Improve algorithm weighting and score calibration

[willdaly]

Summary

The whiteboard calls out better algorithm weighting. The project spec also lists live experiments on scoring and recommendation calibration as a planned extension.

Scope

• review how current matching signals are weighted
• make the scoring model easier to tune and reason about
• improve how strengths, concerns, and recommendations are derived from inputs

Acceptance criteria

• scoring inputs and weights are explicit
• the team can tune weights without rewriting the whole algorithm
• compatibility scores and recommendations feel more consistent in test scenarios