Failed Breakout / Shakeout Trap Pipeline — Design Specification

Date: 2026-03-30 Status: Draft Pipeline ID: shakeout Source: Reminiscences of a Stock Operator (Livermore), Market Wizards (Schwager), Wall Street Phantom (Simpson) Related: Pure price action — no fundamentals required

1. Overview

1.1 Concept

When price breaks through a key level (support or resistance) but immediately reverses back through the level — trapping weak hands who entered on the breakout. The "smart money" has used the breakout to distribute to weak buyers (shakeout) or cover shorts (false breakout), then pushes price back through the level.

This is a mean-reversion / contrarian setup that trades the reversal, not the breakout. It is the mirror image of the breakout pipelines (Darvas, Turtle, Minervini) — where those catch continuation after a clean break, the shakeout catches the failed break.

1.2 Two Variants

Variant	Direction	Mechanism	Entry Trigger
Long Shakeout Trap	Bullish	Price breaks below support → stops hit → immediate reversal above	Strong bar closes above broken support
Short False Breakout	Bearish	Price breaks above resistance → weak buyers chase → immediate reversal below	Strong bar closes below broken resistance

Both share the same scoring logic; direction is determined by:

Trend filter bias: In uptrend (EMA stack rising), prefer long shakeouts. In downtrend, prefer short false breakouts.
Level type: Support break → long shakeout. Resistance break → short false breakout.

1.3 How It Complements Existing Pipelines

Existing Pipeline	Entry Trigger	Opposite / Complementary
Darvas Box	Price breaks above box high	False breakout above box = shakeout short
Turtle	Price breaks above 20-day high	Failed 20-day high breakout = shakeout
Brooks Pullback	Price pulls back to EMA	If pullback accelerates through EMA = potential shakeout
Minervini Breakout	Price breaks above pivot	Failed pivot = shakeout

The shakeout pipeline is a quality filter on all breakout pipelines: if a breakout immediately fails, the shakeout pipeline scores it as a potential reversal.

2. Key Level Detection

2.1 Level Types (by confidence)

Level Type	Description	Confidence
ATH / ATL	All-time high / low	Highest
52-Week High / Low	Within current year	High
Reaction High / Low	Prior swing highs / lows (≥10 bars apart)	Medium-High
Structure Break	Prior consolidation boundary	Medium
Trendline	Diagonal support / resistance	Medium
Gap Level	Prior unfilled gap	Medium

2.2 Level Detection Algorithm

For each candidate level within lookback window (e.g., 100 bars):
  1. Count how many times price approached the level (within 1 ATR)
  2. If approached ≥2 times → valid level
  3. Rate by: recency of last touch, number of tests, level type

2.3 Shakeout vs. True Breakdown Criteria

Shakeout (false break):

Price closes through level (violation + close beyond level)
Break depth < 2% of price
Reversal occurs within 3 bars
Volume on reversal bar is lighter than volume on break bar

True breakdown (invalid for shakeout — price continues in the direction of the break):

Price closes through level
No reversal within 5 bars
OR: Reversal bar has weaker conviction than the break bar

3. Scoring Architecture

3.1 Score Components

Shakeout Score = TrendFilter(30) + Setup(40) + SignalBar(15) + Volume(15) - RiskGates

TrendFilter (shared, 30 pts): Same as all pipelines. Biases toward:

Long shakeouts when EMA stack is rising (uptrend)
Short false breakouts when EMA stack is falling (downtrend)

Setup (pipeline-specific, 40 pts): Level quality + break quality + pre-break behavior

SignalBar (shared, 15 pts): Reversal bar quality — same function as all pipelines

Volume (shared, 15 pts): Volume pattern — same function as all pipelines

3.2 Setup Category Detail (40 pts)

A. Level Quality (0-15 pts)

Factor	Points	Description
Level type	0-6	ATH/ATL=6, 52w=5, Reaction=4, Structure=3, Trendline=2, Gap=2
Level tests	0-5	1 test=1, 2 tests=3, 3+ tests=5
Recency	0-4	Tested in last 20 bars=4, 20-50 bars=2, 50+ bars=1, never=0

B. Break Quality (0-15 pts)

Factor	Points	Description
Break conviction	0-5	Gap break=5, Strong bar=4, Moderate bar=2
Reversal speed	0-5	Next bar=5, 2 bars=4, 3 bars=2, 4+ bars=0
Break depth	0-3	<0.5%=3, 0.5-1%=2, 1-2%=1, >2%=0
Gap fill (for reversal)	0-2	If gap was filled on reversal=2

C. Pre-Break Behavior (0-10 pts)

Factor	Points	Description
Pre-break extension	0-4	Extended from EMA20=0, Near=2, Touching=4
Climax move before break	0-3	Strong climax bar just before break=3 (distribution pattern)
Volume dry-up before break	0-3	Volume contracting into the level=3 (exhaustion)

4. Entry, Stop, and Target

4.1 Entry

Long Shakeout: Buy when:

Price broke below support level
Within 3 bars, price closes above support level
Signal bar is bullish with close in upper 1/3

Short False Breakout: Sell when:

Price broke above resistance level
Within 3 bars, price closes below resistance level
Signal bar is bearish with close in lower 1/3

4.2 Stop

Variant	Stop Level
Long Shakeout	Below the shakeout bar's low (tight stop — 1-2 ATR)
Short False Breakout	Above the false breakout bar's high (1-2 ATR)

Stop distance is typically tighter than other pipelines because the entry is "late" (after reversal confirmation). Penalty based on ATR distance:

≤2 ATR: 0 penalty
2-4 ATR: -2 penalty
4 ATR: -5 penalty

4.3 Target

Conservative: Back to the broken level (now support/resistance)
Aggressive: Measured move = height of the breakout range, from the level

5. Sub-Factor Summary

Sub-Factor	Category	Range	Description
`levelType`	ShakeoutLevel	[0, 6]	ATH, 52w, reaction, etc.
`levelTests`	ShakeoutLevel	[0, 5]	Number of prior approaches
`levelRecency`	ShakeoutLevel	[0, 4]	When last tested
`breakConviction`	ShakeoutBreak	[0, 5]	Gap vs bar break
`reversalSpeed`	ShakeoutBreak	[0, 5]	Bars until reversal
`breakDepth`	ShakeoutBreak	[0, 3]	How far through level
`gapFill`	ShakeoutBreak	[0, 2]	Gap was filled
`preExtension`	ShakeoutContext	[0, 4]	Pre-break EMA distance
`preClimax`	ShakeoutContext	[0, 3]	Climax before break
`preVolumeDryup`	ShakeoutContext	[0, 3]	Volume contraction
`stopPenalty`	ShakeoutPenalties	[-5, 0]	ATR-based stop distance

6. Integration Points

6.1 New Files

File	Purpose
`convex/services/scoringShakeout.ts`	Shakeout setup scoring (level quality, break quality, context)
`convex/services/scoringShakeoutLevels.ts`	Key level detection (ATH/52w/reaction/trendline)
`convex/services/scoringShakeoutBreak.ts`	Break validation and reversal detection

6.2 Modified Files

File	Changes
`convex/types/stockBreakout.ts`	Add `'shakeout'` to `PipelineId` union + `entryMode` union
`convex/services/scoringThresholds.ts`	Add `SHAKEOUT` namespace with level/break/context thresholds
`convex/services/scoringBreakout.ts`	Add `buildShakeoutResult()` builder; call shakeout scoring in orchestrator
`shared/src/core/scoring/pipelineRegistry.ts`	Add `shakeout` pipeline definition
`convex/services/scoringRiskGates.ts`	Ensure SRG rule applies to shakeout

6.3 Pipeline Color

Suggested: amber-500 (#f59e0b) — orange/amber for "trap" concept
Or red-500 for short variant connotation

7. Relationship to Existing Pipelines

7.1 Interaction with Breakout Pipelines

The shakeout pipeline should be computed in addition to the existing pipelines, not replacing them. A stock can simultaneously have:

A clean Darvas box breakout → Darvas score high
AND a prior failed breakout (now reversed) → Shakeout score high

The max(pipelines) selection means: if a stock has both a clean breakout AND a shakeout signal, whichever scores higher wins.

7.2 Shakeout as a Filter

Alternatively, shakeout can be used as a negative signal: if any breakout pipeline is active AND the shakeout pipeline is also scoring high (failed break detected), reduce the breakout pipeline's score or flag it as suspicious.

This is a design choice: "trade the breakout OR the failed breakout" vs "avoid breakouts that are failing."

8. Edge Cases

8.1 No Reversal Within 5 Bars

→ Not a shakeout. If the break continues, the relevant pipeline is whatever the trend suggests (Turtle for long continuation, or no trade).

8.2 Very Deep Break (>2%)

→ Likely a true breakdown, not a shakeout. Score 0.

8.3 Multiple Level Breaks

→ Score the most recent valid level break. Older levels are lower priority.

8.4 Low Volume Break

→ If the break has very low volume (no one was watching this level), it's less likely to be an intentional stop hunt. Score lower on breakConviction.

8.5 Earnings Proximity

→ Apply earnings risk gate. Shakeouts near earnings are risky (catalyst may resolve against you).

9. Pseudocode for Key Logic

interface ShakeoutResult {
  score: number;
  direction: 'long' | 'short' | 'none';
  levelPrice: number;
  levelType: 'ath' | '52w' | 'reaction' | 'structure' | 'trendline';
  breakConviction: number;
  reversalSpeed: number;
  reversalBarQuality: number;
  stopLevel: number;
}

// 1. Find candidate levels (ATH, 52w, reaction highs/lows, trendlines)
// 2. For each level, check if price recently broke through it
// 3. If break detected, validate it's a shakeout (not true breakdown):
//    - Reversal within N bars
//    - Reversal bar conviction
//    - Volume pattern (contracting on reversal)
// 4. Score the shakeout quality
// 5. Return best result if score > threshold

function detectShakeout(ohlc: OHLC[], atr: number): ShakeoutResult[] {
  const levels = detectKeyLevels(ohlc);  // ATH, 52w, reaction highs/lows
  const results: ShakeoutResult[] = [];
  
  for (const level of levels) {
    const breakEvent = detectLevelBreak(ohlc, level);
    if (!breakEvent) continue;
    
    const reversalBar = findReversalBar(ohlc, breakEvent);
    if (!reversalBar) continue;
    
    const isShakeout = validateShakeout(breakEvent, reversalBar, atr);
    if (!isShakeout) continue;
    
    const score = scoreShakeoutSetup(breakEvent, reversalBar, atr);
    results.push({ score, ... });
  }
  
  return results.sort((a, b) => b.score - a.score);
}

10. Open Questions

Bidirectional vs. separate pipelines: Should shakeout be one pipeline (bidirectional) or two (shakeout_long, shakeout_short)? The current architecture uses separate long/short pipelines (e.g., pullback/short_pullback).
Interaction with other pipelines: Should a high shakeout score on a stock with a breakout active reduce the breakout score? Or should they compete independently via max()?
Level types to support: ATH/52w are easiest to detect reliably. Reaction highs/lows require swing detection. Trendlines require drawing. Should we support all three or start with just ATH/52w?
Lookback window for level detection: 100 bars? 200 bars? Should be long enough to catch multi-month levels but not so long that stale levels are scored.
Reversal bar threshold: What constitutes "immediate" reversal — 3 bars? 5 bars? This affects how many shakeouts are detected vs. missed.