Problem

When a strategy requests a strike that doesn't exist in the parquet data (e.g. far OTM wings with 2000+ point hedges), the engine queries DuckDB on every bar because empty results are never cached.

In _lazy_fetch() (context.py:414):

if data:
    self._cache[instrument_key] = data  # only caches non-empty results

A missing strike returns {}, which is falsy, so it's never written to cache. The next bar sees a cache miss and queries DuckDB again.

Impact

For a 60-day backtest with a 365-tick clock and 2 missing strikes:

• ~43,800 wasted DuckDB queries (60 × 365 × 2) that all return empty
• 2400-point hedge: 68.7s runtime vs 10.8s after fix

Fix

Cache the empty result so subsequent bars skip the query entirely:

# Before:
if data:
    self._cache[instrument_key] = data

# After:
self._cache[instrument_key] = data  # cache {} for missing strikes too

Subsequent bars find the empty dict in cache, hit the if not inst_data: return None, -1 guard, and return immediately with no database call.

Benchmarks

All results verified — PnL, trades, win rate, max drawdown are identical before and after.

Scope

• src/fastbt/backtest/context.py — _lazy_fetch() method (1 line change)
• No changes to DataSource ABC, engine, or strategy interfaces
• 312 backtest tests pass

Reported by: @darkseid during iron condor backtest with wide hedge wings.