Pattern Similarity Search vs Traditional Technical Indicators: What's the Difference?

Two Approaches to Reading Charts

If you use Finviz, TradingView, or any traditional charting platform, you're familiar with technical indicators: RSI, MACD, Bollinger Bands, moving average crossovers, and named patterns like head-and-shoulders or double bottoms. These tools have been the foundation of technical analysis for decades.

Chart Library takes a fundamentally different approach. Instead of applying predefined rules to detect known patterns, it converts the entire price action into a mathematical vector and searches for the most similar historical charts — regardless of whether the pattern has a name. These are two very different philosophies, and understanding the distinction helps you use each one more effectively.

How Traditional Indicators Work: Rule-Based Detection

Traditional technical indicators work by applying mathematical formulas to price and volume data. RSI computes the ratio of average gains to average losses over 14 periods. MACD calculates the difference between two exponential moving averages. A 'golden cross' fires when the 50-day MA crosses above the 200-day MA.

Named pattern detectors (like those in Finviz or TradingView's built-in scanner) use geometric rules: a head-and-shoulders requires three peaks where the middle peak is the highest, with neckline support connecting the two troughs. A bull flag requires a sharp move up followed by a consolidation that retraces less than 50%.

The key characteristic of this approach is that humans define the rules. An expert decides what constitutes a bull flag, writes the detection logic, and the scanner finds instances that match those rules. The system can only find patterns it's been told to look for.

How Similarity Search Works: Vector-Based Matching

Chart Library's approach skips the rule-writing step entirely. Instead, every historical trading day is converted into a 384-dimensional embedding vector that encodes the full shape of the price action — including nuances that no rule set would capture.

When you search for a pattern, the system doesn't ask 'is this a bull flag?' It asks 'which historical charts look most like this one?' using L2 (Euclidean) distance in the embedding space. The search runs across 24 million pre-computed embeddings spanning 10 years and 19,000+ symbols, and returns the 10 closest matches in about 9 milliseconds.

The critical difference: similarity search doesn't need pattern names. It finds matches based on the actual shape of the price action, including complex, multi-phase patterns that don't fit neatly into any textbook category. Many of the most interesting matches are patterns that have no name at all — they're just shapes that recur across different stocks and time periods.

Pros and Cons of Each Approach

Traditional indicators have clear advantages. They're interpretable — you know exactly why RSI says 'overbought' (it computed a ratio above 70). They're fast to compute. They produce simple, actionable signals (buy when MACD crosses zero). And they've been studied extensively, with decades of backtesting literature.

The downsides are equally clear. Rule-based detectors can only find patterns someone has defined. They reduce rich price action to a handful of numbers, losing nuance. And different implementations of the same pattern (what exactly counts as a 'head-and-shoulders'?) produce different results, making it hard to validate claims about pattern performance.

Similarity search captures the full shape of price action, including subtle features that no rule set would encode. It finds patterns that don't have names. And because it returns actual historical matches with real outcomes, it answers the question traders actually care about: 'what happened next?' The downsides: it's less interpretable (you see that two charts are similar, but not always why), it requires a large database of pre-computed embeddings, and the results require more judgment to act on than a simple buy/sell signal.

Why Similarity Search Finds What Indicators Miss

Consider a chart that shows a slow, grinding rally with three small pullbacks, each slightly shallower than the last, on steadily declining volume — and then a sudden gap up on high volume followed by two days of tight consolidation. This pattern doesn't have a standard name. It's not quite a bull flag. It's not an ascending triangle. A traditional scanner would either miss it entirely or force it into an ill-fitting category.

Similarity search doesn't care about categories. It converts the full 30-day shape into a vector, searches the database, and returns 10 historical instances where the same shape appeared. Maybe 7 of those 10 continued higher over the next week. Maybe 3 pulled back to fill the gap. Either way, you now have data-driven context that no indicator would have given you.

This is particularly valuable for modern markets, where algorithmic trading has made price action more complex than the textbook patterns devised in the 1990s. The patterns that actually recur in 2026 are often messy, multi-phase, and unlabeled — exactly the kind that similarity search was built for.

When to Use Which

The two approaches are complementary, not competing. Here's a practical framework for when to use each.

Use traditional indicators for screening and filtering. RSI, volume, and moving averages are excellent for narrowing a universe of 10,000 stocks down to 50 candidates. They answer 'what is happening right now?' — overbought, trending, consolidating.

Use similarity search for context and outcome data. Once you've identified an interesting setup, search for historically similar charts to see what happened next. Similarity search answers 'what happened after charts that looked like this?' — which is the question that matters for trade planning.

• Screening 10,000 stocks for setups: Traditional indicators (RSI, volume filters, MA crossovers)
• Understanding a specific chart's historical context: Similarity search
• Setting stop losses and profit targets: Similarity search (use the fan chart and forward return data)
• Quick intraday decision-making: Traditional indicators (faster to interpret)
• Research and trade journaling: Similarity search (richer data, actual historical precedent)

The 'What Happened Next' Advantage

The single biggest advantage of similarity search over traditional indicators is the forward return data. When RSI says 'overbought,' it tells you a math formula crossed a threshold. It doesn't tell you what happened after the last 1,000 times that same condition appeared.

When Chart Library finds 10 similar charts, each match comes with real forward returns at 1, 3, 5, and 10-day horizons. You see the average return, the win rate, the best case, the worst case, and the full distribution. The fan chart shows all 10 trajectories overlaid so you can see the range of outcomes at a glance.

This doesn't make pattern similarity a crystal ball — nothing is. But it gives you something traditional indicators don't: a data-driven answer to the question 'if I take this trade, what range of outcomes should I expect based on history?' That's the foundation of good risk management.