Black Friday: all 12 tools. Until 1 December.Black Friday $606.98 one by one

$606.98 one by one $49 once Save $557.98Get the SuiteGet it
ProEA Lab · MT5, TradingView and trading research · Evidence standards
Systems

We turned the TTrades-style Fractal Model into rules and tested 108 versions on 7 markets. Before costs, it was a coin flip.

C2 sweeps C1, the lower timeframe confirms with a CISD, the candle closes back: we coded the public model as one mechanical rule, wrote the test down before running it, and ran it on 8¾ years of gold, FX, indices and Bitcoin. Every number, the losing ones first.

PL · 7 min read
Native TradingView capture, OANDA:XAUUSD on the 15-minute chart with Fractal Model Lab v1.0.0: C2 sweep labels, dashed CISD levels from each extreme, and the Lab's record panel.

A trader asked us a simple question.

"Is your Fractal Model Desk like TTrades' fractal model?"

It is built from his public material. So we asked the question behind the question.

Does the model, written as rules, make money?

We wrote the test down first, then ran 108 versions on seven markets from January 2018 to September 2026. Nothing passed.

This article is where the numbers live. The desk page carries none of them.

The rule, in one replay

The model (explained here, with a one-page cheat sheet) starts when one 4-hour candle (C2) sweeps the last one (C1). The 15-minute chart then has to close back through the open of the run that made the extreme (the CISD), and C2 has to close back inside C1. The stop sits at the extreme; the target is C1's other end.

The same sweep-and-turn runs on a clock in ICT's Judas swing: the Asian range instead of C1, and a close back through the midnight open instead of the CISD.

The rule, one setup at a time.Setup 1 of 10 · long · confirmed entry, C1 target · a simulated market

Simulated 15-minute candles, not market data, run through the study's rule. Setup 1 is the first setup whose trade reaches C1's far side, so every step shows; Another setup plays the rest in order, whatever they did. R is after a 0.40 round-trip cost.

Read as text
  1. C1, the four-hour candle, closes at 12:00 UTC: high 2394.41, low 2382.89.
  2. 12:15: a candle of C2 trades below C1's low. A long setup starts.
  3. 15:30: a 15-minute close above 2386.36, the open of the run of down candles that made the low (2377.80).
  4. C2 closes back inside C1's range: the swing is confirmed.
  5. 16:00: entry at 2390.96, stop 2377.80, target 2394.41 (C1's high).
  6. 19:30: the target fills at 2394.41: +0.23R after a 0.40 round-trip cost.

That is the whole skeleton, and it is what we tested. Three timeframe pairs, three entries, three targets and four filters make 108 versions. The part TTrades adds by hand (a point of interest, a daily bias, an objective) cannot be turned into rules, so it was not tested.

Without the CISD, the same skeleton (a candle sweeps the last one's high or low and closes back inside it, and the target is its other side) is what traders call candle range theory (CRT).

How we measured it

The test was written down and fingerprinted (SHA-256) before the first run, so no rule could be bent after we saw a result. We chose on 2018–22, checked on 2023–24, and ran one final test on January 2025 to September 2026.

  • Data: 5-minute bars, about 4.7 million of them: XAUUSD, EURUSD, GBPUSD, USDJPY, Nasdaq 100 and S&P 500 CFDs (Dukascopy) and BTCUSDT (Binance).
  • Costs per round trip: gold 0.40, EURUSD 1.0 pip, GBPUSD 1.4 pips, USDJPY 1.0 pip, Nasdaq 2.0 points, S&P 0.6, Bitcoin 0.12%. A stressed run used 1.5×. No overnight financing.
  • Fills: decide at the close, fill at the next open; if a candle touches both the stop and the target, we count the stop.

Which versions made money?

Almost none, and never the same ones twice. Flip the periods and watch the grid.

Which of the 108 versions made money after costs?2018–2022: 9 of 108 above zero. Best +0.032R per trade (Daily on 1H · Retest · 2.5× leg · no filter), worst −0.219R.
C1 target
2.5× leg
4× leg
None
Time
Fast
Bias
None
Time
Fast
Bias
None
Time
Fast
Bias
1H on 5mEarly CISD
Confirmed
Retest
4H on 15mEarly CISD
Confirmed
Retest
Daily on 1HEarly CISD
Confirmed
Retest
Table: all 108 versions, R per trade after costs
Version2018–20222023–20242025–26
4H on 15m · Early CISD · C1 target · no filter−0.117 (21,275)−0.141 (8,733)−0.113 (7,740)
4H on 15m · Early CISD · C1 target · Time−0.097 (15,211)−0.155 (6,149)−0.110 (5,540)
4H on 15m · Early CISD · C1 target · Fast CISD−0.123 (18,264)−0.148 (7,665)−0.123 (6,772)
4H on 15m · Early CISD · C1 target · Daily bias−0.141 (11,882)−0.168 (4,868)−0.135 (4,306)
4H on 15m · Early CISD · 2.5× leg · no filter−0.105 (22,168)−0.124 (9,222)−0.099 (8,072)
4H on 15m · Early CISD · 2.5× leg · Time−0.090 (16,066)−0.139 (6,600)−0.101 (5,848)
4H on 15m · Early CISD · 2.5× leg · Fast CISD−0.107 (19,169)−0.129 (8,120)−0.114 (7,055)
4H on 15m · Early CISD · 2.5× leg · Daily bias−0.123 (12,657)−0.151 (5,259)−0.119 (4,583)
4H on 15m · Early CISD · 4× leg · no filter−0.100 (20,540)−0.118 (8,610)−0.088 (7,495)
4H on 15m · Early CISD · 4× leg · Time−0.087 (15,299)−0.134 (6,317)−0.081 (5,570)
4H on 15m · Early CISD · 4× leg · Fast CISD−0.101 (18,127)−0.126 (7,745)−0.099 (6,697)
4H on 15m · Early CISD · 4× leg · Daily bias−0.118 (12,802)−0.154 (5,375)−0.104 (4,657)
4H on 15m · Confirmed · C1 target · no filter−0.091 (12,465)−0.110 (4,701)−0.107 (4,279)
4H on 15m · Confirmed · C1 target · Time−0.075 (7,570)−0.122 (2,782)−0.096 (2,552)
4H on 15m · Confirmed · C1 target · Fast CISD−0.093 (9,004)−0.110 (3,469)−0.119 (3,207)
4H on 15m · Confirmed · C1 target · Daily bias−0.090 (6,689)−0.126 (2,489)−0.086 (2,277)
4H on 15m · Confirmed · 2.5× leg · no filter−0.078 (12,528)−0.099 (4,782)−0.080 (4,149)
4H on 15m · Confirmed · 2.5× leg · Time−0.074 (7,915)−0.112 (2,935)−0.080 (2,575)
4H on 15m · Confirmed · 2.5× leg · Fast CISD−0.080 (8,683)−0.085 (3,358)−0.093 (2,920)
4H on 15m · Confirmed · 2.5× leg · Daily bias−0.076 (6,930)−0.123 (2,657)−0.058 (2,285)
4H on 15m · Confirmed · 4× leg · no filter−0.078 (15,025)−0.103 (5,847)−0.074 (5,152)
4H on 15m · Confirmed · 4× leg · Time−0.074 (9,894)−0.111 (3,736)−0.065 (3,350)
4H on 15m · Confirmed · 4× leg · Fast CISD−0.081 (11,174)−0.084 (4,407)−0.083 (3,909)
4H on 15m · Confirmed · 4× leg · Daily bias−0.070 (8,628)−0.120 (3,351)−0.057 (2,920)
4H on 15m · Retest · C1 target · no filter−0.086 (9,193)−0.139 (3,299)−0.101 (2,997)
4H on 15m · Retest · C1 target · Time−0.053 (5,526)−0.150 (1,944)−0.075 (1,793)
4H on 15m · Retest · C1 target · Fast CISD−0.092 (6,257)−0.137 (2,273)−0.116 (2,118)
4H on 15m · Retest · C1 target · Daily bias−0.064 (4,922)−0.147 (1,731)−0.106 (1,569)
4H on 15m · Retest · 2.5× leg · no filter−0.072 (9,884)−0.115 (3,684)−0.077 (3,208)
4H on 15m · Retest · 2.5× leg · Time−0.049 (6,279)−0.126 (2,270)−0.069 (2,004)
4H on 15m · Retest · 2.5× leg · Fast CISD−0.072 (6,583)−0.094 (2,464)−0.101 (2,183)
4H on 15m · Retest · 2.5× leg · Daily bias−0.070 (5,450)−0.137 (2,019)−0.067 (1,752)
4H on 15m · Retest · 4× leg · no filter−0.081 (11,488)−0.134 (4,312)−0.063 (3,786)
4H on 15m · Retest · 4× leg · Time−0.064 (7,489)−0.161 (2,710)−0.050 (2,428)
4H on 15m · Retest · 4× leg · Fast CISD−0.087 (8,127)−0.114 (3,089)−0.073 (2,744)
4H on 15m · Retest · 4× leg · Daily bias−0.071 (6,526)−0.143 (2,431)−0.073 (2,115)
1H on 5m · Early CISD · C1 target · no filter−0.207 (66,357)−0.197 (27,075)−0.142 (24,024)
1H on 5m · Early CISD · C1 target · Time−0.202 (48,310)−0.211 (19,519)−0.142 (17,333)
1H on 5m · Early CISD · C1 target · Fast CISD−0.210 (57,999)−0.197 (24,055)−0.143 (21,409)
1H on 5m · Early CISD · C1 target · Daily bias−0.219 (37,374)−0.209 (15,110)−0.163 (13,466)
1H on 5m · Early CISD · 2.5× leg · no filter−0.192 (69,471)−0.195 (28,385)−0.158 (24,943)
1H on 5m · Early CISD · 2.5× leg · Time−0.184 (51,490)−0.194 (20,840)−0.155 (18,278)
1H on 5m · Early CISD · 2.5× leg · Fast CISD−0.194 (61,336)−0.197 (25,390)−0.155 (22,419)
1H on 5m · Early CISD · 2.5× leg · Daily bias−0.202 (40,026)−0.201 (16,225)−0.164 (14,320)
1H on 5m · Early CISD · 4× leg · no filter−0.194 (65,009)−0.189 (26,533)−0.145 (23,323)
1H on 5m · Early CISD · 4× leg · Time−0.188 (49,161)−0.196 (19,963)−0.143 (17,506)
1H on 5m · Early CISD · 4× leg · Fast CISD−0.197 (58,196)−0.191 (24,029)−0.146 (21,243)
1H on 5m · Early CISD · 4× leg · Daily bias−0.198 (40,657)−0.205 (16,565)−0.157 (14,573)
1H on 5m · Confirmed · C1 target · no filter−0.165 (47,880)−0.181 (18,747)−0.133 (16,833)
1H on 5m · Confirmed · C1 target · Time−0.161 (29,285)−0.178 (11,196)−0.130 (10,097)
1H on 5m · Confirmed · C1 target · Fast CISD−0.169 (35,704)−0.191 (14,306)−0.136 (12,856)
1H on 5m · Confirmed · C1 target · Daily bias−0.181 (26,522)−0.198 (10,475)−0.158 (9,298)
1H on 5m · Confirmed · 2.5× leg · no filter−0.150 (49,841)−0.173 (19,505)−0.142 (17,271)
1H on 5m · Confirmed · 2.5× leg · Time−0.145 (32,325)−0.168 (12,321)−0.142 (11,074)
1H on 5m · Confirmed · 2.5× leg · Fast CISD−0.156 (35,646)−0.183 (14,241)−0.143 (12,671)
1H on 5m · Confirmed · 2.5× leg · Daily bias−0.162 (28,414)−0.183 (11,092)−0.154 (9,862)
1H on 5m · Confirmed · 4× leg · no filter−0.148 (58,565)−0.169 (23,221)−0.128 (20,679)
1H on 5m · Confirmed · 4× leg · Time−0.145 (38,938)−0.164 (15,182)−0.126 (13,661)
1H on 5m · Confirmed · 4× leg · Fast CISD−0.153 (44,446)−0.175 (17,941)−0.132 (16,085)
1H on 5m · Confirmed · 4× leg · Daily bias−0.150 (34,428)−0.183 (13,664)−0.142 (12,106)
1H on 5m · Retest · C1 target · no filter−0.175 (33,753)−0.215 (12,746)−0.154 (11,582)
1H on 5m · Retest · C1 target · Time−0.178 (21,015)−0.210 (7,722)−0.143 (7,078)
1H on 5m · Retest · C1 target · Fast CISD−0.181 (23,523)−0.234 (9,083)−0.161 (8,305)
1H on 5m · Retest · C1 target · Daily bias−0.178 (18,657)−0.218 (7,070)−0.176 (6,398)
1H on 5m · Retest · 2.5× leg · no filter−0.154 (38,925)−0.196 (14,872)−0.155 (13,286)
1H on 5m · Retest · 2.5× leg · Time−0.153 (25,817)−0.187 (9,605)−0.140 (8,749)
1H on 5m · Retest · 2.5× leg · Fast CISD−0.161 (26,614)−0.207 (10,412)−0.154 (9,326)
1H on 5m · Retest · 2.5× leg · Daily bias−0.159 (22,125)−0.197 (8,452)−0.159 (7,598)
1H on 5m · Retest · 4× leg · no filter−0.167 (43,330)−0.205 (16,715)−0.147 (15,057)
1H on 5m · Retest · 4× leg · Time−0.170 (29,050)−0.202 (10,979)−0.137 (9,993)
1H on 5m · Retest · 4× leg · Fast CISD−0.178 (30,961)−0.215 (12,223)−0.147 (11,094)
1H on 5m · Retest · 4× leg · Daily bias−0.161 (25,126)−0.208 (9,746)−0.155 (8,729)
Daily on 1H · Early CISD · C1 target · no filter−0.032 (4,361)−0.015 (1,731)−0.064 (1,572)
Daily on 1H · Early CISD · C1 target · Time−0.038 (2,878)+0.022 (1,110)−0.010 (1,053)
Daily on 1H · Early CISD · C1 target · Fast CISD−0.045 (3,776)−0.033 (1,521)−0.056 (1,371)
Daily on 1H · Early CISD · C1 target · Daily bias−0.070 (2,375)+0.036 (854)−0.090 (840)
Daily on 1H · Early CISD · 2.5× leg · no filter−0.044 (4,477)−0.035 (1,764)−0.058 (1,600)
Daily on 1H · Early CISD · 2.5× leg · Time−0.059 (2,901)−0.006 (1,122)−0.051 (1,061)
Daily on 1H · Early CISD · 2.5× leg · Fast CISD−0.050 (3,882)−0.037 (1,544)−0.045 (1,393)
Daily on 1H · Early CISD · 2.5× leg · Daily bias−0.065 (2,402)−0.013 (874)−0.088 (853)
Daily on 1H · Early CISD · 4× leg · no filter−0.020 (4,159)−0.063 (1,646)−0.046 (1,490)
Daily on 1H · Early CISD · 4× leg · Time−0.035 (2,829)−0.043 (1,100)−0.007 (1,012)
Daily on 1H · Early CISD · 4× leg · Fast CISD−0.023 (3,700)−0.048 (1,470)−0.051 (1,328)
Daily on 1H · Early CISD · 4× leg · Daily bias−0.061 (2,438)−0.037 (896)−0.093 (855)
Daily on 1H · Confirmed · C1 target · no filter−0.037 (2,228)−0.087 (845)+0.008 (752)
Daily on 1H · Confirmed · C1 target · Time−0.039 (1,071)−0.040 (350)+0.103 (342)
Daily on 1H · Confirmed · C1 target · Fast CISD−0.056 (1,723)−0.086 (654)+0.007 (567)
Daily on 1H · Confirmed · C1 target · Daily bias−0.054 (1,191)−0.070 (426)−0.018 (429)
Daily on 1H · Confirmed · 2.5× leg · no filter+0.008 (1,857)−0.090 (718)+0.028 (643)
Daily on 1H · Confirmed · 2.5× leg · Time+0.000 (783)−0.113 (237)+0.114 (270)
Daily on 1H · Confirmed · 2.5× leg · Fast CISD+0.013 (1,321)−0.103 (526)+0.014 (444)
Daily on 1H · Confirmed · 2.5× leg · Daily bias−0.034 (1,011)−0.059 (393)+0.014 (395)
Daily on 1H · Confirmed · 4× leg · no filter−0.004 (2,347)−0.085 (921)−0.009 (788)
Daily on 1H · Confirmed · 4× leg · Time−0.008 (1,121)−0.040 (378)+0.112 (383)
Daily on 1H · Confirmed · 4× leg · Fast CISD−0.004 (1,798)−0.069 (717)−0.025 (591)
Daily on 1H · Confirmed · 4× leg · Daily bias−0.030 (1,339)−0.058 (523)−0.021 (487)
Daily on 1H · Retest · C1 target · no filter−0.020 (1,589)−0.106 (597)−0.014 (543)
Daily on 1H · Retest · C1 target · Time−0.034 (737)−0.080 (222)+0.157 (220)
Daily on 1H · Retest · C1 target · Fast CISD−0.053 (1,175)−0.127 (438)−0.065 (386)
Daily on 1H · Retest · C1 target · Daily bias−0.050 (830)−0.063 (292)−0.097 (314)
Daily on 1H · Retest · 2.5× leg · no filter+0.032 (1,449)−0.084 (566)+0.071 (507)
Daily on 1H · Retest · 2.5× leg · Time+0.030 (609)−0.082 (187)+0.217 (202)
Daily on 1H · Retest · 2.5× leg · Fast CISD+0.020 (990)−0.106 (395)+0.051 (336)
Daily on 1H · Retest · 2.5× leg · Daily bias−0.019 (781)−0.010 (311)+0.045 (312)
Daily on 1H · Retest · 4× leg · no filter+0.023 (1,737)−0.093 (683)+0.010 (597)
Daily on 1H · Retest · 4× leg · Time+0.005 (797)−0.011 (270)+0.186 (269)
Daily on 1H · Retest · 4× leg · Fast CISD+0.010 (1,277)−0.072 (500)−0.036 (429)
Daily on 1H · Retest · 4× leg · Daily bias−0.022 (958)−0.062 (381)−0.052 (367)

In 2018–22, 9 of the 108 were above zero after costs, all on daily candles, the best at +0.032R per trade over 1,449 trades. Every intraday version was red. The three we carried forward then lost in 2023–24, positive in one market of seven each, so the main test stopped there: no validated edge.

2025–26 flatters the daily row, 15 of 36 above zero. But 34 of those 36 lost in 2023–24. We saw the same with the Engulfing Volume Desk: a rule that follows what markets happened to do, not an edge that carries across years.

Gold only: the final test

Many of our readers trade gold, so we registered a gold-only track too. One version passed 2023–24 on gold: daily candles, retest entry, 4× target, the extreme in the first half of C2 (46 trades, +0.104R per trade). We ran it once on January 2025 to September 2026.

CostsTradesAbove zeroNetR per tradeProfit factorDeepest fall
Base5019 (38%)−1.09R−0.0220.968.8R
Stressed, 1.5×5019 (38%)−1.37R−0.0270.958.9R

Random entries with the same stops and targets did better in 58% of 2,000 shuffles. Fail.

What the costs did

Before costs the model is a coin flip. After costs it only goes down, and the shorter the timeframe the faster.

Before costs it goes nowhere. After costs it only goes down.Cumulative R of the textbook version (confirmed entry, C1 target), seven markets pooled, January 2018 to September 2026.
2018–22 · choose2023–24 · check2025–26 · test−10,000R−5,000R0R201820192020202120222023202420252026−318R−13,546R
Table: cumulative R at each year end
ThroughBefore costsAfter costs
Dec 2018+59.9R−1,553R
Dec 2019+169R−3,319R
Dec 2020+49.1R−4,924R
Dec 2021+36.1R−6,558R
Dec 2022−15.7R−7,914R
Dec 2023−83.4R−9,492R
Dec 2024−332R−11,313R
Dec 2025−436R−12,721R
Sep 2026−318R−13,546R

The 5-minute textbook version averaged −0.004R per trade before costs over 83,460 trades. Costs then took about 0.16R from every trade, because a 5-minute stop is small next to a normal spread. It lost in every year from 2018 to 2026.

Against random entries with the same stops, targets and holding time, it was better by under 0.01R per trade (p = 0.01). Real, perhaps, and about twenty times smaller than its cost. The 4-hour and daily versions could not be told apart from random entries at all.

Check it on your own chart

Fractal Model Lab is a TradingView strategy (Pine v6, in Pine Pro Library) with the same 108 settings. Here it is on the 15-minute gold chart, defaults on, with the study's cost in Properties.

TradingView Strategy Tester for Fractal Model Lab v1.0.0 on OANDA:XAUUSD 15-minute, 1 July to 3 October 2026, 0.20 USD per ounce per side commission: total P&L −749.53 USD (−7.50%), max drawdown 863.97 USD (8.54%), 29 of 83 trades profitable, profit factor 0.57, the cumulative P&L line falling from August
TradingView's Strategy Tester, Lab defaults (confirmed entry, C1 target), OANDA:XAUUSD 15-minute over the history TradingView loads on our plan (1 July to 3 October 2026), with 0.20 USD per ounce per side, the study's 0.40 round trip. A 10,000 USD account.

TTrades sells its own Fractal Model indicator by subscription; how it compares with the desk is on its own page. The free scripts that draw the model on TradingView are read from their source.

We checked the Lab against our research engine trade by trade on that chart, eight settings on the 15- and 5-minute timeframes: TradingView took 737 trades, and 736 matched the engine's entry time and price, 730 its exit too. The rest are candles where TradingView guesses whether the stop or the target came first, a Labor Day session that closed early, and one same-candle re-entry TradingView did not take.

Two of our own mistakes

That trade-by-trade check caught two errors in our research. Neither changes the verdict, and both are written into the study's amendments next to the plan as it was registered.

  • When the 4-hour candles start. We assumed TradingView starts gold's and the indices' 4-hour candles at 18:00 New York. It starts them at 17:00. Re-run that way, the 36 4-hour versions are still 0 of 36 above zero in every period.
  • A gap through the stop. On a Sunday open that jumped through both a retest order and its stop, our engine exited at the stop price, a small gain that could not happen. Fixed, 1,502 retest trades moved by at most 0.006R per trade per version, and the gold final test went from −1.09R to −1.14R.

What you can take from it

  • The mechanical skeleton carried no edge in 8¾ years of seven markets. That is not a verdict on TTrades, or on a trader who adds the parts that cannot be written down.
  • Lower timeframes make it worse. On the 1-hour/5-minute pairing, costs are a large share of every stop. Check yours.
  • It is still a clear way to read a chart: which candle swept which, where the run began, whether the close came back. That is what the desk shows, with no buy or sell signal attached.

We are now following three daily versions forward, on data that did not exist when we registered them (fills from 5 October 2026). Monthly numbers will land here, with no verdict until October 2027.

The deep version, as a PDF

The Fractal Model Field Guide

TTrades' Fractal Model, CRT and CISD as exact rules, with every result of our 108-version, 7-market test.

  • The model in TTrades' own dated words, then as a seven-check rule you can apply candle by candle
  • Every result of our 108-version test on 7 markets, 2018 to 2026, before and after costs
  • What each choice changes: timeframe pair, entry, target and filter, each shown on the chart
See what's inside →

One payment, instant download. Refund within 7 days if you haven't downloaded it.

Limits

One data vendor per market (bid prices; costs stand in for the spread), a fixed cost per market across eight years, no overnight financing, the stop counted first when a candle touches both levels, and our own reading of public TTrades material. Three periods are three samples of history, not a forecast. Nothing here is advice to trade, or not to trade, this model.

Source: ProEA Lab Fractal Model study, protocol fingerprint ad2e6616…0587, engine 33db1554…7c02 (v1 as registered) and 46db28e5…5d66ef (v1.1, amendment A2), registered 2026-10-02 before the first run.

Use this animation on your site.

Frequently asked questions

Is the TTrades Fractal Model profitable?

Not as a mechanical rule. We coded the public model (C2 sweeps C1, the lower timeframe confirms with a CISD, C2 closes back inside C1), fixed the test before the first run, and ran 108 versions on seven markets from January 2018 to September 2026. Before costs it was a coin flip; after costs nothing passed, and the three best versions of 2018–22 lost in 2023–24. That is not a verdict on TTrades, or on a trader who adds the point of interest, bias and objective that rules can't capture.

What is the Fractal Model's win rate?

It depends on the version, and none had an edge. The one gold version that passed 2023–24 (daily candles, retest entry, 4× target) won 19 of 50 trades, 38%, in the final test from January 2025 to September 2026, and still finished below zero after costs. On TradingView's Strategy Tester, Fractal Model Lab's defaults on 15-minute gold won 29 of 83 trades from 1 July to 3 October 2026. A win rate alone says little here, because both targets sit far from the entry (four legs out, or C1's far side), so a win is bigger than a loss.

Does the Fractal Model do better than random entries?

Barely, and by far less than it costs. The 5-minute textbook version was better than random entries with the same stops, targets and holding time by under 0.01R per trade (p = 0.01), about twenty times smaller than its cost. The 4-hour and daily versions could not be told apart from random entries at all, and on gold's final test random entries did better in 58% of 2,000 shuffles.

Which timeframes did the backtest use, and can I run it myself?

Three pairings: 1-hour candles on the 5-minute chart, 4-hour on 15-minute and daily on 1-hour, each with three entries, three targets and four filters. Lower timeframes did worse, because costs are a larger share of a small stop: on the 5-minute version they took about 0.16R from every trade. Fractal Model Lab, a TradingView strategy with the same 108 settings, is in Pine Pro Library with its full Pine v6 source, beside the Fractal Model Desk indicator.

How was the Fractal Model backtest run?

On about 4.7 million 5-minute bars of XAUUSD, EURUSD, GBPUSD, USDJPY, Nasdaq 100 and S&P 500 CFDs (Dukascopy) and BTCUSDT (Binance). The test was written down and fingerprinted before the first run: versions were chosen on 2018–22, checked on 2023–24 and tested once on January 2025 to September 2026, with a cost on every round trip, fills at the next open, and the stop counted first when a candle touches both levels.

Continue reading

We Built a Liquidity Sweep Engine. Then We Measured Both Confirmations.

A thread said finding the liquidity is easy and the entries and exits are where the gold is. So we built the liquidity sweep sequence with a real stop and both targets, ran it with costs on gold and bitcoin, and published every number, including the ones that lose.

Sep 4 · ProEA Lab

Ink sketch on ivory paper: a resting pool of equal highs, a candle wicking through it and closing back, the gap below it inverted into an orange zone, a stop above the sweep and two targets below; the statement reads: the pools were easy, the exits were the work.

Gold Sweep Desk: What We Measured Before Release.

Two pre-registered studies of a gold liquidity-sweep rule on 2026 data: every candidate, every period we ran, and why no setting became a tested default.

Oct 1 · ProEA Lab

Ink sketch on ivory paper: price falls into the Asia low, a candle wicks through it and closes back inside, a later close breaks up through the orange opposing gap, with the stop beyond the sweep's extreme and a 2R target above; the statement reads: two studies, no setting we could call tested.

ICT OTE: Optimal Trade Entry and Fibonacci Retracement, Animated

ICT's optimal trade entry is the 62% to 79% Fibonacci retracement of the leg that broke structure, with 70.5% inside it. The levels are the easy part; the leg is the hard part. One written rule, four replays: Fibonacci on the leg, levels that move until the high is confirmed, OTE against the 50% line, and what ends the leg after the tap.

Oct 4 · ProEA Lab

Explainer cover titled OTE = the 62–79% pullback: simulated candles with a Fibonacci retracement drawn on a bullish leg from the swing low to the high. Grey levels at 0.236, 0.382 and 0.5, ICT's orange OTE band from 0.62 to 0.79 with 0.705 inside, and a ring on the first pullback wick that reaches 0.62.