Free check · gold signals

Does that gold signal beat chance?

A signal channel's win rate depends first on its distances. With a small take-profit and a wide stop, price reaches the take-profit first most of the time whoever picks the entry. We measured it on gold: from 12,887 random hourly entries, a $3 take-profit against a $10 stop was reached first 76.1% of the time. After a $0.30 spread, those distances need 79.2% just to break even. An "80% win rate" at 30 pips take-profit and 100 pips stop is roughly what random entries score.

What random entries score at common distances

Every hourly close of gold from 13 May 2024 – 5 Oct 2026 (12,887 entries over 697 days), bought and sold. Each entry counts as reaching take-profit first or the stop first within the next 48 hourly bars. Measured 7 Oct 2026; these are fixed historical figures, not live and not a forecast.

• Random entries pools buys and sells at the same distances, which takes out the trend of the period (gold rose from about $2,350 to about $4,200, with large swings between) and keeps its volatility. The small grey range under it is the 95% interval. With equal distances the pooled share is exactly 50%: a buy's take-profit is a sell's stop.
• Buys / sells differ by a few points. Part of that is a known fault in the futures price series at contract rolls, the rest is this period's path; it changed between sub-periods and is not something to trade.
• Net per trade is what random entries kept after the spread, as a share of the amount risked (−4.1% means 4.1 cents lost per dollar at risk). No row is clearly above zero: the best, $20 / $10 at +0.9%, is inside its own uncertainty.
• Pips at $0.10, the usual gold pip in signal channels; some brokers count $0.01. At $50 / $25, 15% of entries reached neither level within 48 hours and are left out, which is why that row sits below the formula.

How to read a channel's claimed win rate

1. Write down the take-profit and stop distance in dollars, from the entry the channel posts. Use the first take-profit if there are several, because that is the one the win rate usually counts.
2. Work out what the distances alone give: stop ÷ (take-profit + stop). $3 against $10 is 10 ÷ 13 = 76.9%. The table shows that real gold entries land close to this.
3. Add your costs: (stop + spread) ÷ (take-profit + stop) is the share needed to break even. With a $0.30 spread: 10.3 ÷ 13 = 79.2%.
4. Compare the claim with the break-even, not with 50%. A claimed 80% at $3 / $10 is 0.8 points over break-even: about +$0.10 per trade per ounce before slippage, commission and swaps, if every signal was counted.
5. Check what was counted. Signals deleted or edited after the fact, "TP1 hit" counted as a win when the stop was hit later, stops moved to entry and called break-even, and screenshots of good weeks only all push a reported rate up.

How many signals does a claim need?

Even an honest record carries luck. For a true rate of 80%, chance alone moves the observed rate within about this band (95%):

So 80% over 100 signals cannot be told apart from the 76% random entries score at $3 / $10. It takes several hundred signals, all of them counted, before a few points over break-even mean anything.

The spread matters most for small take-profits

The spread is paid on every trade, so it eats a larger share of a small take-profit: $0.30 is 10% of a $3 take-profit and 0.6% of a $50 one. Spreads also widen around news and the daily rollover.

One signal service we tested

In October 2026 we scored one signal service: its first signal per market per day on 15 forex, metal and energy markets, May – Oct 2026. Its take-profit was typically a quarter of its stop distance or less (median), and on average its distances alone implied a 72.6% hit rate, which is why its record looked strong.

A 72.6% hit rate is not evidence of skill when the distances alone ask for 72.6%. Both halves of the period lost after costs, although the loss is not clearly beyond chance on this sample. One service and under five months is a small sample, so this describes that service in that period, not every service.

What this page does not say

• It scores nothing live and checks no particular channel. The table is random entries, measured once on 7 Oct 2026.
• It does not say a signal is good or bad. A service could have an edge; its win rate alone cannot show it, only a win rate measured against its own distances and costs, over many signals.
• It is not a forecast and not investment advice. Past frequencies on past prices.
• We sell no signals and recommend no service. Ampeld models where the crowd's stops may sit, and publishes what our own tests find, including when they find nothing.

Method

• Data: hourly bars of the front gold future (COMEX, via Yahoo Finance), 13 May 2024 – 5 Oct 2026. A broker's spot gold differs by the futures basis and its own spread; distances in dollars are the same.
• Entries: every hourly close, as a buy and as a sell, with no selection of any kind. Take-profit and stop are fixed dollar distances from that close.
• Outcome: which level the bars' highs and lows reached first within 48 hourly bars. When both were inside one bar the order is unknown and the entry is left out (16.5% at $3 / $10, 7.4% at $10 / $10, 1% or less at $30 / $15 and wider); a check on 15-minute bars moved the shares by under 2 points. Entries that reached neither are left out too.
• Contract rolls: around each roll the futures series mixes two contracts about $40–60 apart. 131 bars with a wick to the other contract were found by a rule fixed before the results, and the 819 entries near them left out; looser and no cleaning moved the pooled shares by under 1 point.
• Costs: a flat $0.30 spread per trade. Slippage, commission, swaps and gaps through the stop are not counted, so real results are worse than the table.
• Uncertainty: entries an hour apart share most of their path, so the 95% intervals come from resampling whole days. Sub-periods (2024–25, 2025–26, 2026 alone) stayed within about 3 points of the formula, except $50 / $25 in calmer 2024–25 (27%), when more entries reached neither level in 48 hours.
• Only the first take-profit is scored against the stop. Channels with several take-profits, or that move the stop after the first one, need their own arithmetic.

The same approach - every signal counted, against what chance gives at the same distances - is how we score the textbook indicators on each chart (see does RSI work? and the EMA crossover test) and our own levels on the methodology page.

Educational content, not financial or investment advice and not a recommendation to buy or sell anything. Historical frequencies on 13 May 2024 – 5 Oct 2026 gold futures data, measured 7 Oct 2026. Trading leveraged products such as CFDs carries a high risk of losing money. To size a gold trade with your own stop, use the gold position size calculator; for how often a stop at a given distance gets hit, see gold stop-loss hit rates.