Limitations and Trust Boundaries

This is the page the pack is built around. Every other page points here when a misread gets heavy, because this is where the heaviest misreads are treated as main instruction rather than as disclaimers. If you only read one page of this pack thoroughly, make it this one — the others teach the tool, and this one teaches the shape of the ways the tool is most often misread in the wild.

The page is ordered by damage potential. The highest-cost misreads come first. The two items at the top are the ones most readers come to this tool pre-loaded with, usually from other RSI-family indicators or from general chart-reading habits that transfer imperfectly here. The last items are the ones that only surface after some live experience — they are less dramatic but can be more corrosive, because they distort a reader's model of what the pane is saying without ever producing a loud failure.

Read in order. Returning later to skim the headings is useful; skimming on a first read is not. Each item carries a reason the misread is expensive, a description of what the tool is actually doing at the point of confusion, and a plain-language correction you can install as a reading habit.

1. Divergence triangles are not entries

This is the most expensive misread on the pane, and it is not close.

A confirmed divergence triangle describes a pair of confirmed pivots. Bullish: chart price made a lower low at the confirmed pivot pair while the blended RSI at the same offsets made a higher low. Bearish: the mirror. Both evaluated only inside confirmed bars. That is geometry. It is a description of two pivots and the blend values at matching offsets. It is not a prediction, it is not a directional conclusion, and it is not a decision rule.

What the triangle is not, in order of how people misread it:

• A directional call. The engine does not predict the next bar, the next session, or the next move. It reports a geometric relationship that has already formed, in confirmed price action, evaluated on a confirmed bar.

• An entry. Nothing about the triangle's existence promises anything about what happens next. The same triangle can appear at a meaningful trend exhaustion and at a minor pullback that resolves against it; the engine cannot distinguish those two cases because both of them satisfy the same geometric rule. Acting on every triangle is a posture this pack actively argues against, because the triangle carries information about the past and no information about the future.

• A confirmation of something else. A divergence triangle does not confirm a Keltner touch, a Donchian press, a BBWP reading, or a slot color. They are all scoped to the same blend. See the next item.

The way to use the divergence engine well is in Workflows' divergence-as-question card. The triangle is a prompt to look at the underlying slot stack and at the two pivots that produced the geometry; it is not a prompt to act. A reader who routes every triangle into an entry is trading the engine's visual authority, not the engine's actual reporting.

A corollary worth stating forcefully: do not lower Pivot Len to see more divergences. Lower values trade strict pivot logic for triangle spam. Every loose pivot gets counted, and a looser pivot is not a stricter one — it is a pivot that required less price action to confirm, so the geometry it participates in is correspondingly weaker. Pivot Len is the strictness dial, not the sensitivity dial, and chasing more markers is the fastest way to drown in noise that will still look like evidence on the chart.

2. The four structure features are not independent confirmations

Second most expensive misread, and for many readers it is the most insidious because it looks like rigor. A reader who wants to be disciplined will often reach for multi-feature agreement as evidence of caution; on this pane, multi-feature agreement is exactly where caution gets undermined by construction.

Divergence, Keltner, BBWP, and Donchian all derive from the same blended RSI line. When all four point the same way at the same moment, they are one line reported from four framings — not four independent witnesses agreeing. The feeling of confluence is built in by how the features are constructed.

What this means in practice:

• Apparent agreement between features is not evidence. It is the geometric consequence of deriving four things from one input. The four features are useful because each one asks a different question about the blend; they are not useful as mutual confirmations because they share an input.

• The word "confirms" does not belong between two structure features. They cannot confirm each other; they share a source. Usable language: "co-exists with," "is consistent with," "happened at the same time as." Avoidable language: "confirms," "validates," "corroborates."

• If the blended line is misconfigured (duplication trap, mixed-posture inputs, mismatched cross-ticker), the misconfiguration propagates into all four features. The confluence of wrong looks exactly like the confluence of right — and this is the dangerous case, because the pane will look most decisive and most coherent at precisely the moment it is least trustworthy. A misconfigured pane does not fail loudly; it fails with visual confidence.

The co-movement inspection drill in Workflows is the cleanest way to feel this. Toggle one slot's weight or enable state and watch every one of the four features move together. That is the dependency. Do it once before you trust this pane for a live decision, so that when you see a four-feature agreement later, some part of your reading immediately discounts the visual weight.

3. Alignment is not breadth — and STR makes the trap larger than Base

The alignment alerts (All RSI Slots Bullish, All RSI Slots Bearish) report that every enabled slot is in the same RSI-vs-signal regime on a confirmed bar. What they do not report is independent agreement among five independent witnesses.

When enabled slots share source, RSI length, smoothing, and MA family — differing only in timeframe — alignment is one measurement sampled at several cadences. A reader who reads five-way alignment as five independent votes is reading the same value five times and counting it as five pieces of evidence.

The trap is larger on STR than on Base because the visual authority of five-way self-agreement is larger than three-way self-agreement. Five matches on a pane feel convincing in a way three matches do not. The feeling has no relationship to whether the slots are actually differentiated.

The prevention is differentiation. If every enabled slot carries at least one axis of variation beyond timeframe (different source, different length, different MA family, different ticker), alignment starts to mean something closer to breadth. If not, alignment is decoration.

4. Slot-05 alignment current behavior

A current behavior in the script, as of this pack's writing: the All RSI Slots Bullish and All RSI Slots Bearish alerts do not fire when slot 05 is enabled. The alignment counter's slot-05 arm contains additional increments past the intended single increment; the count is inflated past what the per-slot bullish count can ever reach. Enabling slot 05 therefore disables the alignment alerts for as long as the script contains the current behavior.

What the pack says and does not say:

• This pack does not edit Pine source and does not promise the behavior is fixed.

• The default ships slot 05 disabled, so most first-session configurations are unaffected.

• The workaround is plain: until upstream remediation, treat the all-slot alignment alerts as four-slot-maximum signals, and do not rely on them when slot 05 is enabled.

• Per-slot alerts, blend-state alerts, and divergence alerts are unaffected regardless of whether slot 05 is enabled.

See Alerts for the alignment-alert block that carries this disclosure in context, and Troubleshooting for the symptom-to-cause row.

5. Per-slot On Bar Close? is an upgrade with new failure modes

Base offered one global On Bar Close? switch. STR offers five, one per slot (shared between the slot's RSI pass and signal pass). The finer control is an expressive upgrade. It is also where the specific STR mixed-posture misread originates.

Two failure modes worth naming up front:

• Unintentional mixed posture. A reader changes slot 03's On Bar Close? to false because they want a faster HTF read, and leaves the other slots at true. The blend is now fed by a mix of confirmed (from the true slots) and live-and-repainting (from the false slot) members. Structure features over that blend inherit the mix. Keltner and Donchian can look jumpier than a reader expects for reasons that live in the input posture, not in the feature math.

• Intentional mixed posture without the structure-feature caveat. A reader deliberately wants a mixed posture for a specific purpose and then forgets to account for it when reading BBWP or Keltner. The posture is legitimate; the blind spot is the problem.

The full treatment is on MTF and Repainting.

6. BBWP is not price volatility

The BBWP columns rank the blended RSI's own Bollinger-band width as a percentile of the blend's own history. Every part of that read is scoped to the blend.

This creates a reliable misread, because BBWP columns look like the familiar price-side BBWP many readers have seen on overlay tools:

• A tall BBWP column on this pane means "the blend's width is ranking high against its own history." It does not mean price volatility is high.

• A short BBWP column on this pane means "the blend's width is ranking low against its own history." It does not mean price is quiet.

The two can diverge cleanly. A session with wild price movement and a relatively stable blend (say, because slot weights and smoothing are calm) can produce short BBWP columns while price is loud. A session with calm price movement and a widening blend (say, because a cross-ticker slot is carrying volatility) can produce tall BBWP columns while price is quiet. Both are the tool reporting accurately at its layer.

The qualifier "from the blend, not from price" rides every mention of BBWP in this pack. Apply it when you teach BBWP to anyone else, or the misread keeps propagating.

7. Keltner touch is not overbought or oversold

The Keltner envelope wraps the blended RSI's own smoothed basis with bands sized by the blend's own bar-to-bar range times a multiplier. A touch of the upper band means the blend is stretched against its own basis, measured by its own recent range.

What it does not mean:

• It does not mean "overbought."

• It does not mean a reversal is likely.

• It does not say anything about price. The envelope is in blend-space.

Keltner upper touches happen during sustained bullish blend regimes and do not resolve themselves into exhaustion calls on their own. Reading them as overbought is a habit from classical single-timeframe RSI and from some price-side Keltner uses. Neither transfers here.

8. Plot On Pivot is not historical visibility

Plot On Pivot is a visual choice. ON back-shifts the divergence triangle by Pivot Len bars to the original pivot. OFF places it at the right-shoulder confirmation bar.

The setting does not change when the alert fires. The alert evaluates only on confirmed bars; both modes fire on the confirmation bar. The marker's back-shifted position is a more-honest picture of where the geometry formed. It is not a record of what a reader could have seen in real time.

Both truths stay side by side. ON is honest about the geometry's origin. ON is misleading-looking about when a reader could have acted. If you use ON, remind yourself on every chart review that the marker's presence at the back-shifted bar is a post-hoc visualization, not a real-time observation.

9. Hide Plot and weight-zero are not kill switches

Two visual-looking inputs that feel like kill switches and are not:

• Hide Plot removes the slot line from the pane. The slot continues to compute, contribute to the blend, and fire its own alerts as long as Enable is on.

• Weight = 0 removes the slot from the weighted-mean blend. The slot continues to compute, continues to plot (subject to Hide Plot), and continues to fire its own alerts.

Only the slot's Enable input is a genuine kill switch. If you want a slot fully off — no computation, no contribution, no alerts — disable it. Using Hide Plot or weight-zero as a silent kill leaves the slot active in ways you cannot see from the pane.

10. Master smoothing is not a reliability upgrade

Master smoothing applies a single MA pass to both blended values after the weighted-mean aggregation. The result is a calmer, slower blend.

What the calming costs:

• Blend-based alerts — the two blend-state alerts — fire later, not earlier.

• The blend becomes less reactive to underlying slot changes.

• Per-slot alerts are unaffected; only the blend's timing changes.

The misread is "the pane looked choppy, so I enabled master smoothing, and now it looks cleaner, therefore my read is more reliable." Calmer is not truer. If the chop carried information, the smoothed version is carrying less of it. Enable master smoothing when you have a stated reason — for example, you trade on a cadence slower than the blend's natural reactivity, and the intra-session chop is noise to you — not when you want the pane to look neater.

11. 70 and 30 are not triggers

The reference bands at the user-configured overbought and oversold levels are visual bumpers. They help you orient on the bounded axis. They do not participate in any alert condition; they do not drive any color rule; they do not interact with the structure features.

Moving them does not trigger anything. Crossing them does not trigger anything. If your workflow depends on threshold-crossing, that condition is not wired in the current script. See Alerts for the full inventory and the set of deliberate absences.

12. Warm-up periods are real

Several parts of the pane return nothing until they have enough data:

• Each slot's signal pass needs enough history before its color-driving signal is mature. In the common warm-up path, the helper substitutes the slot's current smoothed RSI until the signal MA can return a value, so equality produces the faded/down tone and no bullish or bearish state alert. A defensive 50-line fallback exists only if the signal value is truly unavailable.

• The BBWP percentile-rank needs a full lookback window (default 252) of prior blend widths before columns print. On sparsely-populated instruments, or near the start of a chart, the BBWP strip can be silent for a long stretch.

• The divergence engine needs a confirmed prior pivot in each direction before it has a comparison to make.

A reader who scrubs to the left edge of a chart or loads a thin instrument and reads the missing values as bugs is misreading warm-up behavior. Wait for enough history.

13. Cross-ticker mismatch is a silent misread

The cross-ticker slot is one of STR's best features and one of its easiest mis-configurations. A cross-ticker slot inherits the full per-slot pipeline — source, length, MA families, weight — and feeds its RSI into the blend.

If the chosen cross-ticker's market dynamics (liquidity regime, trading session, scale) are structurally different from the chart symbol's, the blend is reporting a hybrid read across two regimes. That can be exactly the question you want to ask; it can also be a silent source of a read you will not be able to interpret.

Preventative discipline: before you weight a cross-ticker slot aggressively, do the before-and-after drill in the cross-ticker scenario in Workflows. Know what the blend looks like without the cross-ticker before you read it with the cross-ticker contributing.

What the pane is, and what it is not

The pane is a reading tool. It describes a blended RSI, stacked across cadences and optional tickers, with four framings of the blend to inspect it from different angles. It is an input to decisions you are making elsewhere, with a posture, a method, and a risk model that belong to you. The pane does not replace those; it informs them.

The pane is not a signal service. It does not predict prices. It does not call turns. It does not produce a trading edge on your behalf — any edge you find using this tool will be one you built by combining the readings here with your own method, your own risk discipline, and your own history of live decisions. This is not a sales posture. It is a working description of what a reading tool can and cannot do, and why the pack keeps saying it.

Hold both of those at the same time — the pane is useful, and the pane is not a shortcut — and this page does its job.