Feature: Decomposition clustering — uncovered code paths
  As a developer
  I want to exercise the cluster_by_language method and the
  estimate_tokens_for_path fallback in cluster_by_size
  So that lines 122-131 and 160 of decomposition_clustering.py are covered

  # ---- cluster_by_language (lines 122-131) --------------------------------

  Scenario: cluster_by_language groups files by extension
    Given a list of mixed-language file paths
    When I cluster by language with max_per_cluster 10
    Then the language clusters should separate Python and TypeScript files
    And each language cluster should contain only one extension

  Scenario: cluster_by_language splits oversized language buckets
    Given a list of 8 Python file paths
    When I cluster by language with max_per_cluster 3
    Then the language clusters should contain at least 3 clusters
    And no language cluster should exceed 3 files

  Scenario: cluster_by_language with single language returns one cluster
    Given a list of 4 Python file paths
    When I cluster by language with max_per_cluster 10
    Then exactly 1 language cluster should be returned

  Scenario: cluster_by_language with empty list returns no clusters
    Given an empty list of file paths
    When I cluster by language with max_per_cluster 5
    Then no language clusters should be returned

  Scenario: cluster_by_language sorts files within each cluster
    Given an unsorted list of mixed-language file paths
    When I cluster by language with max_per_cluster 20
    Then files within each language cluster should be in sorted order

  Scenario: cluster_by_language orders clusters by extension
    Given a list of files with js py and ts extensions
    When I cluster by language with max_per_cluster 20
    Then the language clusters should be ordered by extension alphabetically

  # ---- cluster_by_size fallback to estimate_tokens_for_path (line 160) ----

  Scenario: cluster_by_size falls back to estimate_tokens_for_path without token_map
    Given a set of temporary files with known byte sizes
    When I cluster by size with max_tokens 50 and no token_map
    Then the size clusters should respect the estimated token limits

  Scenario: cluster_by_size falls back for files missing from partial token_map
    Given a set of temporary files with known byte sizes
    And a partial token_map covering only some files
    When I cluster by size with max_tokens 50 and the partial token_map
    Then the size clusters should include all files