Multi-timeframe scanning is the practice of running the same detection procedure across several chart intervals at once, rather than looking at one interval and forming a view. Automated, it is cheap. The interesting part is not the scanning — it is what to do when the timeframes disagree, and what it costs you to have looked at all of them.
This engine scans five intervals — 15 minutes, 1 hour, 4 hours, 1 day and 1 week — across twenty-five markets. That is 125 chart states, refreshed continuously. Nobody watches that by hand, which is the honest case for automating it. But scanning 125 states also multiplies a specific failure, and most descriptions of multi-timeframe analysis leave that part out.
What a scan actually returns
For one market, a scan reports, per interval: whether a textbook formation is present, its name, its confidence and the component breakdown that produced that confidence, plus the five-lane composite score for that interval.
A typical result is mostly empty. On any given market, most intervals have no textbook formation on them, and the scan says so. That emptiness is the normal case rather than a failure to find content — the detectors return nothing far more often than they return something, by design.
The scan does not rank the intervals, and it does not pick one for you. It reports what each shows. The reason is in the next section.
Timeframes disagree, and the disagreement is the information
A market can carry a falling wedge on the 4-hour and nothing at all on the daily. It can be scored bullish by the composite on the 1-hour and bearish on the 1-week. These are not contradictions to be resolved. They are statements about different questions.
An interval defines what counts as a swing. The pivot threshold is 2.5 × ATR(14) floored at 0.8% of price, computed on that interval's bars — so a move that is a major reversal on the 15-minute chart is not even a pivot on the daily. The 15-minute chart is answering "what has structure done in the last day or two", and the weekly is answering "what has structure done over the last two years". Both answers can be true.
The failure mode this creates is well known and worth naming: scanning until you find the interval that agrees with you. With five intervals available and a scan that surfaces all of them at once, someone who already has a view can nearly always locate a chart that supports it. The tool has made that easier, not harder. A scanner that presents five intervals side by side and then invites you to pick has automated the selection step of a bias.
The defence is procedural rather than technical: decide which interval you are asking about before looking at the scan, and treat the others as context rather than as candidates.
The evidence is thinner on the intervals people trust most
Here the data contradicts a common intuition. Traders generally treat longer intervals as more reliable — fewer false breaks, less noise. That is a reasonable belief about signal quality. It is the opposite of the truth about evidence quantity.
Replaying the detectors over nine years of ten major pairs on the daily and 4-hour intervals produced 661 completed formations. Split by interval:
| Interval | Formations reaching the 30-sample floor |
|---|---|
| Daily | One shape only — symmetrical triangle, at 30 |
| 4-hour | Eight shapes — symmetrical triangle 181, descending triangle 68, falling wedge 50, double top 49, ascending triangle 41, bull pennant 36, double bottom 35, rising wedge 30 |
On the daily interval, falling wedge appears 16 times in nine years. Double bottom 15. Double top 6. Head and shoulders once. Rectangle, rounding bottom, triple top and triple bottom never appeared on the daily at all.
The consequence is direct: every hit rate published in the pooled table is, in practice, a 4-hour statistic. Double top at ten bars measured on the 4-hour alone is 28.6% over 49 samples, against 29.1% pooled — the daily sample is too thin to move the combined figure.
So when someone looks at a daily chart and recalls that "this formation historically resolves such-and-such", the statistic they are recalling was almost certainly counted on a much shorter interval. Nine years is only about 2,900 daily bars, and a strict detector over 2,900 bars leaves a dozen or so formations. The longer the interval, the thinner the evidence behind any claim about it.
This does not mean daily formations are less meaningful. It means nobody — including us — has enough of them to say.
Only two of the five scanned intervals have any historical record
This is the point about this product most worth stating bluntly.
The scan covers 15 minutes, 1 hour, 4 hours, daily and weekly. The statistics counting what actually followed completed formations exist for the daily and 4-hour intervals only. For 15-minute, 1-hour and weekly charts, there is no record of how those formations resolved.
The reasons are practical. Nine years of weekly bars is only about 470 candles, and a strict detector over 470 bars leaves almost nothing. At 15 minutes the bar count is enormous, but it is also the region where fees and slippage dominate the outcome, so a rate computed there would carry little practical meaning.
| Interval | Formations detected | Historical record |
|---|---|---|
| 15 minutes | Yes | None |
| 1 hour | Yes | None |
| 4 hours | Yes | Yes — eight shapes above 30 instances |
| Daily | Yes | Very limited — one shape above 30 |
| Weekly | Yes | None |
So when a double bottom is detected on a 15-minute chart, any "here is what happened historically" attached to it is not from that interval. Presenting the same statistics across all five intervals without saying so would not be honest, which is why this table exists.
Most scanners do not draw this distinction. If a tool shows the same success rate on every interval, it is worth asking which interval that rate was measured on. Attaching a statistic to an interval it was never computed for is technically trivial and thoroughly misleading.
Scanning more states surfaces more noise
This is the cost of coverage, and it is arithmetic rather than opinion.
Forty independent 400-bar random walks run through these detectors produced a nameable formation in fourteen of them. Pure noise, no structure, and roughly a third of the series still yielded something the detector was willing to name at confidence 0.72 or above.
Now scan 125 chart states instead of one. If a comparable fraction of states can produce a formation from structureless price action, then on any given day the scan will surface some formations that are not there. Not because the detector is broken — because the detector has a threshold, thresholds have false-positive rates, and running the test 125 times means encountering that rate 125 times.
This is the multiple-comparisons problem, and automated scanning is an efficient way to walk into it. Two implications follow:
- A formation found by scanning everything is weaker evidence than the same formation found on a chart you were already examining for other reasons. The scan searched a much larger space to find it.
- The count of formations a scanner reports is a property of its threshold, not of the market. A tool reporting many formations per day across a market list is not seeing more; it is filtering less.
Our own density gives a reference point: 661 completed formations across 182,341 bars is about 3.6 per thousand bars. A scanner surfacing far more than that on comparable coverage is using looser thresholds, which is a legitimate choice — but only if it publishes them.
What "nothing changed" means, and why it is reported
On most days, on most of the 125 states, the honest reading is that structure has not changed. Publishing that is a design rule rather than a judgement call.
The reasoning is straightforward. A daily format that requires fresh drama will produce fresh drama. If the template has a slot that must be filled with something noteworthy, the threshold for noteworthy will drift down until the slot is full — not through dishonesty, but through the ordinary pressure of having a slot to fill.
Reporting "no structural change on the majority of markets" is also the most useful thing to say to someone deciding whether to look closer. And it is the single clearest behaviour separating a tool with a threshold from one without.
Reading a multi-interval result
A practical sequence, given the constraints above:
- Fix the interval first. Decide which question you are asking before you look. The other intervals are context.
- Read the confidence breakdown, not the label. A named formation at 0.73 confidence and one at 0.91 are very different objects with the same name. The components say which.
- Check the sample count for that shape. If it is one of the nine below the floor, there is no published rate, and the shape's reputation is doing the work instead of evidence.
- Treat agreement across intervals as weak corroboration. Intervals of the same instrument are not independent observations; they are the same price series measured at different resolutions.
- Treat an empty scan as a result. It is the most common one, and it is not the tool failing to find something.
What multi-interval scanning is genuinely for
Not prediction, and not finding more opportunities. Two things:
Coverage without drift. A person cannot watch 125 chart states, and a person watching even five applies a threshold that moves with what they hope to find. A fixed procedure applied uniformly is the only way to have looked at everything by the same standard.
A denominator. Because the scan covers every state on a schedule, the number of formations it did not find is knowable. That is what makes a rate definable at all — and it is exactly the thing manual review cannot supply, because nobody can count the charts they never looked at.
In short
- Five intervals across twenty-five markets is 125 chart states. Automation buys uniform coverage, not insight.
- Intervals disagreeing is normal: each defines a different swing threshold and answers a different question. The risk is scanning until one agrees with you.
- Evidence is thinnest where trust is highest. On the daily interval only one shape reached thirty instances in nine years; head and shoulders appeared once. Every published rate is effectively a 4-hour statistic.
- Fourteen of forty pure random walks produced a nameable formation. Scanning 125 states meets that false-positive rate 125 times.
- Detection density is a property of the threshold: 661 formations across 182,341 bars is about 3.6 per thousand bars.
- An empty scan is the normal output. Nothing here is investment advice.