What an Indicator Is Actually Computing

A study is a function of the bars, and nothing else

Whatever an indicator is called, and whatever its description says it detects, its input is the series of numbers already on the chart. It takes the bars, applies arithmetic, and returns another series to be drawn. It does not read order flow, positioning, anybody’s intentions, or news that has not yet reached the price.

And the bars it reads are themselves a summary — a handful of numbers standing in for everything that happened in a period. So a study sits at two removes from the market: arithmetic over a compression. That is not a criticism of any particular tool. It is the boundary of what any of them can be asked.

Four settings decide the output, and one of them is not in the indicator

The first is the price source: which number per bar the calculation reads. The close is the usual default, but the open, the high, the low, the midpoint of the high and the low, or an average of several are ordinary alternatives. The second is the lookback, the count of completed bars consumed to produce one value. The third is smoothing, where the result is averaged again before it is drawn.

The fourth is not an indicator setting at all. It is the chart’s time frame, which silently decides how much real time a fixed number of bars covers, so the same lookback spans minutes on one chart and months on another. Change any of the four and every value moves — every crossing, every extreme, every divergence anyone reads off it. Two people running what they both call the same indicator can disagree completely without either of them misreading anything.

Where the line is drawn tells you its units

A study plotted on the price chart itself is expressed in the instrument’s own prices, which is why it can sit alongside the bars and be compared with them directly. A study plotted in a panel below is not a price at all: its output is a difference, a ratio, or a position within some range, on a scale its formula defines.

The distinction is about units, and it settles what may honestly be compared with what. A panel value is not a level. Two panel studies on different scales cannot be read against one another without knowing what each scale is. And "the reading is high" is not yet a statement — it becomes one only when it says high relative to what, which is a property of the formula and is documented with it.

The delay is arithmetic, not a defect

Any calculation that averages a window of past values responds to a change only as new values enter the window and old ones leave it. A long window is pulled toward a new price gradually, because the old prices are still inside it. A short window turns sooner, and also turns for movement that comes to nothing. Weighting recent values more heavily shortens the delay without abolishing it, because the window still contains the past — that is what a window is.

The consequence is worth stating without hedging. A study describes prices that have already occurred. A line turning up records that recent prices were higher than earlier ones. Tools described as leading are computed from the same closed bars as everything else; they differ in the transform, not in what they are able to see. No setting converts a function of the past into information about the future.

A bounded reading is a fact about the scale

Some studies are built to stay inside a fixed range. When such a reading sits near its limit, what has been established is that the formula’s own arithmetic has reached the end of its scale. Price has no equivalent limit, so during a sustained move the study can arrive at its extreme early and stay there while the move continues — the behaviour people find surprising, and simply what a bounded function does.

Nothing in the calculation contains a valuation of the instrument. The level at which a reading is called extreme is a threshold somebody chose, published with the study or set on the chart, and it can be moved. Any meaning past "this is near its limit" was supplied by the reader rather than by the tool.

A signal line is a second derivation, not a second opinion

Many studies draw two series: the calculation, and a smoothed copy of it. Because the second is derived from the first, it moves the same way with more delay, and the distance between them expresses whether the first is changing faster than its own recent average. Some tools plot exactly that distance as a histogram, which is the same information a third time.

A crossing of the two is therefore not the arrival of information. It is the moment an existing change in the first series grew large enough to overcome the smoothing, and its timing is set by the smoothing length. Shorten that length and the identical chart produces more crossings, earlier. Which series a platform calls the signal, and how it smooths it, is defined in the study’s own specification.

Agreement between studies is often one measurement counted twice

Several tools pointing at the same place feels like corroboration, and the vocabulary around it treats them as independent observations. Usually they are not. Almost everything on a chart is computed from the same series of prices, so an average, a band constructed around that average, and a panel study built from the same closes are three transformations of one input rather than three witnesses. Correlation by construction is not evidence.

Adding more lines makes coincidence easier to find as well, because with enough of them something is always near something else. What can actually be established is narrower and more useful: where each reading came from — the price source, the lookback, the smoothing and the time frame. That is the difference between a reading somebody else can reproduce and one they can only agree or disagree with.