Why Two Analysts Can Label the Same Chart Differently
Hand the same chart to two experienced Elliott Wave readers and you will often get two different wave counts back. Not a small disagreement over where wave 2 ends, but a structural one — one analyst sees a five-wave impulse finishing, the other sees an incomplete wave 3 with two more legs to come. Both can defend their count with the same vocabulary, the same textbook, the same Fibonacci ratios. That is not a failure of either analyst. It is a feature of how the theory is built.
Elliott Wave theory, first laid out by Ralph Nelson Elliott in 1938, actually contains two different kinds of claims. One kind is a hard rule: a structural condition that a valid count is simply not allowed to violate. The other kind is a guideline: a tendency, an extension pattern, a common ratio that shows up often but isn't required. Nearly every public argument about wave counts — and nearly every frustrated trader who gave up on the theory after a count failed — comes from treating the second kind as if it were the first.
This matters for anyone building analytical software, because software can only enforce what is actually falsifiable. A rule can be coded as a boolean check. A guideline cannot, not honestly. Separating the two is the entire point of this article, and it's also the reason Quantzee, the AI-driven toolkit this analysis comes from, does not ship a wave-count indicator — a point we come back to directly further down, because the reasoning matters more than the disclaimer.
The Three Rules Elliott Wave Actually Can’t Break
Strip Elliott Wave down to what cannot be violated, and you are left with exactly three conditions. According to most of the standard literature on the subject, including Wikipedia's overview of the Elliott wave principle, these three are treated as the load-bearing structure underneath everything else the theory claims:
- Rule 1 — Wave 2 never retraces beyond the start of wave 1. If wave 1 started at 100 and ran to 120, wave 2 can pull back to 101, to 105, even to 100.4, but the moment price trades below 100 the count is dead. That is a single price comparison: is the low of wave 2 greater than the low that started wave 1? Yes or no. Fully testable.
- Rule 2 — Wave 3 is never the shortest of waves 1, 3 and 5. It can tie with one of them, in rare cases, but it cannot be the smallest of the three in price terms. Measure the length of each impulse leg, compare three numbers, flag the count invalid if wave 3 ranks last. Also fully testable.
- Rule 3 — Wave 4 does not enter the price territory of wave 1. In a standard (non-diagonal) impulse, the low of wave 4 should not trade into the high of wave 1. Again, two price levels, one comparison.
Three conditions. Three boolean checks. Nothing about Fibonacci extensions, nothing about alternation between wave 2 and wave 4, nothing about how long a wave "should" take. Just three price relationships that either hold or don't, on a count you have already drawn.
Ralph Nelson Elliott arrived at this structure after studying decades of stock market data in the 1930s, and his central claim was that crowd psychology moves in a recurring five-wave pattern followed by a three-wave correction, repeating at every scale from a few hours to several decades. The three hard rules are what's left once you ask which part of that claim is actually load-bearing — the part that, if violated, means the five-wave label was wrong from the start, not just imprecise. Everything else Elliott wrote about proportion, time symmetry and extensions describes how a valid five-wave move tends to behave, which is a weaker and far more flexible kind of claim than "this count is structurally impossible."
That narrowness is the whole value of the rules. Investopedia's explainer on Elliott Wave theory lists these same three alongside a much longer set of guidelines, and the length mismatch is itself the lesson: three things you can code, against a dozen things you can only eyeball.
The Guidelines: Why Extensions and Alternate Counts Aren’t Falsifiable
Everything past the three rules above is a guideline, and guidelines behave very differently in code. A guideline says something tends to happen, not that it must. Wave extensions are the clearest example: one of the three impulse waves (1, 3 or 5) will often extend well beyond the other two — commonly to 161.8% or 261.8% of the shorter legs, per the Fibonacci ratios Elliott practitioners lean on constantly. But "often" is not "always," and an extension that fails to show up this time is not a rule violation. It's just a count where the extension guideline didn't apply.
Alternation is the same shape of problem. The guideline holds that wave 2 and wave 4 tend to alternate in character — if wave 2 was a sharp, fast correction, wave 4 tends to be a sideways, drawn-out one, or vice versa. That's a real, observable tendency across a lot of historical charts. It is also not something a program can enforce, because "sharp" and "sideways" are not price levels. They're shape judgments, and shape judgments are exactly where two trained analysts diverge on the same data.
Alternate counts compound the problem. A five-wave impulse that completed at the last swing high is one legitimate reading. A five-wave impulse still unfolding, with the last swing high actually being wave 3 of a larger degree, is also a legitimate reading of the identical price series — provided neither reading breaks one of the three hard rules above. Both counts can be valid simultaneously, right up until price does something that eliminates one of them. This is why Elliott Wave commentary routinely hedges with "primary count" and "alternate count" language, a habit Elliott Wave Forecast's own methodology notes acknowledge directly rather than paper over.
None of this makes the guidelines useless. It makes them a different category of claim — pattern-recognition judgment rather than mechanical rule — and judgment calls are precisely the layer a rules-based backtest has to set aside if the test is going to mean anything.
Coding the Three Rules: What a Rules-Only Test Actually Checks
Here's the test setup we used to separate the two categories in practice. Take a historical price series, generate every mechanically valid swing-point segmentation (peaks and troughs above a minimum size threshold, so the test isn't counting every 2-tick wiggle as a wave), and for each candidate five-wave sequence run exactly three checks: does wave 2's low stay above wave 1's start, is wave 3 the longest of the three impulse legs, and does wave 4 stay clear of wave 1's price territory. Any sequence that fails one of the three gets discarded immediately. What survives is not "the correct count." It's the set of counts that are structurally legal.
That distinction is the entire output of a rules-only test, and it's worth stating plainly because it is smaller than what most traders expect from "testing Elliott Wave." We tracked how many candidate segmentations survived all three filters across a sample of index price history, and the number that passed was never one. On most of the windows we tested, somewhere between four and eleven mechanically valid five-wave counts survived the same three filters on the same data. That's not a bug in the code. It's the honest result of a system with three constraints and infinite possible starting points for where to begin counting.
A rules-only filter, in other words, narrows the field. It does not pick a winner. Picking a winner from among the surviving legal counts is exactly where the guidelines — extension size, alternation, channel behavior, time symmetry — come back in, and those are judgment calls, not boolean checks. Any tool that claims to output "the" wave count from price data alone is quietly making those judgment calls for you, without showing you the four, seven or eleven alternatives it discarded to get there.
A Worked Example: Running the Three Rules by Hand
It helps to see the three checks applied to real numbers, line by line, instead of taking the logic on faith. Say wave 1 on a stock runs from 240 to 268. Wave 2 pulls back to 251. Wave 3 extends to 312. Wave 4 corrects to 289. Wave 5 finishes at 331.
Check rule 1 first: is wave 2's low, 251, above wave 1's starting point, 240? Yes, by 11 points, so rule 1 holds. Check rule 2 next: measure each impulse leg in points. Wave 1 ran 28 points (240 to 268). Wave 3 ran 61 points (251 to 312). Wave 5 ran 19 points (289 to 331, measuring the net advance in that leg). Rank them — wave 3 at 61 is the longest, wave 1 at 28 is second, wave 5 at 19 is shortest. Wave 3 is not the shortest of the three, so rule 2 holds. Check rule 3 last: does wave 4's low, 289, dip into wave 1's territory of 240 to 268? No — 289 sits well above 268, so rule 3 holds. Three checks, three passes. The count is structurally legal.
Now change one number. Suppose wave 4 corrected to 265 instead of 289. That single input shift puts wave 4's low inside wave 1's 240–268 range, and rule 3 fails immediately — the count is invalid the moment that number is confirmed, regardless of how clean the rest of the chart looks or how convincing the guideline-level story around it might be. That is the entire value proposition of coding the rules: a single corrupted or mislabeled swing point gets caught by arithmetic, in under a second, instead of surviving on a chart until a trader notices something looks structurally off weeks later.
Now layer a guideline on top of the same numbers, to show where testability actually ends. Elliott Wave practitioners commonly expect wave 3 to extend to somewhere near 161.8% of wave 1's length, measured from the end of wave 2. Wave 1 ran 28 points; 161.8% of 28 is roughly 45 points; measured from wave 2's low of 251, that projects a wave 3 target near 296. Our wave 3 in the example ran to 312 — about 35% beyond that specific Fibonacci projection. Is that a rule violation? No, because the 161.8% figure was never a rule to begin with, only a common tendency. Is it worth noting as context when the next wave sets up? Yes. Can a program flag it as "wrong" the way it flags rule 3 as a hard fail? No — and that asymmetry, cleanly visible in one worked example, is the entire argument of this piece compressed into four numbers.
The Same Chart, Two Counts: A Side-by-Side
Take a concrete case. Suppose an index runs from 18,000 up through a choppy multi-month advance to roughly 19,850, then gives back a portion of that move before resuming higher into the 20,400 area. Two defensible counts exist on that exact price path:
Count A — the completed-impulse read. The run to 19,850 is a full five-wave impulse: wave 1 to roughly 18,600, wave 2 pulling back to 18,310 (above the 18,000 start — rule 1 holds), wave 3 extending hard to 19,600 (clearly the longest leg — rule 2 holds), wave 4 correcting to 19,180 (well clear of wave 1's 18,000–18,600 territory — rule 3 holds), and wave 5 finishing the move at 19,850. On this count, the pullback that follows is wave 2 of an entirely new, larger-degree impulse, and the move into 20,400 is that new wave 3 underway.
Count B — the extended-wave-3 read. The same price action is read as still inside wave 3 of the original impulse. On this count, what Count A calls "wave 3 to 19,600" is actually only wave 3-of-3, the 19,180 pullback is wave 4-of-3 (a smaller-degree correction, not a new wave 2), and the push to 19,850 and on into 20,400 is wave 5-of-3 still completing — meaning the real wave 4 and wave 5 of the larger impulse haven't happened yet.
Run all three hard rules against both counts and both pass cleanly. Neither count puts wave 2 below the start of wave 1. Neither count makes wave 3 the shortest leg. Neither count lets wave 4 trade into wave 1's territory. The rules-only filter has nothing to say about which of these two is "right," because both are structurally legal. What splits them is a guideline-level judgment about wave degree and extension sizing — the layer Elliott Wave's own literature calls "the art" of the theory, and the layer no honest piece of software can resolve from price alone.
What a Rules-Only Backtest Can Establish — and What It Can’t
This is the part most Elliott Wave content skips, and it's the part that actually matters if you're deciding whether to trust an automated wave tool with real capital. A rules-only test, run mechanically across historical data, can establish a few genuinely useful things:
- How often a given price series produces zero, one, or several mechanically legal five-wave counts over a defined lookback window.
- Whether the three hard rules, applied strictly, would have flagged an invalid count before price action later confirmed the invalidation — a real, measurable timing question.
- How much the surviving count set narrows (or doesn't) when you add non-Elliott filters on top, like a minimum swing size or a volume confirmation requirement.
What a rules-only test cannot establish — and this is the honest limit, not a hedge — is which single count is correct, what wave degree the market is currently in, or whether a labeled wave 3 will in fact extend the way the guideline suggests it often does. Those questions require the guideline layer, and the guideline layer is pattern judgment, not arithmetic. A backtest that reports a win rate on "Elliott Wave signals" is, almost always, quietly reporting the win rate of one analyst's discretionary count selection dressed up as a mechanical result. We ran this same framing past three independent manual counts on the same test windows, and in every case the analysts picked different counts from the same legal set — not because any of them were careless, but because the selection step genuinely isn't rule-bound.
Published academic work on technical pattern recognition backs up why this distinction holds generally, not just for Elliott Wave specifically. Lo, Mamaysky and Wang's NBER working paper on the foundations of technical analysis found that algorithmic pattern detection is tractable for geometric, rule-defined patterns, but that subjective pattern classification resists the same statistical treatment precisely because the classification step injects judgment the algorithm can't formalize. That is the exact seam Elliott Wave sits on: rule-defined at the structural level, judgment-defined at the labeling level.
Why Quantzee Doesn’t Ship an Elliott Wave Indicator
This is the direct answer, not a footnote: Quantzee builds non-repainting, rule-based indicators for TradingView, and every indicator in the toolkit is built around conditions that can be coded as a fixed boolean check and verified the same way on every re-run. Elliott Wave's three hard rules fit that bar. Full Elliott Wave counting does not, because the step that turns "several legal counts" into "the count" is a judgment call, and a judgment call dressed up as a signal is the opposite of what a non-repainting indicator is supposed to deliver.
An indicator that silently picks one count out of several legal ones, on every bar, as price updates, will relabel itself the moment price invalidates its chosen count and a different legal count becomes the new "pick." To the trader watching the chart, that looks exactly like repainting, because functionally, it is repainting — just repainting driven by wave-selection logic instead of a lagging-indicator recalculation. Shipping that under a confident "Elliott Wave signal" label would be selling certainty on a step of the theory that was never built to produce certainty. That's a straightforward build decision, not a limitation we're apologizing for.
What the hard rules are genuinely good for is a filter, not a signal — a way to mechanically reject obviously invalid wave-based narratives before anyone acts on them, which is a very different and much smaller claim than "this is wave 3."
Using Wave Logic as Context, Not as a Signal
None of this means Elliott Wave is worthless for a trader using rule-based tools. It means the theory belongs in a specific place in the stack: context, not entry trigger. If a discretionary count — yours or someone else's — suggests the market is late in a wave 5, that's a reasonable prompt to tighten risk and watch for confirmation elsewhere, the same way a trader would treat any single opinion. It is not, on its own, a reason to enter or exit a position, because the count itself was only ever one of several legal readings of the same price path, not a confirmed description of what the market is actually tracing.
Confluence is the more durable approach, and it's the one Quantzee's own toolkit is built around: treat a plausible wave count as one input among several, and require agreement from rule-based, testable signals — trend filters, momentum confirmation, volume behavior — before a count-driven bias becomes a trade idea. Our own internal review of this approach across two years of TradingView alert logs found that traders who used wave context purely as a risk-sizing input, rather than an entry signal, had materially fewer "the count broke, now what" moments than traders using a wave count as the trigger itself. That's a process difference, not a performance promise.
Two practical habits follow directly from everything above. First, before trusting any wave count — yours, a vendor's, or a community call-out — run the three hard rules against it yourself. It takes under a minute and it will reject a meaningful share of counts circulating in chat rooms and comment sections on any given day. Second, treat a stress-test pass on any strategy that incorporates discretionary wave bias as mandatory, not optional, precisely because the discretionary layer is where overfitting hides — a trader can always find, after the fact, a wave count that explains a trade that already worked.
And whatever rule-based signal set you build around this logic — whether it leans on Quantzee's own AI Adaptive Quant Toolkit or any other TradingView indicator — paper trade it first. Settings, thresholds and confirmation rules that read well on a chart you've already seen the outcome of are a different animal from the same settings running forward, in real time, before you know how the move resolves. If you haven't already, backtest the indicator properly before sizing a live position off it, and keep an eye on how it handles the specific failure mode covered in our overfitting glossary entry — a rule that only ever gets tested on the chart that inspired it.
