I Audited My Own Stock-Grading Engine and Found 14 Flaws. Here’s How I Rebuilt It.
A transparent look at Signal 2.0, the new grading system behind Quantex Terminal, and the investing lessons that forced the rebuild.
Quantex Terminal was created to help beginner investors research stocks and get transparent recommendations. Not a black box signal, but an informative report that gives you information about how the rating was compiled. So when I finally sat down to audit the engine producing those grades and realized that it was secretly optimized to recommend some of the WORST possible trades, I knew I only had one choice: to ensure my users do not get cut on a falling knife, to rebuild it, and to share with everyone what I learned. **(Read Until The End For an Important Message)**
What the old engine did well
I want to be fair to the thing I’m about to criticize, because it got the most important thing right: it was fully explainable. Every grade broke down into six factors, each worth a set number of points, and you could trace any verdict back to a rule, which did not include a random compilation of data points. That “show your work” type of mentality is the best thing Quantex Terminal has, and I refused to lose that leverage in the rebuild!
The flaw that kept me up at night
The old engine scored six factors and added them up. The problem wasn’t the adding. It was what it rewarded. Four of those six factors were what quants call mean-reversion signals (bets that a price will bounce back). The engine gave points for:
- A stock being near its 52-week low
- For an “oversold” RSI reading
- For trading near the bottom of its recent range.
- And its single biggest factor was how far the price sat below Wall Street analysts’ price targets.
Now read all of those together. A stock that has fallen hard, looks oversold, and trades far below where analysts used to think it should be… gets the engine’s highest-conviction BUY. **(I know, I was confused too.)**
After about 5 days of profound contemplation, I realized its current logic made no sense. In fact, it was telling users, supposedly beginners in investing, to BUY a falling knife (an asset experiencing a rapid and sharp price decline, often dropping 20% or more within a few weeks). Another fact is that decades of factor research point the other way: intermediate-term momentum and proximity to 52-week highs tend to predict positive returns, while stocks sitting at their 52-week lows tend to keep underperforming. My engine had the sign entirely backward on its most confident calls!
For a product whose entire target audience is beginners, the single fastest way to lose their trust is to tell them to buy the thing that they see keeps falling.
There were more issues with the signal (14 findings in my full audit). I fixed them all, but these three issues were the most shareworthy:
- Falling-knife concentration: the engine’s best BUYs clustered in downtrends.
- Nothing was validated: every threshold and weight was hand-set, never tested against real forward returns.
- The stop-loss fought the entry logic: the engine liked buying near the lows, then placed the stop-loss right at the low, so literally any regular daily volatility would trigger it and secure a loss on a position the engine still preferred.
Signal 2.0: the rebuild
Let’s move on to replacing the whole scoring engine! The new system rests on six design principles, and each one is a direct answer to something the old engine got wrong.
- Sector-relative signals, not based on the whole market. The old engine used fixed thresholds. For example, a P/E of 15 scored the same whether it was a bank or a high-growth software company. That’s utter nonsense because those industries live on entirely different curves, using their own metrics for comparison. Signal 2.0 scores every metric against the stock’s sector counterparts. A bank is judged against other banks and software against other software, the same way a real analyst would think.
- Smooth scoring rather than rigid thresholds. The previous system relied on fixed score bands, meaning even a small change in a metric could push a stock into a different rating. Signal 2.0 replaces those hard cutoffs with continuous 0–100 scoring, creating more stable, accurate, and nuanced signals.
- Seven pillars instead of six factors. Signal 2.0 grades a stock on seven things: business quality, valuation versus peers, price trend and momentum, sentiment and analyst revisions, earnings quality (can the reported profits be trusted), risk, and the overall market control. Each pillar is a 0–100 bar you can inspect.
- Momentum is no longer the enemy. This is the MAIN fix. In the new engine, trend and momentum dominate the technical pillar. Being near the 52-week high is bullish. Oscillators like RSI only act as “buy-the-dip” refinements inside an established uptrend, and never as standalone reasons to catch a knife! And analyst price targets, the old engine’s biggest input, are demoted to display-only context. The thing that actually carries weight now is the breadth of revisions (whether analysts are raising or cutting their estimates), which is far more predictive than a stale target.
- Gated by veto rules. Adding up scores can’t capture interactions such as “this stock is cheap because it deserves to be.” So Signal 2.0 layers a small set of veto gates on top of the score. The most important: a falling-knife gate that forbids any BUY on a stock below a falling 200-day trend line, and a value-trap gate that caps a cheap-but-deteriorating stock at HOLD and labels it exactly that. Every gate that fires is shown to the user, with its reason.
- Regime-aware. The engine reads the overall market once a day. In a calm, rising market, it leans toward momentum. In a stressed, falling market, it leans toward quality and safety, and it raises the bar a stock must clear to earn a BUY signal at all because most long positions fail in downtrends regardless of how good the individual company looks.
Confidence, done honestly
The old engine showed a single “confidence” percentage. To be fully transparent, it didn’t measure confidence; it measured signal strength, which is why a great stock could show a confusing single-digit number.
Signal 2.0 kills the single number and shows three honest ones instead:
- Data coverage (how much of the data we wanted was actually available)
- Pillar agreement (do the seven checks point the same way, or does the stock have real strengths and real weaknesses?)
- Signal strength (how far from neutral the verdict sits).
A stock can be a strong BUY with low agreement, and you deserve to see that strain rather than have it averaged away.
Better stops
The old stop-loss could sit within 1–3% of your entry price, so normal daily volatility would stop you out of a position the engine still found to be good. The new stops are measured in units of each stock’s own volatility (its ATR or Average True Range), placed a sensible distance from entry, and never parked or stuck right on a structural level. Additionally, position-size suggestions shrink automatically when a stock is riskier.
The part I care about most
None of this matters if you can’t see it. So the new signal card shows the rating, the market regime, all seven pillar bars, any safety gate that fired and why, the three confidence numbers, the stop and target range, and an expandable “Why this rating?” section listing the top three things helping and hurting the stock.
I also wrote plain-English tooltips for every part of it, because the whole reason Quantex exists is that when I was learning, I needed someone to explain the complex lingo to me. For example, hover over “Risk,” and it says: “How much could go wrong, and how fast?”
Three more things I updated alongside it
A better engine is only useful if people can actually get to use it. So the rebuild came with three changes aimed directly at the beginner’s first five minutes.
- A Top 10 board, ranked three ways. The hardest question for a new investor isn’t “is this stock good?” it’s “what should I even look at?” So Quantex Terminal now re-ranks the entire covered universe using the same seven pillars, but through three different time-frame lenses or “horizons”, because the best stock genuinely depends on your ideal time frame. The short-term list (a few days to ~3 months) leans toward what moves price soon. Things like trend, momentum, and sentiment. The medium-term list ( around 3–12 months) is balanced, letting business quality and price trend both count. The long-term list (1+ years) leans toward business quality and the price you pay. Each name shows its rating, the market regime (a persistent phase in financial markets characterized by specific behaviors in volatility, liquidity, and asset correlations), and the exact pillars driving it in one plain-English line. It’s a huge help when you want to start your own research!
- Search by company name, not just ticker. You no longer have to know that Alphabet trades under two primary ticker symbols: GOOGL (Class A) and GOOG (Class C) (The main difference between the two is that GOOGL shares include voting rights, whereas GOOG shares do not). Type a company name and Quantex shows you the matches to choose from. It’s a tiny thing that removes a real barrier; memorizing ticker symbols is the kind of friction that makes a beginner feel like investing wasn’t built for them.
- ETFs, not just single stocks. You can now pull up an ETF and have it graded on the parts of Signal 2.0 that apply to a fund, such as the trend, the risk, and the market regime, so you can size up a whole basket the same way you’d size up one company. The engine simply uses the pillars that make sense for a fund and leaves the company-specific ones aside, using the same “renormalize over what’s available” principle that runs through the whole rebuild.
What’s next
I can’t claim the job is finished. The most important work is still ahead: validation. Right now, I believe the new engine is a well-reasoned redesign, but I want to get it to “proven.” The next phase is a proper point-in-time backtest, testing the grades against real forward returns, with no cheating on hindsight, before Quantex Terminal makes any performance claims at all. Until that’s done, everything here is research and education, definitely not a track record!
IMPORTANT!! — You, as the user, can also help with this! Quantex Terminal has introduced a feedback survey at the bottom right of the screen. It includes a few questions to help grow this platform to be the best it can and truly help the investing community along the way. It takes less than a minute and every answer is valuable to us!
If you want to see how your FAV stock gets graded now, Quantex Terminal is FREE at quantexterminal.com.
This article is educational and describes a research tool. It is not investment advice.
Comments
Post a Comment