Everything in this series so far ran on 5-minute bars. Every certification, every graveyard verdict, every chart — 5-minute bars, because that's the data we owned. This final chapter is about the day we bought finer data and re-asked every question we thought was settled. One strategy died. One got a retrial and lost anyway. And the champion came out with a better certificate than it went in with.
A reader-inspired idea — trade reactions at structural price levels (hourly color-flip levels, the prior day's high and low, the opening print), morning window only, two trades a day, tight 20-point stop, 2R target. We pre-registered the whole family before running anything, then put it through the full gauntlet from chapter one:
| gate | result |
|---|---|
| out-of-sample CPCV (5 paths) | mean 1.37 — double our champion |
| probability of backtest overfitting | 0% |
| deflated Sharpe | 86% |
| +1 tick/side slippage | 1.06 — still strong |
| recent window, 2024→now | 2.22 — stronger lately |
| stricter fill assumption | survives |
Six gates, six passes. By every rule this series has taught, that's a green light. We started planning its paper-trading slot. Then we spent $2.59.
The strategy's stop was 20 points and its target 40. An MNQ 5-minute bar routinely spans 20–60 points. Which means the entire trade — entry, stop, and target — frequently lived inside a single bar. And a bar is just four numbers. It does not tell you the order in which price visited them.
Here is a real trade, September 26, 2024, drawn from the actual data:
We bought the same two years at 1-minute resolution and re-ran the exact same strategy, same window, same configuration:
The full pre-registered grid at 1-minute: CPCV mean −0.61, overfit probability 73%. Not weakened — gone. And here is the uncomfortable lesson: this was not overfitting, so no overfitting test could catch it. PBO, deflated Sharpe, slippage stress, recency checks — all of them validate the consistency of a measurement. Our measurement was consistently wrong. The only cure for a measurement error is a finer instrument.
New instrument in hand ($9.46 for the full seven years at 1-minute), we re-ran everything — the champion's full 108-configuration certification, the funded-sizing certification, and every graveyard verdict whose shape allowed it. The bias, it turns out, cuts both ways: bar-scale targets get flattered (that's what killed the level strategy), while bar-scale stops get punished. One old verdict — a stop at the opening-range midpoint — flipped from "loses money" to "makes a little," and still lost its case: a third of the champion's risk-adjusted return. Every other verdict held. Not one adoption changed.
And the champion? It went in certified at 85% deflated-Sharpe confidence and came out at 92%:
| 5-min record | 1-min record | verdict | |
|---|---|---|---|
| champion certification | 0.71 · PBO 0% · DSR 85% | 0.66 · PBO 0% · DSR 92% | certified |
| funded sizing (432 configs) | 0.92 · DSR 95% | 0.83 · DSR 98% | certified |
| slippage decay | ~0.05/tick | ~0.05/tick | reproduces |
| level-reaction strategy | 1.37 · PBO 0% | −0.61 · PBO 73% | artifact — rejected |
| midline stop | −0.22 | +0.25 | real, but beaten — rejected |
| fixed-tick stops | −6 to −16 | −2.6 to −9.3 | still dead |
| afternoon session | 0.27 best | 0.22 best | still dead |
The question everyone asks, answered with the 1-minute numbers and no varnish. One micro contract, the champion, seven-plus years:
Per micro, that's modest by design — this edge is certified, not spectacular. The economics come from three multipliers, each covered earlier in the series:
| lever | the honest number |
|---|---|
| a $50K eval, one micro | 42% pass per attempt · ~2.4 attempts expected · median 133 days |
| funded account, vol-scaled sizing | certified (DSR 98%): risk-matched 1–5 micros on cushion — scales the grind without scaling the bust odds |
| five funded accounts, copy-traded | mean $4–8K/yr, median near $0 — same signal five times buys speed, not certainty (chapter five) |
If those numbers look small next to what strategy vendors advertise, re-read the vendor autopsy. Small, certified, and real compounds. Large, imaginary, and leveraged doesn't.
Eight chapters, ~1,900 sessions of real data, three markets, four strategy families, eighteen recorded negatives, two certified positives, two adversarial audits that found real bugs in our own favor, one false champion executed the day before promotion — total data bill for the final chapter's decisive evidence: $12.06.
The champion trades on. The paper campaign is next. And when live results start diverging from these backtests — they will, everyone's do — we'll publish that, too.