ProfileClickWeight is 0.0: profile clicks add nothing to For You ranking

Problem Phoenix predicts the probability that a viewer clicks the author's profile (README "Scoring and Ranking", Clicks: post · profile · ...), but RankingScorer multiplies that prediction by ProfileClickWeight = 0.0, so it adds nothing to the final score.

The default ValueModelMode is "weighted", so this is the live scoring path. The value has been 0.0 in every commit touching param.rs since 2026-08-17 (15 commits checked).

• ProfileClickWeight = 0.0: https://github.com/xai-org/x-algorithm/blob/6bb4594/home-mixer/params/param.rs#L344
• Applied here: https://github.com/xai-org/x-algorithm/blob/6bb4594/home-mixer/scorers/ranking_scorer.rs#L490
• ValueModelMode default "weighted": https://github.com/xai-org/x-algorithm/blob/6bb4594/home-mixer/params/param.rs#L483
• For comparison, FollowAuthorWeight = 4.0: https://github.com/xai-org/x-algorithm/blob/6bb4594/home-mixer/params/param.rs#L378

Why it matters

• A profile click is the step between seeing a post and following its author. The follow is weighted 4.0; the step that leads to it is weighted 0.
• An impression says someone saw a post. A profile click says they stopped and wanted to know more about the person behind it.
• Posts that make strangers curious about the author are the ones most worth showing to more non-followers.

Proposal

1. Evaluate a small positive ProfileClickWeight in an experiment (somewhere between ClickWeight = 0.4 and FollowAuthorWeight = 4.0).
2. Alternatively, apply it to out-of-network candidates only, where a profile click is a direct discovery signal (related: #200).

Open questions

• Was 0.0 chosen because the profile-click head is poorly calibrated, or to avoid rewarding curiosity-bait? If the latter, pairing it with dwell on the profile could guard against that.