Research

We tested our own trading signals. They lost money.

Most signal services show you their wins. We wanted to know the real answer, so before looking at any results we wrote down the rules for what would count as "working", then tested on thousands of past setups.

Test 1

Our intraday setups

Across 5,774 past long setups that reached their target or their stop, on 26 liquid crypto pairs (Jan 2022 – Oct 2026), the setups won 25% of the time but needed 29% to break even. Average result: −0.38R per trade after low fees (0.20% round trip), and −1.69R at typical retail fees (1.20% round trip). In the original test, run before we looked at the most recent months, none of the 226 strategy, direction and timeframe combinations we could judge passed.

Test 2

Daily trend-following

12 versions tested on 2016–2024. One passed, then failed on 2025–2026 data it had never seen (−0.079R per trade after fees).

Test 3

"Hold BTC/ETH in uptrends, cash otherwise"

5 rules × 2 coins. None cut the worst crash by the 40% we required: the deepest falls were 64.3–87.4%. They did better in 2022–2026, but a rule that only works in the period you happen to like isn't a rule.

What "R" means

R is the amount you risk on a trade: the distance from your entry to your stop. −0.38R means that for every $100 risked per trade, you lost about $38 on average.

The lesson that surprised us most: fees

Short-term setups use tight stops. When your stop is 0.7% away and a round trip costs 1.20% in fees, fees alone cost you more than your whole risk. Try the calculator below with your own numbers.

Free tool

What do fees cost you per trade?

For a long trade. Everything runs in your browser; nothing is sent anywhere.

Simplified: both legs priced at entry.

Why we publish this

SignalForge now shows the real track record next to every setup, including the bad ones. We'd rather you trade less than trade blind.

Every record

Track records by setup

"All setups" plus every setup with at least 30 past cases. Same columns and labels as in the app.

Setup	Past cases	Expired	Won	Break-even	Average R after fees	At retail fees	By year	Label
All setups	5,774	1,499	25%24% to 26%	29%	−0.38R−0.43R to −0.33R	−1.69R	2022: −0.26R (n 162) 2023: −0.08R (n 193) 2024: −0.27R (n 1,270) 2025: −0.38R (n 1,624) 2026: −0.48R (n 2,525)	Has lost money historically
Momentum breakout · long · 15m	1,690	466	23%21% to 25%	29%	−0.35R−0.41R to −0.29R	−1.36R	2024: −0.33R (n 677) 2025: −0.34R (n 609) 2026: −0.40R (n 404)	Has lost money historically
Momentum breakout · long · 5m approximate	1,367	294	23%21% to 26%	29%	−0.59R−0.65R to −0.53R	−2.74R	2025: −0.62R (n 304) 2026: −0.57R (n 1,063)	Has lost money historically
Momentum breakout · long · 1h	620	179	24%21% to 28%	29%	−0.23R−0.31R to −0.14R	−0.72R	2022: −0.36R (n 94) 2023: −0.27R (n 97) 2024: −0.16R (n 171) 2025: −0.16R (n 153) 2026: −0.28R (n 105)	Has lost money historically
Multi-timeframe continuation · long · 5m approximate	389	67	25%21% to 29%	29%	−0.58R−0.73R to −0.46R	−2.84R	2025: −0.50R (n 68) 2026: −0.60R (n 321)	Has lost money historically
Breakout retest · long · 15m	379	121	20%16% to 24%	29%	−0.43R−0.53R to −0.33R	−1.38R	2024: −0.32R (n 149) 2025: −0.50R (n 122) 2026: −0.50R (n 108)	Has lost money historically
Breakout retest · long · 5m approximate	344	63	27%23% to 32%	29%	−0.45R−0.64R to −0.26R	−2.49R	2025: −0.53R (n 78) 2026: −0.42R (n 266)	Has lost money historically
Multi-timeframe continuation · long · 15m	271	66	33%27% to 38%	29%	−0.10R−0.27R to +0.05R	−1.10R	2024: −0.25R (n 118) 2025: +0.01R (n 99) 2026: 0.00R (n 54)	No clear edge either way
Momentum breakout · long · 4h	157	98	27%21% to 35%	29%	−0.10R−0.26R to +0.06R	−0.34R	2022: −0.01R (n 18) 2023: −0.15R (n 26) 2024: +0.12R (n 35) 2025: −0.42R (n 45) 2026: +0.01R (n 33)	No clear edge either way
Breakout retest · long · 1h	144	52	24%17% to 31%	29%	−0.23R−0.42R to −0.03R	−0.73R	2022: −0.46R (n 27) 2023: −0.07R (n 22) 2024: −0.12R (n 40) 2025: −0.10R (n 32) 2026: −0.53R (n 23)	Has lost money historically
Multi-timeframe continuation · long · 1h	144	26	37%29% to 45%	29%	+0.10R−0.10R to +0.29R	−0.44R	2022: +0.62R (n 10) 2023: +0.47R (n 31) 2024: −0.33R (n 31) 2025: −0.09R (n 34) 2026: +0.20R (n 38)	No clear edge either way
Liquidity sweep reversal · long · 15m	42	1	38%25% to 53%	36%	−0.24R−0.63R to +0.15R	−1.78R	2024: −0.15R (n 12) 2025: −0.38R (n 18) 2026: −0.13R (n 12)	No clear edge either way
Breakout retest · long · 4h	41	31	22%12% to 37%	29%	−0.18R−0.41R to +0.07R	−0.40R	2022: −0.62R (n 4) 2023: +0.11R (n 11) 2024: −0.04R (n 7) 2025: −0.32R (n 12) 2026: −0.36R (n 7)	No clear edge either way

How we measure this

A track record shows what happened when SignalForge's current rules were replayed over past market data. It is a count of past outcomes, not a prediction, and it is separate from a setup's rule match.

What a replay is

We fed past Coinbase candles, one bar at a time, through the same scanner code that runs live, and recorded every setup it would have published. Each setup was then followed on later candles until it reached its take profit, hit its stop loss, or expired. When one candle touched both levels, we counted the stop.

What is covered

Setups created from Jan 29, 2022 to Oct 10, 2026 on 26 pairs: AAVE-USD, ADA-USD, AERO-USD, ARB-USD, ATOM-USD, AVAX-USD, BCH-USD, BTC-USD, DOGE-USD, DOT-USD, ETH-USD, HYPE-USD, ICP-USD, INJ-USD, LINK-USD, LTC-USD, NEAR-USD, PUMP-USD, QNT-USD, SOL-USD, STRK-USD, SUI-USD, TIA-USD, UNI-USD, XRP-USD, ZEC-USD. Timeframes: 5m, 15m, 1h, 4h. Long setups only, because short setups are switched off in the live scanner. Records on 5m are marked approximate (see below).

Fees

Every average subtracts a 0.20% round-trip fee, converted to R as the fee divided by the setup's stop distance. Expired setups pay the fee too. The second line on each card repeats the calculation at 1.20% round trip, which is typical for retail taker orders (about 0.6% per side). Tight stops make fees large in R, so short timeframes are hit hardest.

Why history may look better than it was

The pairs were chosen because they are listed and liquid on Coinbase today. Coins that faded away or were delisted are missing, and that survivorship tends to flatter the past.

How closely the replay matches live

We compared the replay with what the live scanner actually published. From Sep 14, 2026 to Oct 10, 2026, the replay also found 52 of the 56 live setups (92.9%). On 15m bars that live verifiably scanned (Sep 14, 2026 to Sep 28, 2026), 21 of 24 replay setups matched a live one (87.5%). 5m precision could not be verified against live (every 5m bar with live scan evidence was a bar where live itself signalled), so 5m records are approximate.

How to read a record

Past cases counts setups that reached their target or their stop. Expired setups are listed separately.
Won is the share of past cases that reached the target. Break-even needs is the win rate the setup's reward-to-risk requires just to break even.
Average R per trade is the average result in units of the risk taken, after fees, across every resolved setup including expired ones. The range in the table is a 95% interval from resampling whole calendar months.
Each record gets exactly one label, decided in this order: fewer than 30 past cases is "Not enough history yet"; a whole 95% range below zero is "Has lost money historically"; a whole range above zero is "Has made money historically"; anything else is "No clear edge either way".

Past results do not guarantee future results. Markets change, and your own fills, fees and slippage will differ. SignalForge provides information and tools, not financial advice.

Data generated Oct 12, 2026 from replay commit 2100547. See the records