Fib Reaction Memory

Features

• Confirmed structural A→B reference leg from alternating swing pivots, frozen while price retraces (a new impulse only replaces it when a confirmed pivot extends beyond B, or the pullback invalidates past A)
• Micro-pivot reaction detection inside the structural leg, sized in ATR units so weak stalls don’t count the same as real reversals
• Reaction-first, fib-second: historical reactions are located and weighted before they’re checked for Fib proximity, not the other way around
• 0–100 Fib Memory Score per level, combining reaction strength, proximity to the level, optional recency weighting, consumption decay (later tests at the same level count for less) and a broken-level penalty (a reaction whose level was later broken counts for less than one that held), run through a saturating (not linear) aggregation
• Per-level Hold Rate — % of historical reactions near this level that were never subsequently broken, shown alongside the score
• Per-level state machine: APPROACHING / TESTING / REJECTED / ACCEPTED / RECLAIMED, derived from the current retracement and the deepest retracement reached during this pullback
• Score-based NEXT level highlight — the next level ahead of price with the strongest evidence, not simply the nearest geometric Fib
• Persistent line/label objects reused every bar instead of recreated, keeping draw-object usage bounded

Reference leg construction

The structural leg comes from alternating ta.pivothigh()/ta.pivotlow() confirmations (Structural Pivot Length, confirmed only after the right-side lookback bars have closed). A→B stays fixed while price pulls back inside it — a naive “last two pivots” approach would silently swap the reference range out from under an in-progress pullback the moment a smaller counter-pivot confirms. The leg only changes when either a new confirmed pivot extends past B (using the most recent opposite-type pivot after old B as the new A, if one exists), or the pullback exceeds Reference Leg Invalidation (retracement beyond 1.00, i.e. beyond A).

Reaction memory

Inside the frozen A→B leg, smaller pivots (Reaction Pivot Length) are located and each is scored by how far price moved away from it afterward, in ATR units (Minimum Reaction · ATR filters out noise, Strong Reaction · ATR is the size that earns full weight). Each qualifying reaction is normalized onto the 0.0–1.0 A→B range and, if it falls within Fib Reaction Radius of a standard Fib ratio, contributes evidence to that level — weighted by reaction strength, proximity, and optionally recency (Recency Influence). Evidence per level is aggregated with a saturating curve (Score Saturation) so repeated touches don’t produce an unbounded score. Maximum Memory Scan bounds how far back the historical scan runs, for performance.

Reactions are scanned in chronological order (oldest inside the leg first) so that Level Consumption can discount repeated tests of the same level: the first qualifying reaction near a level counts fully, each later one at that same level counts less (1 / (1 + consumptionDecay × testsSoFar)), reflecting that a level typically weakens rather than staying equally valid after repeated tests. 0 disables this (every test counts fully, matching v1.0/v1.1 behavior).

Broken Level Penalty discounts a reaction if the level it’s near was later broken anyway. This is checked by projecting every bar’s own raw high/low (not a historical live-leg series — the leg wasn’t necessarily active back then) onto the current fixed A/B range, then taking the deepest such value reached between the reaction’s bar and now. If that later exceeds the level plus tolerance, the reaction’s contribution is multiplied by 1 - brokenLevelPenalty (default halves it). A reaction near a level that was never subsequently broken always counts fully. 0 disables this.

The same held/broken check per reaction also feeds a Hold Rate per level: the % of qualifying reactions near that level that were never subsequently broken (na, hidden, until the level has at least one qualifying reaction). This is a descriptive statistic about what already happened in this leg, not a probability forecast — see “What this is not” below. Show Hold Rate toggles the label suffix.

NEXT level selection

Among enabled Fib levels still ahead of the current retracement, the level with the highest Memory Score at or above Minimum Score for NEXT is preferred. If no ahead level clears that threshold, the nearest geometric level ahead is used instead. This means the highlighted NEXT level is not necessarily the closest one — a distant level with strong historical reaction evidence can outrank a close one with none.

Level state machine

Each level’s state is derived from two numbers: the current retracement (now) and the deepest retracement reached anywhere during this pullback (pullbackDeepest, which is always at least as deep as the current retracement). This keeps the states mutually exclusive and avoids storing a separate per-level history:

• APPROACHING — the pullback has never reached this level’s tolerance band
• TESTING — the current retracement is inside the level’s tolerance band right now
• REJECTED — the pullback reached the level’s band at some point, then retreated shallower, without ever breaking decisively past it
• ACCEPTED — the pullback broke decisively past the level and is still beyond it
• RECLAIMED — the pullback broke past the level at some point, then retreated back shallower than it again

Reading the chart

Each Fib line’s color and width encode its Memory Score (Strong Score / Medium Score thresholds: gray/thin = little memory, orange = relevant structure, bold/lime = strong structure). The highlighted NEXT level (when Highlight NEXT Level is on) is drawn solid and in a distinct color regardless of score tier. Labels show the level, its score, its state (APPROACHING/TESTING/REJECTED/ACCEPTED/RECLAIMED) and, if Show Hold Rate is on and the level has been tested, its Hold Rate. Show Weak Levels can hide low-score levels below Medium Score to reduce clutter (the NEXT level stays visible regardless). A gray A/B leg line and optional anchor labels mark the current reference leg.

What this is not

A high Fib Memory Score or Hold Rate means the area showed meaningful historical structure within this specific leg — neither is a probability that the level holds going forward, and neither makes a forecast. Both are descriptive statistics about reactions that already happened earlier in this same leg. The state machine is diagnostic only; it doesn’t gate or weight anything. See todo.md for known gaps (pause/acceptance detection, reaction zones independent of Fib ratios, multi-leg history) that were deliberately left out.