For the Geeks

This page exists because STR's mechanics are non-trivial and a careful reader deserves a mental model that builds real trust without publishing the pieces that would make the tool cloneable. The whole page sits at mental-model altitude — what things are, why they exist, what they trade off against simpler alternatives, and how you can verify each claim yourself on a live chart.

The pack's posture on this page is specific. A reader who wants to trust the tool deeper than "the vendor said so" has a legitimate need and deserves an honest answer. A reader who wants enough information to reconstruct the tool is asking for something the pack is not going to provide, and the line between those two readers is what shapes what follows. Everything below is told at a level where you can verify it, argue with it, and build a working mental model that survives live use — but not at a level where the specific internal choices can be lifted into a new script.

What this page will and will not tell you

What it will tell you. The shape of the pipeline. The stages it runs through. What each stage is asking, why that stage exists, and what you give up when you add or remove it. The structural framings of the four blend-derived features. The tradeoff each feature represents versus a simpler alternative. The structural distinction between this pane, classical single-timeframe RSI, and Stoch-RSI. A reader verification path at the end that lets you confirm the mental model empirically without needing to see the source.

What it will not tell you. Internal coefficients of the Axiom MA library's Pro trim beyond what that library's own manual already publishes. Parameter choices that would let you reconstruct the pane from scratch. "Best" tunings for any control. Internal constants of the structure-feature constructions. The specific handling of pivot-boundary edge cases. Any line that doubles as a recipe.

This is not caginess and it is not a puzzle. It is the scope the pack deliberately takes, for two reasons. First, the Axiom MA library is a separate product with its own teaching surface — family internals, tuning hypotheses, and their tradeoffs live in that library's own manual, and this pack's job is to cross-link rather than duplicate. Second, the structure-feature constructions are accessible at conceptual altitude — you will leave this page able to describe what each feature does and why, to debate whether a different construction would be better for your use, and to verify the page's claims on a live chart, without needing the specific numbers the script uses internally. That is the trust this page is trying to build.

Ring one — the double-smoothing plus weighted-blend pipeline

The first ring of mechanics is what happens inside each slot and how the slots aggregate into a blend.

The per-slot pipeline

Per slot, the pipeline runs in stages. At each stage, there is a choice the configuration has made, and that choice has a cost.

• Stage A — raw RSI. The slot computes classical RSI on its chosen source at its chosen timeframe, at its chosen length. If the slot carries an optional cross-ticker, the RSI runs on that ticker at the chosen timeframe. This stage produces a fast, noisy, familiar read.

• Stage B — RSI smoothing pass. The raw RSI goes through an MA pass. The family is the slot's RSI Type; the length is the slot's RSI Smoothing Length. This is where the slot's first layer of character is installed. An EMA pass feels different from a Jurik pass feels different from a Laguerre pass — and each family has its own defaults inside the Axiom MA library (cross-reference the MA library manual's Pro trim for family internals).

• Stage C — signal pass. The smoothed RSI goes through a second MA pass at the slot's Signal Length and Signal Type. This second pass produces the slot's signal line, which is never plotted but which drives the slot's color and the slot's alerts.

• Stage D — clamp. Both the smoothed RSI and the signal are clamped into 0..100. The clamp is not a rescaling. It only prevents arithmetic excursions outside the native RSI range; it does not move values proportionally. This is an important distinction from Stoch-RSI, covered below.

• Stage E — bar-close return posture. If the slot's On Bar Close? is true, the slot returns the previous bar's clamped values from its HTF context; if false, it returns the current bar's values. This is the repaint-posture choice treated in MTF and Repainting.

Tradeoffs at each stage, named at altitude. The length choice at stage B is a reactivity-versus-smoothness tradeoff; the family choice is a lag-versus-responsiveness-versus-noise tradeoff whose specifics live in the MA library manual. The signal pass at stage C is doing the same thing one more time, and its length is how much slower you want the signal to be than the smoothed RSI. The clamp at stage D costs you nothing meaningful; the alternative (letting values excurse) would add failure modes without adding information. The repaint posture at stage E is honesty-versus-reactivity.

The blend

After the slots compute, the blend is a weighted mean. Each active slot contributes its clamped smoothed RSI value and its clamped signal value, weighted by the slot's weight input. Slots with weight zero are skipped at the blend stage. Slots with non-available RSI (warm-up) are skipped. The weighted mean produces two blended values — a blended RSI and a blended signal — which are then optionally processed through master smoothing.

Why a weighted mean. An unweighted mean would treat every active slot as equally important, which is rarely what a reader actually means when they set up a five-slot stack. A weighted mean lets the reader express "slot 02 is my primary read, slot 05 is a proxy voice at 10% influence." The blend's shape is a direct function of weight allocation.

Tradeoff versus a simpler alternative. A plain stacked RSI without a weighted blend is easier to read visually and requires less configuration discipline. The cost is that the reader is informally computing a blend in their head, with weights that are opaque even to themselves. The weighted-mean blend makes the reader's informal blend explicit. Readers who configure carelessly make worse reads with this tool than they would on a simpler one; readers who configure deliberately make better ones.

Master smoothing — an optional third pass

If enabled, master smoothing applies a single MA pass to both blended values, using its own family and length, then re-clamps into 0..100.

• What it costs. Timing. A smoother blend is a slower blend, and blend-state alerts fire later.

• What it buys. A visual read that is less reactive to short-term oscillation.

• What it does not do. It does not affect per-slot alerts or the all-slot alignment alerts, which read per-slot states upstream. It does affect blend-state alerts and every structure feature, because divergence, Keltner, BBWP, and Donchian read the post-smoothing blend when master smoothing is enabled. Read Settings and Limitations and Trust Boundaries for the full trade.

Ring two — the four blend-derived structure features

The second ring is what happens around the blended line. All four features read from the same blend. None of them writes back into the blend. They are four independent framings of one underlying line.

The key structural truth, stated once here and reinforced across the pack: these features co-move by construction. They cannot confirm each other. Their agreement is geometric, not observational.

Divergence — confirmed-pivot geometry

What it considers. Chart low and chart high series. The blended RSI line at pivot-lookback offset. The Pivot Len parameter (used as both left and right pivot strength). The barstate.isconfirmed gate.

What it does. Waits for a chart-price pivot to confirm — Pivot Len bars of price action on each side of the pivot. Pairs the confirmed pivot with the blended RSI value at the same offset. Compares to the most recent confirmed pivot pair in the same direction. Bullish rule: chart price lower-low with blend higher-low. Bearish rule: chart price higher-high with blend lower-high.

Tradeoff versus a simpler alternative. Manual eye-ball divergence detection — a reader drawing lines on a chart — is flexible and cheap to compute. It is also notoriously subject to cherry-picking. The engine here trades flexibility and real-time responsiveness for strict pivot logic and confirmed-bar gating. You get late-by-design timing in exchange for a triangle that does not lie about which pivots are being compared.

What the pack will not publish. The internal pivot comparison beyond "left and right lookback match at Pivot Len." Any description of pivot-state caching or the comparison's handling of edge cases at the pivot boundary. "Best" Pivot Len for a market. These protect the engine from reconstruction and the reader from prescription.

Keltner envelope — stretch framing

What it considers. The blended RSI line. The basis MA family and length. The KC Length (range EMA length). The KC Mult (multiplier).

What it does. Smooths the blend into a basis. Estimates the blend's bar-to-bar absolute change as an EMA at KC Length. Multiplies the range EMA by KC Mult to produce an offset. The upper band is basis plus offset; the lower band is basis minus offset.

What this means. A Keltner touch is a statement about how stretched the blend is against its own smoothed basis, relative to its own recent bar-to-bar range. It is a blend-space reading. Nothing about price-space volatility or exhaustion is being asserted.

Tradeoff versus a simpler alternative. A Keltner on price is a familiar chart feature. Keltner on the blend trades direct price-space interpretability for clarity about the blend's own stretch — which is the thing the reader of this pane is actually trying to evaluate. The cost is the transfer-misread risk ("Keltner means overbought"); the benefit is that the answer is scoped to the line the reader is reading.

What the pack will not publish. The internal rationale for that range-estimate choice, or protected tuning details around the envelope construction.

BBWP — regime framing

What it considers. The blended RSI line. The Length parameter (used for both the Bollinger basis and the standard-deviation window — a matched-length Bollinger construction). The Lookback parameter. The basis MA family.

What it does. Computes a classical Bollinger-band width on the blended RSI. Ranks the current width as a percentile of prior blend widths across the Lookback window. Returns nothing until the Lookback window is fully populated.

What this means. A BBWP column is a regime statement about the blend's own width history. Tall means the blend's width is high against its own history; short means low. The read is entirely in blend-space — the most common BBWP misread is reading the columns as a statement about price volatility, and this pack names the misread on every page that mentions BBWP.

Tradeoff versus a simpler alternative. A price-side BBWP is a direct volatility regime read. A blend-side BBWP is a regime read of the thing you are already reading — the blend. The cost is the transfer-misread risk. The benefit is that the question "is the blend's width regime high or low" is answerable without leaving the pane.

What the pack will not publish. The Bollinger-width construction constant. The percentile-rank helper's full validity rule beyond "a full lookback window is required."

Donchian channel — range-edge framing

What it considers. The blended RSI line. The Channel Length. The basis MA family and length for the optional midpoint smoothing.

What it does. Marks the blend's highest and lowest over the channel length as steplines. Optionally smooths the midpoint with the chosen basis MA. When the blend prints a fresh extreme, the corresponding stepline moves to it.

What this means. A Donchian press is the blend sitting at its own recent range edge. A channel breakout on this pane is a blend breakout, not a price breakout.

Tradeoff versus a simpler alternative. A price Donchian is a familiar range-edge tool. A blend Donchian tells you where the blend is in its own range, which is the thing you are already reading. Same tradeoff shape as the Keltner and BBWP comparisons: transfer-misread risk for the benefit of an answer scoped to the blend.

The co-movement truth, stated structurally

Divergence, Keltner, BBWP, and Donchian all read the blended RSI line. None writes back into the blend. When the blend changes — because a slot changed its enable state, weight, source, timeframe, MA family, length, or repaint posture — all four features change with it. They change simultaneously because they are looking at the same line from different angles.

That means:

• Apparent agreement between features is geometric, not evidentiary.

• The word "confirms" cannot honestly link two structure features. They share a source.

• A reader who weights multi-feature agreement heavily is often double-counting one observation.

Internalizing this is the main deliverable of this page. The Workflows co-movement inspection drill is the empirical verification.

Structural distinction from classical RSI and Stoch-RSI

This is the structural clarification the pack considers most important for a careful reader, because it fixes a class of misreads that come from prior RSI tools.

• Classical single-timeframe RSI. Runs on one source at one timeframe, produces a single RSI line. No blending. No signal line. No re-normalization. A reader of classical RSI has one number to interpret, and the familiar references at 30 / 50 / 70 are single-timeframe single-symbol observations.

• Stoch-RSI. Takes the classical RSI output and applies a stochastic oscillator over it. The stochastic step re-normalizes RSI into a secondary 0..100 range — the output is the percentile position of RSI within its own recent range, not RSI itself. The secondary re-normalization is the defining structural feature. The read at 70 on a Stoch-RSI is a different thing from the read at 70 on a classical RSI; they are categorically different observations.

• This pane. Runs multiple classical RSIs (slots) at possibly different sources, tickers, and timeframes. Smooths each one with an MA pass (stage B). Smooths the smoothed RSI with a second MA pass (stage C, produces the slot signal). Clamps both into 0..100. Blends the slots with a weighted mean. Optionally applies master smoothing to the blended values. Four structure features frame the blend.

The critical distinction: this pane does not re-normalize RSI into a secondary range. The 0..100 axis is native RSI, clamped but not rescaled. The clamp prevents arithmetic excursions outside the native range; it does not move values proportionally. A reader who transfers Stoch-RSI habits onto this pane is importing the wrong category of tool.

Structurally:

If this table leaves you with one sentence, let it be: the 0..100 on this pane is native RSI clamped, not stochastic-rescaled RSI.

Reader verification path

You can confirm the mental model above without reading any source. Do these five things on a live chart, in order.

1. Verify the pipeline is double-smoothed, not rescaled. Enable only one slot with defaults. Open a classical RSI indicator alongside this pane using the same source and length. The two lines will not match — this pane's slot is the classical RSI after a smoothing pass, then compared against a second-smoothed signal. Watch them diverge at sharp price turns; watch the lag structure make sense. Then add a Stoch-RSI to a third pane; notice that Stoch-RSI's output is qualitatively different from either.

2. Verify the blend is a weighted mean. With slots 01, 02, 03 at equal weight and defaults, set slots 04 and 05 to weight zero. Confirm the blend sits roughly where the three slots' average sits. Raise slot 01's weight to 60 and the others to 20 each. The blend pulls toward slot 01. Restore equal weights.

3. Verify the structure features read only the blend. With the blend and all four features active, change one slot's weight meaningfully. Keltner shifts, BBWP columns re-rank, and Donchian steplines re-anchor because their input changed. The divergence engine also recalculates from the changed blend, though you may only see the effect when confirmed pivot geometry is present. That is the co-movement dependency.

4. Verify On Bar Close? timing. Set one HTF slot to true, watch it step at HTF closes. Flip to false, watch it update intra-bar. Flip back.

5. Verify Plot On Pivot is a visual, not a timing claim. On a recorded chart with a confirmed divergence, toggle Plot On Pivot ON and OFF. Watch the marker shift; watch the alert log remain anchored to the confirmation bar.

If any of those five produced a surprising result, re-read the relevant section above. If any of them produced a result that contradicts this page, surface the specific case and the pack owes you an update.

Where to go next

If you read this page because you wanted to trust the tool more, the thing the pack asks you to do next is to put it in front of your actual charts and let the four-stage reading order in Visuals and Logic do its work over a few sessions. Mechanical trust is earned through use, not through documentation; the documentation is here to keep the use honest while you accumulate the experience. A pane that makes sense on paper and a pane that makes sense in front of live charts are not the same pane, and only one of them is the one you will eventually rely on.

If the page raised a question it did not answer, that is the intended outcome for some questions — the protected-details boundary exists on purpose — and the unintended outcome for others. If you are sitting on a question that feels like it should have been answerable here and was not, surface it. The pack is maintained, and a question that should have been addressed and was not is exactly the kind of feedback that improves the next revision.