For the Geeks

This page is a trust surface, not a reference implementation. It is here for the reader who wants to understand the conceptual shape of the pipeline — enough to decide whether the tool is doing something they can defend — without being handed a recipe that would let a third party rebuild it. If that sounds like a compromise, it is a deliberate one: the value of the tool is in using it well, and a reader who wants an exact recipe is a reader building their own tool rather than using this one.

Before the shape, the boundary. The next section names what this page does and does not disclose, so you know what kind of material you are reading and you can stop when you have what you came for.

The disclosure boundary

Disclosed on this page:

• The three conceptual stages of the pipeline and what each stage does at a level of abstraction.

• The tradeoff each stage introduces.

• The structural difference between this pane, textbook RSI, and Stoch-RSI.

• The reader-side verification moves that let you feel each stage responding.

Not disclosed on this page, and not available elsewhere in the pack:

• Formulas or pseudocode for any stage.

• The internal implementation details of the Lite moving-average library beyond the family names exposed in the settings.

• Specific numeric constants inside the pipeline beyond the reader-visible 0 and 100 rails.

• Reproducible recipes that would let a third party rebuild the smoothing cascade or the blend rule from this page alone.

The boundary lives where it does for one reason: the tool's value is in using it well, not in reconstructing it. A reader who wants to understand why a slot is calmer than they expected, or why the blend sits near 50 when the slots disagree, can get there through the conceptual shape below. A reader who wants an exact recipe has wandered off the use case this pack is built for.

Structural comparison — three constructions, one column each

The table is the cheapest way to install the structural distinction. If the comparison you carried in from textbook RSI or Stoch-RSI no longer maps cleanly, that is expected; this is a different pane and it was built to be a different pane.

One honest note on the Stoch-RSI row: Stoch-RSI earns its extra reactivity by rescaling the RSI into a smaller band before smoothing, which saturates at the band's edges during sustained trends. Base does not do that rescaling. The plotted values stay in the native RSI band, and the second smoothing pass adds directional color without compressing the value range. The two instruments answer different questions; neither replaces the other.

The three conceptual stages

Stage 1 — Per-slot double smoothing

What happens: a standard RSI is computed on the slot's source at the slot's higher timeframe. That RSI is smoothed by a moving-average pass into the slot's RSI line. The RSI line is smoothed again by a second moving-average pass into the slot's signal line. Both outputs are defensively held inside 0..100 by the clamp each stage applies after its smoothing.

What the clamp is and is not: the clamp is a belt-and-suspenders step after smoothing. The inspected Lite families are ordinary TradingView moving averages over a bounded RSI input, so the pane's native bounding comes from RSI itself; the clamp keeps the output inside the frame instead of turning the tool into a different oscillator. A line pinned at 0 or 100 means the final plotted value is at the rail. Do not read a pin as a stronger claim than a non-pin.

Stage 1 tradeoff: smoothness versus delay. Short smoothing keeps the line close to the raw RSI; long smoothing stabilizes the line at the cost of lag. The second smoothing pass compounds with the first — a long RSI smoothing plus a long signal smoothing produces a slot that is meaningfully slower than either length would suggest on its own.

Reader-side verification: set one slot's RSI length to 14, its RSI smoothing to SMA 3, and its signal smoothing to SMA 3. Watch the slot. Raise the RSI smoothing to SMA 9. Watch it again. You will see the slot line pull toward the midline and respond more slowly. Raise the signal smoothing to SMA 9 on top of that. The color flips will arrive later than they did with the short-signal configuration.

Stage 2 — MTF weighted blend

What happens: for each enabled slot with non-zero weight and a non-na RSI, the blend accumulates the slot's RSI scaled by its weight into a running sum and divides by the total weight at the end. Same for the signal. If no slot qualifies on a given bar — every slot disabled, every weight zero, every slot still warming up — the blend returns na for that bar. That na is expected, not a bug.

Why a weighted average is a coherent blend across timeframes: RSI is unit-less and bounded by construction. There is no price-scale term to reconcile when averaging a 5m RSI with a 60m RSI, unlike a MACD blend which has to deal with price magnitudes changing across timeframes. Averaging bounded, unit-less values across timeframes is cheap to do honestly; the hard part is not the math, it is the interpretation. The blend is a composite the reader shaped by setting weights, not a vote and not a consensus.

Stage 2 tradeoff: coverage versus autocorrelation. More slots on more timeframes give you wider cadence coverage. The slots are not independent — they share a single chart instrument, so a move on the fast slot cascades into the slower slots. Alignment across the blend's inputs is a cadence-coverage read, not a breadth read. The Limitations and Trust Boundaries page carries the full consequence.

Reader-side verification: set slot 02's weight to 0. The blend recomposes from slots 01 and 03 only. Raise slot 03's weight to 60. The blend leans visibly toward slot 03's values. The weighted-average shape is visible in the response.

Stage 3 — Optional master smoothing

What happens: if master smoothing is enabled, one additional moving-average pass runs on the blended pair before plotting. The final output is clamped into 0..100. The per-slot lines and per-slot alerts are untouched — master smoothing is a blend-only layer.

Stage 3 tradeoff: calm versus buried evidence. A calmer-looking blend at a lag cost. Long master lengths on top of long per-slot signal smoothing can flatten the blended pair through regime shifts that the slot lines still show clearly. Master smoothing is not a correctness layer; it is a visual calm layer with a cost.

Reader-side verification: during a regime change on the instrument, enable master smoothing with a visible length (say EMA 7). Compare the blended pair's transition to the slot lines underneath. The blend's transition will be delayed relative to the slots. That delay is the master lag, not the tool being wrong.

Why the boundary lives where it does

Two pieces explain the shape of this page.

First, the single most valuable thing for a reader to understand about the pipeline is the order: per-slot first, blend second, optional master pass third. Once that order is internal, the reader can reason about where their symptom is coming from — a slot-level symptom lives in Stage 1, a blend-level symptom lives in Stage 2, a smoothed-blend symptom lives in Stage 3. Most of the confusion this tool produces resolves as soon as the reader locates the symptom in the right stage.

Second, the thing that does not help you use the tool better is a formula-level disclosure of how each stage is parameterized. If you wanted to reproduce the tool, you would not be on this page — you would be writing your own tool. The protected categories (MA family internals, coefficient recipes, reproducible parameterization, library source) are protected because disclosing them trades a trust-building conceptual view for a reproduction manual, and the audience for the latter does not overlap with the audience for this pack.

If you want deeper smoothing controls than the Lite palette exposes here, that is outside this Base surface. The inspected Lite library for this tool exposes SMA, EMA, RMA, WMA, VWMA, and SWMA; it does not expose ALMA, Jurik, KAMA, FRAMA, or per-family inner knobs through this script. The pointer is not a recruitment to another rung; it is a scope statement about what this trim actually contains.

What this page deliberately did not teach

• Specific numeric defaults for any MA family.

• How the blend handles partial na across warming slots in any detail beyond "slots that are na or zero-weight are skipped, and if no slot qualifies the blend is na."

• Any ordering between pre-clamp and post-clamp values that would let a reader reconstruct the master-smoothing input stream.

• Any exact coefficients of the clamp, the smoothing passes, or the blend beyond the reader-visible 0 and 100 rails.

If you read this page and came away wanting that detail, the page did its job; the boundary is where the tool keeps its working internals. The things you need to use the pane well are above this line. The things that would let you rebuild the pane are not on the other side of a hidden link — they simply are not published, by design, and the pack will not route you to them through a back channel.

Where to go next

• The reading order that maps to Stage 1 and Stage 2 of the pipeline — Visuals and Logic.

• The repaint conversation tied to the higher-timeframe request that Stage 1 runs inside — MTF and Repainting.

• The alignment trap in depth, which is a reading consequence of Stage 2 — Limitations and Trust Boundaries.

• The workflows that match the pipeline to practice — Workflows.