PINE DIGGA
— The ledger

Every backtest we have, including the ones that failed.

Fourteen configurations, each measured twice: on the twelve months it was tuned on, and on years that existed before the tuning and could not have influenced it. Both numbers sit in the same row.

3 / 14

Three reproduced outside the window that selected them.Six landed within a few points of break-even. Five lost — and every one of those five is a 5-minute lane. A table that showed only the survivors would be the ordinary kind of table, and it would tell you nothing you could check.

How to read a row

The left half is the tuning window. That is how results are normally published and the reason such tables flatter: a setting chosen because it worked on a stretch of data will, by construction, work on that stretch. The right half is the same configuration on earlier data. Nothing re-fitted, nothing re-chosen.

The convention

Fixed sizing — every position risks 1% of the same $100,000, never of the account as it stands. Compounding would flatter these by roughly 20% and make two settings look far apart when the only real difference was the order the trades arrived in. Gold is OANDA, MGC is COMEX, crypto is Bybit throughout.

Winners — reproduced outside their window

profit factor above 1.05 out of sample
SleeveTFNetPFWinMax DDTradesOut of sampleNetPFTrades
Gold Scalps15m+29,7002.2244%4,10043Jan 2022 – May 2025+47,0901.371183
Split again inside its own out-of-sample window: +11,729 (PF 1.154) then +34,062 (PF 1.543) — it strengthens. Re-run with the original unscaled preset it gives +47,080 against +47,090, so it owes nothing to the price rescaling applied to the other gold rows.
ETH Scalps v115m+41,3001.966,40071Oct 2020 – Jul 2025+62,2291.285310
Halves +52,035 (PF 1.523) then +8,788 (PF 1.072) — positive in both, but decaying. The campaign reported PF 1.96; the honest trajectory is 1.52 → 1.07. Zero adoptions: the inherited engine ships untouched.
BTC Magic5m+6,2121.1725%10,98544Jan 2024 – Jul 2025+20,5451.34870
In-sample corrected 31 Aug 2026: this row first read +11,466 / PF 1.36 / 39, which was the campaign's own 365-day window wearing the label of the longer span every other sleeve is measured over. Over that span the lane is +6,212 / PF 1.167 / 44; over its own 365 days, +10,417 / PF 1.317 / 40. The out-of-sample column is unchanged and was re-measured to the cent. Still the newest and thinnest of the three: one window, and a chosen one. Its fixed-dollar stop was 2.10% of BTC in 2020 and 0.25% in 2025, so the test was restricted in advance to the stretch where that scale is closest to the one it was measured at. Ships behind a toggle, default off, because it earns roughly what it draws down.

Only the first two carry evidence across a long window and a split half. The narrowest honest statement of what has been demonstrated is this: the Scalp lane on the 15-minute chart, on gold and on ETH. Everything else on this page is flat or worse.

Flat — neither reproduced nor collapsed

profit factor 0.90 – 1.05
SleeveTFNetPFWinMax DDTradesOut of sampleNetPFTrades
Gold Regulars5m+21,5002.2443%9,20030Jan 2022 – May 2025+371.00167
ETH Scalps age755m1.40178Oct 2020 – Jul 2025−4,3760.993869
Never shipped as a year-round preset. It is seasonal, and the season is below.
MGC Magic v115m+38,8003.0939%6,60023Jan 2022 – Aug 2025−2,5970.95991
Twenty-three trades in the tuning window is a sample, not a track record — and it produced the highest profit factor on this page.
BTC Scalps v115m+21,1001.388,30081Mar 2020 – Jul 2025−13,0990.953362
BTC Magic v115m+17,1001.486,30049Mar 2020 – Jul 2025−12,8990.939260
Forty-nine trades, and its tuning-window edge rests on three or four winners per half.
Gold Regulars15m+13,1002.6647%2,00015Jan 2022 – May 2025−1,3220.93125
Fifteen trades. The thinnest sample in the catalogue, and it produced the second-highest profit factor.

The two thinnest samples here — 15 and 23 trades — produced two of the highest profit factors on their own year and two of the weakest afterwards. Read the trade count before the profit factor, always.

Losers — and all five are 5-minute lanes

profit factor below 0.90
SleeveTFNetPFWinMax DDTradesOut of sampleNetPFTrades
MGC Scalp v15m+56,6001.8739%6,900105May 2022 – Aug 2024−12,9320.844115
BTC Scalps v25m+40,6001.4713,800128Mar 2020 – Jul 2025−99,4820.828733
Its tuning-window result is also sequence-fragile: blocking a single net entry re-orders the rest of the year and moves the total by 30,232. The 138-against-137 trade count hides two very different lists.
Gold Magic5m+43,9001.8146%6,90093Jan 2022 – May 2025−41,4820.809311
ETH Magic5m+15,2231.4331.3%17,21248Mar 2021 – Jul 2025−37,7590.811223
Both out-of-sample halves lose — −36,284 then −1,475 — and the whole window drew down 66,662, two-thirds of the account. Its tuning year also drew down more than it made: 17,212 against 15,223, a reward-to-risk of 0.88:1, which said the same thing on 48 trades that 223 later confirmed.
Gold Scalps5m+25,1002.7650%4,10028Jan 2022 – May 2025−53,7910.534135
Hot-regime only, so its tuning-window curve covers April–August rather than the full year. The gold rows were re-run with every dollar threshold rescaled by the 0.614 price ratio between the two windows; that roughly doubled the trade count and lifted this row from PF 0.255 to 0.534. The verdict did not change.
The pattern

Not one 15-minute lane on this page falls below PF 0.931out of sample. Five of the eight 5-minute lanes fall below 0.90, and the worst reaches 0.534. Median out-of-sample profit factor: 0.956 on 15m, 0.836 on 5m.

That is not proof the 15-minute chart works — four of its six lanes are still slightly negative. It is evidence about where the damage is, and it is why the shipped catalogue is being reconsidered rather than defended.

— Six months at a time

The three that held, half-year by half-year.

A profit factor over four years says nothing about what those four years felt like. The obvious question about a +47,090 is how many of the halves produced it, so here is every one.

Gold 15m Scalps

five of seven positive
Half-yearNetPFWinMax DDTrades
Jan – Jun 2022+8,1171.56434.8%7,61023
Jul – Dec 2022−460.96625%10,55728
Jan – Jun 2023−4,4970.79721.4%16,46228
Jul – Dec 2023+6,8571.35932.1%5,10028
Jan – Jun 2024+13,4291.72037.9%6,97329
Jul – Dec 2024+11,6591.67737%5,58027
Jan – May 2025+2,6441.17628.6%6,03021

ETH 15m Scalps

eight of ten positive, three of them carrying 84%
Half-yearNetPFWinMax DDTrades
Oct – Dec 2020+5,4202.14442.9%2,2067
Jan – Jun 2021+12,9971.68136.7%9,36430
Jul – Dec 2021+1,0021.05127.3%14,57633
Jan – Jun 2022+25,4002.11038.2%4,82234
Jul – Dec 2022+4,9481.24628.6%9,96728
Jan – Jun 2023−1,0420.95425%7,71832
Jul – Dec 2023+7941.04226.9%10,08626
Jan – Jun 2024−3,9540.87322.5%12,47140
Jul – Dec 2024+11,7931.43432.5%4,70840
Jan – Jul 2025+2,4261.08927%7,98737

BTC 5m Magic

two of three positive
Half-yearNetPFWinMax DDTrades
Jan – Jun 2024+9,6692.03735.7%4,65414
Jul – Dec 2024−10,1390.63210.3%10,98629
Jan – Jul 2025+22,0432.04419.2%8,58626
The money is not spread evenly

ETH's three best halves carry 84% of its total, and January–June 2022 alone carries 42%. The other seven halves together make about 9,600 across four and a half years. Gold is less extreme and the same shape.

Drawdown does not follow profit

ETH's July–December 2021 made +1,002 and went through −14,576 on the way. Gold's worst drawdown lands in a half that lost money. BTC 5m Magic's one losing half is also the one with the most trades.

Do not total these columns

A trade that opens in one half and closes in the next is counted in neither — the tester reports trades that close inside the window. Those are by construction the longest-held trades, and in a lane that lets winners run they are disproportionately the winners. Gold's seven halves sum to +38,163 on 184 trades where the same span measured continuously gives +45,750 on 188: four trades exist only in the continuous run and are worth about 1,900 each against a 243 average. Block sums understate the period, systematically and in a known direction.

— The index book

Four years, in R, and no holdout column.

The INDEX campaign was measured differently from everything above: consecutive trading years on the owner's own tester, at 1% risk with commission in, in R rather than in account currency. There is no out-of-sample column here because there is no out-of-sample window — and inventing one would be the exact failure this page exists to prevent.

R by trading yearAll three years together
Cell23-2424-2525-26TotalPFMax DDTrades
Nasdaq futures · 15m
3 of 3 positive

The product. Two flat years and one that is the year — and the shipped figure is below the one first published, because the old run placed exits a bar after the fill and gave every trade a free first bar.

10.21.045.4+56.7 R1.21-16.0 R453
Nasdaq futures · 15m, both tapes agreeing
4 of 4 positive

Measured over four years, including the one dropped above, so it is not comparable line for line with the rows around it. The cross-feed rule buys drawdown rather than profit — the same money for a little over half the ride — and it is the only filter this campaign tried that survived: nine of nine feed combinations improved. It loses on gold, so it is refused there.

+61.4 R1.27-13.0 R499
Nasdaq CFD · 15m
2 of 3 positive

The noisier tape, and the numbers say so. A −24.3 R drawdown at 1 % risk is a 24 % account drawdown; this cell wants a fraction of that sizing.

-6.53.542.1+39.2 R1.13-24.3 R472
Spot gold CFD · 15m
2 of 3 positive

No aggregate is quoted because none can be verified: the trade list reproduces this cell's published trade counts year by year but lands about 4 R below its published four-year total, so its profit factor and drawdown over this window are unknown rather than estimated. Measured at a 0.30 spread, and it does not survive double that. The smallest cell in the bundle, and it should be the smallest position.

34.6-0.518.8
Nasdaq futures · 5m
2 of 3 positive

The companion, not the product — on this tape. On the CFD the ranking inverts and the 5m is the stronger of the two, which is why neither table should ever be quoted without its feed.

1.6-2.633.8+32.9 R1.21-15.0 R261
Nine ordinary years and three that pay

Nine measured years sit in the three shipped cells. Take out +45.4, +42.1 and +34.6 and what is left is a scratch. That is not a flaw in this edge, it is the shape of it — and it is the whole instruction for using it: size for the six, not the three.

The one filter that survived

Dozens were tried and refuted. The survivor does not look at price at all: it asks whether the same setup is standing on a second tape. Nine of nine feed combinations improved — and it loses on gold, so the script refuses to apply it there.

The window was chosen, and choosing it flatters this table

These three years start at September 2023. The 2022-23 year exists, it was measured, and it is dropped here by decision, not by absence of data — so you are reading a window somebody picked. It was the weakest year in every cell, and removing it improves every ratio on this row: the futures cell goes fromPF 1.19 to 1.21 and its worst drawdown from−24.0 R to −16.0 R, the CFD from 1.11 to 1.13 and −36.9 to −24.3, and the gold cell loses its only heavy losing year. Add roughly+5.5, +3.7 and −12.0 R back if you want the four-year picture.

Do not total the year columns

Same reason as the half-years above: a trade that opens in one year and closes in the next is counted in neither. The Total column is the measured three-year figure, not the sum of the three beside it, and the two do not agree. Where the window was never measured as a whole, the cell is left blank rather than added up here. The same applies to everything under All three years together:max drawdown is the single worst peak-to-trough of the whole run, which may sit inside one year or straddle two. It is not a yearly figure and there is no arithmetic that turns it into one.

— A season, not a system

One configuration that only works half the time.

Published because it explains why this sleeve ships behind a switch instead of as a preset. Same settings, same instrument, split where the behaviour changed.

PeriodNetWinMax DDTrades
Jul 2025 – Feb 2026+62,05138%9,501110
Mar 2026 – Aug 2026−13,29321%21,27268

A yearly average would have hidden both halves inside one mediocre number. We looked for a computable rule that says which season you are in — across every regime feature we have — and did not find one. So the switch is the operator's judgement, stated as such, rather than a filter pretending to know.

— The limits

What these numbers cannot tell you.

  • Blank cells are missing records, not zeros. Where a campaign logged net, profit factor and drawdown but not the win rate, the cell reads “—”.
  • Out-of-sample windows differ in length between instruments, which is why those results are in account currency rather than percent. A −99,482 over five years is not comparable to a −1,322 over three.
  • Absolute-dollar thresholds do not travel. Several configurations carry stops pinned in dollars, and a dollar is a different fraction of price in 2021 than in 2026. Where that mattered the window was restricted in advance and the arithmetic recorded — but it is a real limit on how far back any of this can honestly be read.
  • An indicator prints more signals than its bench took. The bench holds one position at a time; the indicator shows every qualifying setup. Your chart is a superset of the measured trade list, deliberately.
  • Forward validation is not finished. Everything here is a backtest until a forward period says otherwise.

Backtest results — past performance is not indicative of future results. Nothing here is financial advice. Figures are per-lane on the stated account convention and are not cumulative across lanes.