dated correction withdrawing unsupported provider and signal-cohort findings while retaining metric definitions as a proposed method.

Editorial correction: metric claims need an inspectable register

Editorial correction · 9 September 2026. The original February 14 article said that it audited 2,760 timestamped signals from 64 providers and that a composite signal-quality score reached 68% forward precision. The page did not link the provider universe, preserved signal records, collection times, outcome rows, calculation code or an independent check. Those records could not be recovered from the public page or its broad references during this review.

We therefore withdraw the cohort result, the comparison with hit-rate filtering, the reported forward returns, the metric weights and the numerical pass/watch/fail zones as empirical findings. The original URL and publication date remain; this correction is dated separately. Nothing below establishes a provider ranking, a predictive threshold or a trading allocation rule.

Definitions that remain useful as a proposed method

These terms can make a future study clearer, but they are definitions rather than validated findings:

Metric	Proposed definition	What still has to be shown
Hit rate	Winning eligible calls divided by eligible calls	Eligibility, payoff sizes and loss handling
Precision	Share of a preselected group meeting a predeclared future outcome rule	Selection rule, outcome window and the full denominator
Profit factor	Gross winning PnL divided by gross losing PnL	Fill prices, costs, partial fills and whether the result is gross or net
Sharpe ratio	Mean excess return divided by return volatility under a stated frequency and risk-free convention	Return series, annualization choice and estimation uncertainty
Maximum drawdown	Largest peak-to-trough decline in the defined equity curve	Starting wealth, compounding, missing observations and recovery treatment
Call completeness	Eligible calls containing the fields required by the study	The required fields, exclusions and an audit of missingness

No numeric cutoff is promoted here. A replacement study should predeclare its definitions, sampling frame, feature weights and future evaluation window before looking at the outcome, then preserve the procedure and its failures.

Minimum register for a replacement study

Provider identifier, public profile URL, inclusion rule and the full candidate universe.
Original signal URL or capture, UTC publication time and UTC collection time, including an explicit treatment for edits and deletions.
Instrument, direction, entry rule, first executable price, invalidation, exit rule and holding horizon.
Price source, bar convention, spread, fees, slippage, funding, currency, sizing and benchmark conventions.
Raw outcome rows, exclusions with reasons, calculation code, versioned outputs and an independent arithmetic or notebook check.

Only after those fields are fixed could a study compare hit rate, payoff, drawdown or a composite score in a later window. A general definition of a metric cannot stand in for the missing provider and signal rows.

Source context and limits

The original page linked the Statistics of Sharpe Ratios article and SEC Investor Alerts. The Sharpe-ratio article discusses estimation error in a statistic; the SEC page is a general investor-alert index. Neither contains this article's provider universe, signal register, forward labels or calculations. A cited DOI for a paper on financial charlatanism did not resolve during the 9 September 2026 review, so it is not treated as evidence for the withdrawn claims.

Change record

Original publication: 14 February 2026. The unsupported 2,760-signal, 64-provider and 68%-precision findings were withdrawn and the proposed method clarified: 9 September 2026. The original wording remains in version history and is not promoted as current research. AI-assisted editorial review; no independent human expert review.