strategy-pivot-designer

---
name: strategy-pivot-designer
description: Use when backtest score iteration has plateaued, a strategy shows overfitting or excessive drawdown, or the user wants structurally different strategy designs instead of more parameter tweaking.
category: Strategy R&D
version: 1.0.0
tools: []
---

# Strategy Pivot Designer

Detect when a strategy's backtest iteration loop has stalled, and propose structurally different strategy architectures rather than further parameter tweaking. This is the feedback-loop stage of the strategy pipeline (hint-extraction → concept-synthesis → strategy-design → candidate-generation): it breaks out of local optima by redesigning the strategy's skeleton, not by nudging its parameters.

## When to Use

- Backtest scores have plateaued across multiple refinement iterations
- A strategy shows overfitting (strong in-sample results, weak out-of-sample robustness)
- Transaction costs are defeating a thin edge
- Tail risk or drawdown exceeds acceptable thresholds
- The user wants to explore fundamentally different architectures for the same market hypothesis, not just re-tune the current one

## Methodology

**Step 1 — Accumulate iteration history.** Track each backtest evaluation (score, key metrics, what changed) across iterations of the same strategy so a trend is visible.

**Step 2 — Detect stagnation.** Diagnose the iteration history against four triggers:
- Plateau: scores stopped improving despite parameter changes
- Overfitting: high in-sample performance, low out-of-sample/robustness performance
- Cost defeat: the edge is real but transaction costs erode it below viability
- Tail risk: drawdown or tail-loss exceeds the strategy's risk budget

**Step 3 — Generate pivot proposals**, once stagnation is confirmed, using three techniques:
- *Assumption inversion* — take a core assumption of the current design and flip it (e.g. trend-following → mean-reversion) to see if the opposite framing captures the same underlying edge better
- *Archetype switch* — swap the strategy's structural archetype entirely (e.g. breakout → carry, discretionary trigger → systematic filter) while keeping the same market hypothesis
- *Objective reframe* — redefine what the strategy is optimizing for (e.g. Sharpe → max drawdown, win rate → expectancy) to see if a different objective produces a materially different, better-behaved design

**Step 4 — Rank proposals** by a combination of quality potential and novelty (how different the pivot is from the stalled design) so the most promising, most distinct alternatives surface first.

**Step 5 — Route the result.** Proposals that are ready to formalize go forward as export-ready draft specs for the candidate-generation stage; proposals still exploratory stay flagged research-only pending manual strategy design. Either way, feed the selected pivot back into the next backtest iteration cycle.

## Output

For each pivot: which trigger caused it, which technique produced it, a plain-language description of the new architecture, and a quality/novelty score. Present a ranked list in chat, plus a short stagnation-diagnosis summary explaining which trigger(s) fired and why.