Add ASW-based faultline strength scoring to Group

[Copilot]

This PR adds a faultline-strength capability to Group parallel to the existing diversity-score APIs, implementing an ASW-based approach for multiple subgroup structures. It introduces both direct group scoring and proposed-group scoring from email lists.

• Public faultline scoring API

  • Added Group#calc_faultline_strength for persisted groups.
  • Added Group.calc_faultline_strength_for_group(users:) and Group.calc_faultline_strength_for_proposed_group(emails:) to mirror the existing diversity-score method family.

• ASW computation pipeline

  • Builds per-user demographic profiles across categorical and numeric attributes already used by group composition logic.
  • Computes weighted pairwise dissimilarities (including categorical rescaling), evaluates agglomerative clusterings, and returns the maximum average silhouette width over valid subgroup counts (k = 2..n-1).
  • Preserves member multiplicity (duplicate profiles remain distinct members) and returns 0.0 for undersized groups.

• Proposed-group query normalization

  • Normalizes incoming emails (trim, lowercase), removes empty tokens, deduplicates inputs, and performs case-insensitive matching.
  • Applies DISTINCT at query level so repeated email rows do not inflate subgroup/member counts.

• Focused model coverage

  • Added tests for: undersized-group baseline, polarized-vs-homogeneous behavior, and proposed-group email normalization/deduplication paths.

score = Group.calc_faultline_strength_for_proposed_group(
  emails: ' A@example.com, a@example.com,  , B@example.com '
)
# => case-normalized, deduplicated, ASW-based faultline strength

• Fixes Calculate fault-line strength using ASW #304

[mgmodell]

@copilot write Cucumber tests for this (like generate_diversity_scores.feature).

[Copilot]

@copilot write Cucumber tests for this (like generate_diversity_scores.feature).

Added Cucumber coverage in commit 10d32a5 with a new feature (features/generate_faultline_strength.feature) and step definitions (features/step_definitions/generate_faultline_strength_steps.rb) for group scoring and proposed-group email normalization behavior.

[mgmodell]

@copilot , if code is inspired by research, the calculations should be documented with references to the research that served as inspiration (in this case: Keivan et. al and Keivan et. al)

[Copilot]

@copilot , if code is inspired by research, the calculations should be documented with references to the research that served as inspiration (in this case: [Keivan et. al](https://journals.sagepub.com...

Added the research documentation in commit 7b34c9d with inline references in /home/runner/work/CoLab/CoLab/app/models/group.rb for the ASW faultline approach and the diversity-attribute weight rescaling method.

[mgmodell]

@copilot what is the range of possible fault line strengths? Is it bounded?

[Copilot]

@copilot what is the range of possible fault line strengths? Is it bounded?

Yes—it's bounded. In this implementation the returned faultline strength is in [0.0, 1.0] (rounded to 4 decimals). Silhouette values are at most 1, and the method also clamps negatives to 0 ([max_asw, 0.0].max), so you should never see values below 0 or above 1.