How to evaluate a crypto signal provider's accuracy claim

Use this seven-step sequence before reading testimonials or rankings

The order matters. Calculating a percentage before defining the evidence population creates false precision. Asking for an account statement before establishing which signals belong to the claim can also create a disconnected proof packet. Move from population to chronology, then outcomes, execution, costs, account path, and evidence type.

1. 1. Freeze the population

   What exact channels, products, pairs, access tiers, and dates are included, and when was that rule fixed?

   Failure if missing: A provider can change the denominator after outcomes are known or mix incomparable products.

2. 2. Preserve the chronology

   Can every original alert, timestamp, edit, deletion, cancellation, and closure be reconstructed?

   Failure if missing: A later result post can survive while the original instruction or losing call disappears.

3. 3. Reconcile every status

   Do wins, losses, break-even calls, open positions, no-fills, cancellations, and duplicate updates sum to the published population?

   Failure if missing: The numerator can look precise while exclusions remain invisible.

4. 4. Declare execution rules

   Which venue, price, order type, latency, entry range, target weights, stop rule, and missed-fill rule are used?

   Failure if missing: A market move is mistaken for a subscriber fill.

5. 5. Apply costs trade by trade

   Are fees, spread, slippage, funding, borrow cost, and subscription cost included at the same grain as the trades?

   Failure if missing: Gross signal movement is presented as net value.

6. 6. Rebuild the account path

   Are sizing, leverage, concurrency, compounding, open equity, peak margin, drawdown, and recovery time defined?

   Failure if missing: Summed trade percentages are treated as account return.

7. 7. Separate evidence types

   Are actual, paper-traded, backtested, provider-published, third-party, and independently reproduced results labeled separately?

   Failure if missing: A technically valid simulation or selected example is generalized into live subscriber performance.

The sequence is designed to produce a bounded decision, not a provider score. A provider may supply enough records to reproduce one claim while leaving identity, terms, refund, security, or access questions unresolved. Conversely, transparent commercial terms do not verify performance. Keep each question in its own evidence file.

Denominator lab: one set of records can produce several accurate percentages

The lab starts with the twelve-row synthetic example below. Change the counts to match a claim you are evaluating. The calculator does not decide which rate is correct. It shows how the answer changes when break-even, open, cancelled, no-fill, and duplicate records move in or out of the denominator.

A provider can accurately state a 57.1% win/loss hit rate for the synthetic sample because four of seven win-or-loss records won. The same sample has a 50.0% winner share across all resolved entered records and a 33.3% winner share across all published records. None of those rates tells you whether the sequence made money. Payoff size, sizing, leverage, overlap, costs, and the open position still determine the account outcome.

Advertised-rate lab: reverse-engineer the denominator before accepting the headline

A percentage and a win count constrain the denominator. This lab tests every possible integer denominator from the win count through the disclosed published population, then keeps only the denominators that round to the advertised percentage at the selected display precision. It answers a narrow arithmetic question: which denominators could produce this headline?

Enter the number of published records visible in the period, the declared wins, the advertised rate, and the number of decimal places shown. A unique match identifies one arithmetically compatible denominator. Several matches expose rounding ambiguity. No match means the disclosed inputs cannot reproduce the displayed percentage under this rule. None of those states verifies the underlying calls or decides whether an exclusion is legitimate.

Whole-number counts only. Rates must be from 0% through 100%. The lab evaluates at most 100,000 published records in your browser and does not save or submit inputs.

A compatible integer denominator shows only that the displayed percentage can be reproduced from the entered win count under one rounding rule. It does not establish that the denominator is complete, that exclusions are justified, that records are authentic, or that the claim represents profitability. The three cases are synthetic arithmetic fixtures, not observed provider claims. RATE-001 has one compatible denominator. RATE-002 shows that a whole-number percentage can map to more than one denominator. RATE-003 shows a percentage that no integer denominator inside the disclosed population can reproduce.

The twelve-row example keeps every exclusion visible

These twelve records are invented only to demonstrate denominator changes. They are not provider results, market observations, recommendations, or a performance forecast. R is a synthetic per-trade risk unit and is not an account return. The table exists to make the denominator inspectable. It deliberately includes the states that headline result cards often omit.

The resolved rows sum to 4.4R before any cost assumption. That figure is still not an account return. The example does not state capital allocation, concurrent positions, compounding, leverage, spread, fees, funding, slippage, or how the open record is marked. Even a positive sum can describe a path a real account could not follow at the shown size.

For a real claim, replace the synthetic rows with a complete export and preserve the original identifiers. Do not reconstruct only the calls shown on a results channel. Do not treat a result message as the original signal. If a provider publishes one alert and several target updates, establish whether those updates are one record or multiple observations before counting them.

Evidence checklist: a complete packet becomes reproducible, not automatically verified

Mark a field only when the packet contains the underlying record, not when a marketing page mentions the concept. A statement such as "fees included" is not the same as trade-level fee values. A screenshot of one timestamp is not a chronology. A chart image is not an exportable equity curve.

All nine fields create the conditions for reproduction. They do not establish that the source records are authentic, complete, current, or independently controlled. Independent verification requires a verifier to obtain or observe the evidence, apply the declared method without changing it, reproduce the result, document exceptions, and retain enough provenance for another reviewer to repeat the work.

Identify what the percentage measures before comparing providers

Signal marketing often uses one visual language for several incompatible quantities. A 70% hit rate, a 70% leveraged position gain, a 70% sum of targets, and a 70% account return do not share a denominator. Comparing them as if they were the same is a category error.

When a claim is ambiguous, ask the provider to rewrite it as one sentence containing the unit, population, period, and calculation. For example: "Four of seven eligible closed win-or-loss observations in the named channel and date window reached the declared win condition under the stated fill and exit rules." That sentence is less impressive than "57% accuracy," but it can be tested. Account return needs a different sentence and a different dataset.

Original chronology matters more than a polished result feed

A result feed is downstream evidence. It can show what the provider chose to summarize after a trade, but it may not preserve the original entry range, invalidation, stop, target weights, leverage language, publication time, edit history, or deleted calls. Evaluation should begin with the original signal stream and then reconcile every later update back to one stable signal ID.

Preserve platform-native message identifiers where possible. Record the original publication time and every edit time in one timezone. If a message was deleted, retain a tombstone stating that the record existed and when it disappeared. If an alert was cancelled, keep the cancellation reason and time. If a signal remained open at the report cutoff, do not silently remove it or classify favorable movement as a win. The cutoff should be fixed before the report is calculated.

Chronology also constrains hindsight. An entry published after the market moved cannot support the same claim as an actionable pre-move alert. A target added after price touched it should not be treated as an original target. A stop widened after adverse movement changes the strategy path. Corrections are legitimate, but they should append a visible version rather than replace the earlier instruction without a trace.

Execution rules decide whether a market move was available to a subscriber

A crypto market trades across venues with different books, spreads, liquidity, funding, tick sizes, outages, and regional access. A signal that reaches a displayed price on one venue may not fill on another. Entry ranges create further ambiguity: using the most favorable point after the fact can improve every result without changing the signal text.

Fix one fill policy before calculation. State the venue, product, side of market, order type, reference timestamp, latency allowance, and whether a trade is skipped when the quoted range is no longer available. Use the same rule for winners and losses. Record partial fills and failed orders. If a stop gaps through its level, use the declared execution rule rather than the stop label. If the record uses idealized candle highs or lows, label it as a simulation rather than subscriber execution.

Multi-target signals require weights. If 25% exits at each of four targets, the realized trade is the weighted path of those exits and the remaining stop. The furthest touched target cannot represent the full position unless the method says the full position remained open to that point. Counting every touched target as a separate win can enlarge the numerator while one underlying signal remains the unit.

Costs and risk turn signal classification into an account path

Win rate ignores payoff asymmetry. A strategy can win often and lose money if losses are larger than wins. It can win less often and remain profitable if gains are larger, but only after execution and service costs. Average win and average loss should use the same unit, preferably a declared risk unit or account currency, and should show dispersion rather than only one average.

Apply trading fees, bid/ask spread, slippage, funding, borrow cost, copier or automation cost, data or tool subscriptions, and the signal subscription at the grain where each cost occurs. Do not subtract a vague monthly estimate from gross percentages measured with a different capital basis. A fixed subscription has a larger hurdle for a smaller account and can change the incentive to overtrade.

Position size and leverage must be separated. Ten-times leverage can multiply a position-level move while only a small portion of the account is allocated, or it can expose most of the account through cross margin. State margin mode, initial allocation, maximum concurrent exposure, re-entry and averaging rules, compounding, and liquidation treatment. Then build the equity curve in chronological order.

Report maximum drawdown, longest losing streak, peak simultaneous exposure, recovery time, and open equity at the cutoff. These path measures reveal risks that a cumulative sum of positive trade percentages can hide. A large headline total does not show whether the account had enough capital to survive the sequence or whether the same capital was implicitly reused across overlapping trades.

Backtests, paper records, provider summaries, and live accounts answer different questions

A backtest can test a historical model under assumptions. It does not represent actual execution. Useful backtest evidence fixes the data source and version, strategy or model version, parameters, pair universe, date range, in-sample and out-of-sample periods, costs, slippage, latency, failed orders, and inactive periods. Repeated tuning after seeing results increases the risk that the historical fit will not persist.

A paper-traded or forward-simulated record reduces hindsight when instructions were frozen before outcomes, but it still avoids financial risk and may use idealized fills. A provider-published summary can be genuine while remaining incomplete. A connected exchange statement can support actual trading but still needs account ownership, cash-flow, strategy attribution, and privacy-safe independent review. No single evidence label should be generalized into another.

For AI-assisted predictions, preserve model version, feature timestamps, prompt or policy version, decision threshold, revision behavior, and the exact rule that converts a probability into a signal. A later model update should not silently recalculate historical predictions. The result belongs to one version and one evaluation window.

Send one bounded evidence request instead of debating a headline

The following request is intentionally specific. It does not ask for private customer data, exchange credentials, wallet access, or unverifiable testimonials. It asks for the minimum records needed to reproduce one defined claim.

Please provide the complete signal ledger supporting the stated accuracy or performance claim for one fixed date window and named channel or product.

For every published signal, include a stable ID, original timestamp, edit or deletion history, market and venue, direction, entry rule, invalidation or stop, target weights, leverage and sizing rule, and final status: win, loss, break-even, open, cancelled, invalidated, duplicate, or no-fill.

Please also provide the eligibility and exclusion rules fixed before outcomes, the fill and exit model, fees, spread, slippage, funding and subscription-cost treatment, starting capital, concurrency and compounding rules, and the account-level equity curve with maximum drawdown.

Label actual, paper-traded, hypothetical, backtested, third-party, and provider-published results separately. A privacy-safe export is sufficient; no login, API key, customer identity, or wallet access is requested.

If the provider supplies only selected winners, a current screenshot, a testimonial, or an aggregate percentage without the underlying population, record exactly what was supplied and leave the claim unresolved. Do not fill missing losses with zero, infer timestamps from later posts, or accuse the provider based on the gap.

Three evidence states keep conclusions honest

A fourth label such as "verified provider" is too broad for this method. Verification must name the exact claim, period, dataset, method, verifier, and date. A provider can have one independently reproduced historical calculation while current terms, identity, access risk, or later performance remain unresolved.

Use the empirical audit, method, and interval for different jobs

This page explains how to test a claim. The 16-provider performance claims audit reports what a frozen cohort of researched provider dossiers actually established and which proof remained missing. Keeping the method separate from the empirical audit prevents a general checklist from being mistaken for evidence about a named provider. Before interpreting a supplied denominator, use the result-sheet completeness stress test to compare supplied, resolved, open, cancelled, and declared-gap counts under explicit equal-loss stress assumptions. That stress result is a sensitivity calculation, not evidence that missing rows were losses.

After the rows and chronology are defensible, use the concurrent exposure analyzer to measure overlapping historical intervals and summed planned risk within each exact risk-basis ID. Its output describes the supplied schedule; it does not establish account drawdown, provider verification, or future risk.

After a complete binary denominator is defensible, use the win-rate confidence calculator to quantify interval uncertainty without relabeling it as verified performance or profitability. Use the verified result-sheet boundary to understand what CSR will and will not call verified. Use the Signal Passport proof template to assemble provider evidence, the result-sheet guide to structure outcomes, the risk and reward guide to interpret payoff asymmetry, and the alert-to-order attribution library when the missing link is execution evidence.

The historical accuracy URL had demand but redirected to a broader page

Authenticated Search Console showed 128 impressions, 0 clicks, and an average position of 11.6 for this exact historical route in the measured three-month window. URL Inspection recorded the page as "Page with redirect" because it returned HTTP 301 to the broader methodology page. The crawler fetch succeeded, and Google selected the redirect target as canonical.

Restoring the exact route with a focused, interactive evaluation method preserves the observed page association rather than creating another competing slug. These Search Console values explain topic selection only. They do not prove future ranking, indexing, search volume, or that every impression had the same intent.

Official performance guidance supports the evidence questions, not a provider verdict

The CFTC advisory notes limitations in hypothetical trading, including assumed prices, spreads, execution, costs, and whether traders can withstand losses. The NFA interpretive notice explains that hypothetical results do not represent actual trading, can reflect hindsight, may under- or over-compensate for liquidity and slippage, and should disclose material assumptions. These sources concern specific regulatory contexts. CSR uses them only as general measurement guidance.

• CFTC: Commodity Trading Systems Sold on the InternetAccessed 2026-07-12 | HTTP 200 | 44,430 bytes

  General context on hypothetical trading, execution assumptions, spreads, fees, independent verification, and the difference between simulated and actual results.

  Boundary: The advisory is not specific to crypto signal providers and does not evaluate any provider covered by CryptoSignalsReview.

  sha256:35a02c7f7b4b58c4a7ae7176c464b234b653e5cd3f166f135599764c46219a52
• NFA Interpretive Notice 9025: Hypothetical Performance ResultsAccessed 2026-07-12 | HTTP 200 | 158,080 bytes

  General context on hindsight, liquidity, slippage, financial risk, material assumptions, actual-versus-hypothetical separation, and presentation prominence.

  Boundary: The interpretive notice applies to specified NFA members and promotional material. This guide uses it as measurement context, not as a legal finding about a provider.

  sha256:67603c40f5d552e437192af0353d7526270bf38b361ee80c1e430623beb338ce

No official source listed here evaluates a CryptoSignalsReview provider, validates the synthetic example, endorses this site, or converts missing evidence into a legal conclusion. Regulatory applicability depends on facts and jurisdiction. This page is not legal advice.

Method, source capture, and reusable files

The generator reads one fixed JSON source, validates the seven decision steps, six claim types, nine evidence fields, two official source captures, twelve synthetic records, three advertised-rate compatibility fixtures, and the historical Search Console observation. It recalculates every displayed synthetic metric and every compatible integer denominator instead of trusting stored totals. It then writes the canonical page, WordPress mirror, method summary, synthetic ledger CSV, dedicated one-URL sitemap, LLMS summary, manifest, and isolated auxiliary plugin.

The source capture hashes identify the exact response bodies collected on 2026-07-12. A hash can show that a stored response has not changed; it does not prove that every statement on the source page is correct or that the page remains current. The Search Console snapshot is first-party route-selection evidence and is not a public keyword-volume estimate.

Reuse boundary: the method, checklist, and synthetic dataset may be cited with attribution to CryptoSignalsReview, the canonical URL, and the research date. Reuse must preserve that the examples are invented, arithmetic compatibility is not verification, past results do not guarantee future performance, and missing proof is not an accusation.

Continue with the right evidence surface

• Compare 21 evidence files

  Provider-specific identity, access, pricing, result, and risk boundaries. Coverage is not endorsement.

• Read the full CSR methodology

  How evidence changes status, how uncertainty stays visible, and what paid work cannot influence.

• Submit or correct evidence

  Provide attributable records without buying status, ranking position, or a positive conclusion.