Mark Ramos · Founder · August 15, 2026

Every algorithmic trading platform I’ve ever evaluated does the same thing: they show you the best row.

“71% win rate.” (On the one ticker, on the one timeframe, in the one market regime where they got lucky.)

“$47R return over 62 trades.” (With a filter turned on that they don’t mention until you’re already paying.)

“Backtested 2020-2024.” (While quietly hiding the 2022 crash months where the system got wrecked.)

I’ve spent 20+ years in enterprise IT before I started building Trding.ai. In that world, if a system passes tests only under one set of conditions and you ship it to production anyway, you’re getting fired. The engineering standard is: your test suite has to include the cases that break your assumptions. Otherwise you’re not testing — you’re marketing.

We hold the trading platform to that same standard.

What’s actually on our methodology page

Our published headline is real: 61.3% win rate, 2.75 profit factor, +47.1R over 62 trades on SOXL 1H · Default preset + Zone Proximity Filter · 2024-2026 window.

That’s the good row. It’s what we lead with because it’s genuinely the best performing preset + filter combination in our published test.

But scroll down the same methodology page and you’ll find:

• Conservative preset · marked “too few trades” because it only fired 4 signals in the same window. Four trades is not a sample. Four trades is a coincidence. If we published a win rate on 4 trades, it would be marketing, not data — so we don’t.
• Rows where Aggressive under-performed Default in specific market regimes. When the Aggressive preset takes trades that Default rejects, sometimes it wins big and sometimes it eats losses Default avoided. We show both outcomes.
• A candid “what this does and doesn’t show” block at the top of every performance table. It says exactly what the numbers mean, what they don’t mean, and where the sample sizes get thin enough to stop drawing conclusions.

If a competitor showed you a spreadsheet with only wins, would you trust it? I wouldn’t. So we don’t do that to you.

What our out-of-sample validation actually showed

Last week we ran an out-of-sample test — meaning we deliberately tested the system on two windows outside the data our indicators were tuned on. One window before the published dataset (SOXL daily, 2020 → early 2024). One window after (SOXL 1H, April 2026 → August 2026).

The idea: if the edge is real, it should hold up on data the system was never optimized for. If the edge is curve-fit, out-of-sample tests will expose it fast.

Here’s what we found:

After published data · Apr 2026 → Aug 2026 (SOXL 1H):

Before published data · 2020 → March 2024 (SOXL daily):

Aggressive and Default held their edge in both out-of-sample windows. The Conservative preset kept doing what it does — barely triggering, too small a sample to say anything about.

Notice we didn’t hide the Conservative rows. We didn’t quietly delete them and only show you the two presets that worked. The honest answer is that Conservative doesn’t fire enough trades to draw conclusions from — and we say that directly on the row instead of pretending the data doesn’t exist.

Why this matters (beyond ethics)

There’s a business reason for this transparency that most trading companies miss:

The methodology page’s credibility is the entire product.

If you subscribe to Trding.ai and — three months in — you notice your live results don’t match the headline number, you’ll cancel. Fast. And you’ll be right to.

But if we’ve already told you about the weak rows, the small samples, the caveats — and your live results land in the range we set expectations for — you’ll stay. Because the honesty compounded into trust.

Every trading platform that hides the weak rows is optimizing for the sale.
The ones that show the weak rows are optimizing for the renewal.

Twelve-month customer retention math beats one-month conversion math. Every time.

What this means for you

If you’re evaluating Trding.ai:

1. Read the methodology page in full. Not just the headline number. The “too few trades” rows and the “what this does and doesn’t show” block are the actual signal about whether we’re trustworthy.
2. Run the reproduction commands yourself. Our out-of-sample validation ships with the driver scripts (oos_driver.py, make_charts.py) so you can rerun the tests on any window you want. If the numbers don’t reproduce, they weren’t real.
3. Expect losses. Even our HIGH-grade signals miss about 40% of the time. The math works because winners outsize losers, not because we never lose. Any platform selling you “70% win rate guaranteed” is selling you something they can’t deliver.

If you’re thinking about the Charter Members opening September 26 — the first 30 subscribers lock in today’s pricing forever, and I’d rather you look at the weak rows first and decide honestly. That’s a Founding Member conversation I actually want to have.

More next week.

— Mark

Trding.ai is a software tool, not a registered investment advisor. Nothing on this page is financial advice. Past performance does not guarantee future results. Backtest data does not guarantee live-trade outcomes. Trade at your own risk.