Replace flat recency penalty with cycle detection. For each dish with
≥2 appearances in history, compute the median interval between uses and
score by how well weeksSinceLast aligns with that period:
ratio < 0.5 → 0.10 (too soon)
ratio ~1.0 → 3.00 (right on schedule, boost)
ratio > 2.5 → 1.00 (pattern may have changed, neutral)
Expand history window from 4 to 12 weeks so monthly cycles (period=4)
have enough data points to be detected reliably.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
alexandrev-tibco
e13771fa06