We Got the Time Zone Wrong: Correcting Our Audit of the Viral 'Institutional Protocol' ORB Validation

Correction — 31 July 2026

The original version of this article was wrong, and the error was ours. We read the report’s parameters as New York time; they are Chicago time. That put our opening range an hour before the one the strategy actually uses. The article below has been rewritten with the corrected results. Our apologies to Fabio Valentini and to Matteo Conti.

On 20 July we published an audit of the “Institutional Protocol” — Matteo Conti’s statistical validation of Fabio Valentini’s opening-range-breakout model, the one that went round X with P(EV ≤ 0) = 0.001 and a VALIDATED / DEPLOYABLE stamp on the last page. We re-executed the published EasyLanguage on our own NQ tick store and reported that the claimed $151,013 came out at +$42k, with the win rate collapsing from 57.7% to 52.6%.

That result was an artefact of our own mistake. Here is what happened, what the numbers actually are, and the one habit that would have caught it on day one.

The mistake

The report’s input panel reads:

ORB_Start_H_NY   8      ORB_Start_M_NY   30
ORB_Dur_Min      30     Trade_End_H_NY   14

The variable names say _NY. We took them at their word and built the opening range from 08:30 to 09:00 New York time — an hour before the equity cash open, in the thin pre-market tape.

Those values are Chicago time. 8:30 is 08:30 CT, which is 09:30 ET: the cash open, the single most active thirty minutes of the NQ session. Trade_End 14 is 15:00 ET.

The report told us so, and we wrote it off. Its prose says positions are “flattened on the bar that crosses 15:00 ET” while its input panel says 14. We flagged that as an internal contradiction, tested both, and moved on. It was never a contradiction. It was the one-hour offset, sitting in plain sight.

What the corrected numbers look like

Same rules, same costs, same tick store — only the clock label read correctly.

Under the literal New York reading, the strategy is a coin flip: +$18,969 over 957 trades, t = 0.49. Under the Chicago reading it is +$121,088 over 784 trades, t = 2.39. Identical entry logic, identical stop, identical target, identical cost model. One hour.

	Report’s claim	Our corrected run	Our original (wrong) run
Trades	823	784	945
Win rate	57.72%	58.9%	52.6%
Net, 1 contract	$151,013	$121,088	$42,458
Avg per trade	$183.49	$154.45	~$45
Max drawdown	$25,897	$22,898	—
t-stat	—	2.39	0.96

The remaining gap is cost model and fill detail, not structure. Their $4.50 commission plus a one-tick slippage charge is a different assumption from our flat $5.20 per contract, and 823 trades against our 784 means their engine took a handful of entries ours declined. What matters is that the shape matches: same win rate, same drawdown envelope, same order of magnitude.

And the distribution is healthier than the original audit suggested. Our wrong run had 2024 contributing $36k of a $42k total — textbook regime concentration. The corrected run spreads it: every one of the five full years is positive, the weakest (2022) still adds +$6,137, and no single year carries the edge.

The delta filter does nothing

This is the part worth sitting with.

The model’s distinguishing feature is a volume-delta gate: only take the breakout if the signal bar prints at least 200 contracts of net buying. Our original audit measured that filter and reported it as the model’s genuinely good idea — it lifted the average trade from +$3 to +$45.

On the correct session window, it is worth nothing:

Delta threshold	Trades	Net	t
0 (no filter)	819	+$124,131	2.38
200 (the report’s)	784	+$121,088	2.39
500	667	+$65,362	1.42
1000	228	−$20,546	−0.77

The filter costs a little at the report’s own threshold and destroys the edge above it. Its apparent usefulness in our first audit was a symptom of the broken window: on a pre-open range with no real edge, a filter that removes trades removes losses. On a session that actually works, there is nothing to rescue.

That reverses one of the compliments we paid the model. The delta gate isn’t the clever part. The clever part — and it is genuinely clever — is trading the cash-open range at all.

What carries it

We ran the full grid on real ticks: range window × entry mechanic × cutoff, eight combinations, one pass over the tape.

Change	Marginal effect on net P&L
Range 08:30 → 09:30	+$133,606
Entry: 5-min close → resting buy-stop	+$33,253
Cutoff 14:00 → 15:00	+$16,675

With the 08:30 range, all four variants are statistically flat (|t| ≤ 0.36). With the 09:30 range, all four are significant (t = 2.14 to 3.19). The session window isn’t a parameter of this strategy. It is the strategy.

The habit that would have caught this

We had the answer in front of us the whole time: the trade count.

The report says 823 trades. Our first run produced 945 — 15% more, on a spec we believed we had implemented exactly. We noticed, shrugged, and went hunting for the difference in the win rate instead, because the win rate was where the dollars were.

A 15% mismatch in trade count on an identical specification is not a rounding difference. It means you are not running the same strategy. Our corrected run lands on 819 trades against their 823 — that is what agreement looks like.

So, a rule we are adopting for every third-party audit from here: reconcile the trade count before you discuss a single dollar of P&L. If the counts don’t line up, the disagreement isn’t about fills or costs or slippage. It’s about specification, and no amount of statistics downstream will tell you that.

What still stands from the original audit

Two things, both about method rather than this particular model.

The report’s statistical machinery — 1,000 permutations, a 20,000-path Monte Carlo, bootstrap confidence intervals — is competently built and pointed at the wrong object. Every one of those procedures resamples the MultiCharts trade log. Resampling a log measures the sampling error of that log; it cannot detect whether the fills in it are achievable. The report says so itself, to its author’s credit: “no independent re-execution of the strategy on raw market data was performed.”

We performed that step, twice — once badly, once properly. The second time it confirmed the report’s headline rather than refuting it. That is a better outcome for everyone than our first attempt, and it only exists because the specification was published in full. Strategies that ship their source can be checked. Most can’t.

---

Methodology: NQ front-month tick store, 1,556 sessions 2021-01-04 to 2026-04-16. Real trades only (side 2; sides 0/1 are bid/ask quotes). Aggressor via quote rule against the reconstructed BBO — median spread at trade time 1 tick, 0% inverted, 0.01% of 481M prints inside the spread. Opening range and 5-minute signal bars built from the tape; stop-versus-target resolved by tick order, so no same-bar fill assumption. One contract, $5.20 per contract, zero slippage. Delta thresholds 0/200/ 500/1000 and both cutoffs tested; the 2×2×2 grid additionally tests a resting buy-stop entry. All figures per single contract. This is research, not trading advice; our live, disclosed track records are on /algos.