Strategy Insights & Signal Attribution

Overview

The analytics system is built on two core data structures that the backtest engine accumulates during simulation:

All analytics modules are computed in the browser after the simulation loop completes. There is no server round-trip — the entire analytics pipeline runs client-side from the collected EvalLogs and SignalEvents.

Condition Evaluation Logs

Every time a trade signal fires, the engine records the exact state of every condition that was evaluated during that tick, in order. This produces a ConditionRowfor each evaluation:

These rows are what appear in the Trade Detail Modal when you click a trade in the Positions tab. They also form the input to all condition-level analytics.

In live mode, the same condition rows are written to the TRADE_TRIGGER log entry, stored in detail_json.conditions. The Trade Detail Modal reads directly from this log.

Trigger Statistics

The Trigger Stats module answers: when did signals fire, and how often? It computes:

• Total signals: number of buy + sell signals across the simulation
• Signal frequency: average bars between signals
• Signals by hour of day: a heatmap of when signals tend to concentrate
• Signals by day of week: which days generated the most activity
• Signals by market regime: how many fired in trending vs. ranging conditions

This is useful for identifying strategies that are over-trading (too many signals), strategies that are concentrated in specific time windows, and potential data-snooping if signals cluster suspiciously in one period.

Signal Attribution

Signal Attribution answers: which conditions actually caused this signal to fire?It processes the ConditionRow[] arrays from all EvalLogs and builds a table of every unique condition label that appears in the dataset, along with:

• How many times the condition appeared in a signal's evaluation log
• How many times the condition was true when a signal fired
• The condition's truth rate at signal time (see below)
• The percentage of all signals this condition was present in

This makes it possible to see, for example, that RSI < 30 was present in 87% of buy signals but was only true 34% of the time when evaluated — meaning it was a genuine filter, not just always true noise.

Condition Truth Rate

The truth rate of a condition is the fraction of times it evaluated totrue across all its appearances in the condition log:

truth_rate = appearances_where_result_is_true / total_appearances

Truth rate tells you how selective a condition is. Interpreting it:

A useful condition typically has a truth rate between 10% and 50% — selective enough to filter noise but not so rare that it almost never contributes.

Contributor Percentages

The contributor percentage for a condition answers: what fraction of all executed trades had this condition present in the trigger evaluation?

contributor_pct = trades_where_condition_appeared / total_trades × 100

A condition with a 100% contributor percentage was evaluated in every single trade. A condition with a 40% contributor percentage was only evaluated in less than half of trades — likely because it lives inside an else-branch or an AND group that short-circuits.

Combined with truth rate, contributor percentage helps identify which conditions are actually driving your strategy's behavior vs. which are passive passengers in the logic.

Condition Removal Impact

The Condition Removal Impact module asks a counterfactual question for each condition: if this condition were always true (removed from the filter), how would the strategy's total return change?

For each condition, the module re-simulates the trade sequence with that condition removed (i.e., treated as always true), computes the resulting total return, and reports the delta versus the actual return. A condition is useful if removing it makes performance worse — i.e., it is filtering out bad trades.

Strategy Autopsy

Strategy Autopsy is the most comprehensive module. It performs a post-mortem on the entire simulation, identifying:

• Best and worst trades: the top and bottom trades by PnL percentage
• Best and worst periods: rolling windows with the highest and lowest strategy performance
• Win streak and loss streak: longest consecutive winning and losing sequences
• Exit analysis: breakdown of how positions were closed (TP hit, SL hit, time exit, manual)
• Condition correlation with outcomes: which conditions appeared disproportionately in winning vs. losing trades
• Drawdown anatomy: the maximum drawdown period — when it started, when it recovered, and what was happening to conditions during it

The autopsy is displayed as a scrollable report at the bottom of the backtest analytics panel. It is recomputed after every backtest run.

Signal Quality — MFE / MAE

MFE (Maximum Favorable Excursion) is the highest unrealized profit that occurred during a trade — i.e., how far into profit the position went before it closed.MAE (Maximum Adverse Excursion) is the worst unrealized loss during a trade — i.e., how deep the position went negative before closing.

The Signal Quality module classifies each trade as:

A healthy signal quality distribution shows more "Good" signals than "Bad" ones. A high proportion of "Bad" signals suggests the strategy's entries are correct but exits are letting losses run too far.

Regime Analysis

The Regime Analysis module tags each trade with the market regime that was active at entry time, then breaks down win rate and average return by regime. Market regime is determined by two indicators computed independently of strategy code:

This produces a grid of results: how did the strategy perform in uptrend + high volatility? In downtrend + low volatility? In ranging + high volatility? These breakdowns reveal whether the strategy has a structural edge in specific market conditions or relies on a single regime that may not persist.

Strategy DNA

Strategy DNA is a compact visual fingerprint of the strategy's behavior across the entire backtest. It encodes:

• Trade frequency over time (how active the strategy was in each period)
• Win/loss coloring by time period
• Drawdown depth over time
• Condition engagement rate (how often conditions were actually evaluated vs. short-circuited)

The DNA visualization makes it immediately obvious when a strategy's performance is concentrated in one short window (a potential data overfitting signal) vs. distributed evenly over time (a more robust strategy).

How to Use These Insights

The analytics system is designed to help you ask better questions, not to automatically optimize strategies. Here is a practical workflow:

1. Start with Strategy Behavior — look at total return, win rate, drawdown, and Sharpe ratio to get a baseline view.
2. Check Signal Quality — if you have many "Bad" signals, your exits may be the problem rather than your entries. Consider tightening the stop loss.
3. Review Condition Truth Rates — conditions with 90%+ truth rate are not filtering. Conditions with 0% truth rate are dead code. Both should be removed or reconsidered.
4. Use Condition Removal Impact — find conditions that have negligible or negative impact, and consider simplifying the strategy by removing them.
5. Check Regime Analysis — if the strategy only performs in one regime (e.g. bull market), consider whether that regime is likely to persist.
6. Review Strategy DNA — if most profits came from one narrow window, be skeptical of the overall performance number.
7. Run Autopsy on losing periods — look at what conditions were doing during the worst drawdown period. This often reveals structural weaknesses.