# Model Documentation Evaluation Template

# Metadata
model_name: "Gemma 4"
model_link: "https://huggingface.co/google/gemma-4-31B"
organization: "Google"
org_link: "https://deepmind.google"
evaluation_date: "2026-08-04"
public_summary_link: "https://storage.googleapis.com/transparencyreport/report-downloads/pdf-report-nn_2026-7-31_2026-7-31_en_v1.pdf" 
public_summary_date: "2026-07-31"
public_summary_location: "https://transparencyreport.google.com/"
model_publication_date: "2026-04-02"
category: "New model" # New model, fine-tuned model
archive_file_name: "Gemma_4_2028_08_03.pdf"

# Requirement assessments
# Each requirement should be assessed once with a score from 0-10
# Leave blank for N/A, and assign a max score of 0
S1: # Document-level requirements
  D1: # Clarity
    score: 9
    max_score: 11
    notes: ""

  D2: # Completeness
    score: 4
    max_score: 4
    notes: ""

  D3: # Consistency
    score: 7
    max_score: 7
    notes: ""

  D4: # Correctness
    score: 4
    max_score: 5
    notes: ""

  D5: # Accessibility
    score: 6.5
    max_score: 10
    notes: ""

  D6: # Comprehension
    score: 6
    max_score: 9
    notes: ""

S2: # General information
  D1: # Clarity
    score: 29
    max_score: 40
    notes: ""

  D2: # Completeness
    score: 44
    max_score: 72
    notes: ""

  D3: # Consistency
    score: 8
    max_score: 8
    notes: ""

  D4: # Correctness
    score: 22
    max_score: 22
    notes: ""

  D5: # Accessibility
    score: 0
    max_score: 0
    notes: ""

  D6: # Comprehension
    score: 6
    max_score: 11
    notes: ""

S3: # Public datasets
  D1: # Clarity
    score: 0
    max_score: 76
    notes: ""

  D2: # Completeness
    score: 4
    max_score: 76
    notes: ""

  D3: # Consistency
    score: 3
    max_score: 3
    notes: ""

  D4: # Correctness
    score: 4
    max_score: 4
    notes: ""

  D5: # Accessibility
    score: 0
    max_score: 20
    notes: ""

  D6: # Comprehension
    score: 0
    max_score: 0
    notes: ""

S4: # Private datasets
  D1: # Clarity
    score: 32
    max_score: 32
    notes: ""

  D2: # Completeness
    score: 40
    max_score: 88
    notes: ""

  D3: # Consistency
    score: 3
    max_score: 3
    notes: ""

  D4: # Correctness
    score: 24
    max_score: 24
    notes: ""

  D5: # Accessibility
    score: 0
    max_score: 0
    notes: ""

  D6: # Comprehension
    score: 16
    max_score: 16
    notes: ""

S5: # Scraped/crawled data
  D1: # Clarity
    score: 72
    max_score: 129
    notes: ""

  D2: # Completeness
    score: 92
    max_score: 158
    notes: ""

  D3: # Consistency
    score: 3
    max_score: 3
    notes: ""

  D4: # Correctness
    score: 61
    max_score: 93
    notes: ""

  D5: # Accessibility
    score: 0
    max_score: 25
    notes: ""

  D6: # Comprehension
    score: 33
    max_score: 58
    notes: ""

S6: # User data
  D1: # Clarity
    score: 0
    max_score: 15
    notes: ""

  D2: # Completeness
    score: 11
    max_score: 26
    notes: ""

  D3: # Consistency
    score: 3
    max_score: 3
    notes: ""

  D4: # Correctness
    score: 11
    max_score: 11
    notes: ""

  D5: # Accessibility
    score: 15
    max_score: 15
    notes: ""

  D6: # Comprehension
    score: 0
    max_score: 15
    notes: ""

S7: # Synthetic & other
  D1: # Clarity
    score: 41
    max_score: 65
    notes: ""

  D2: # Completeness
    score: 47
    max_score: 47
    notes: ""

  D3: # Consistency
    score: 6
    max_score: 6
    notes: ""

  D4: # Correctness
    score: 35
    max_score: 47
    notes: ""

  D5: # Accessibility
    score: 17
    max_score: 17
    notes: ""

  D6: # Comprehension
    score: 0
    max_score: 12
    notes: ""

S8: # Data processing
  D1: # Clarity
    score: 18
    max_score: 18
    notes: ""

  D2: # Completeness
    score: 10.5
    max_score: 18
    notes: ""

  D3: # Consistency
    score: 3
    max_score: 3
    notes: ""

  D4: # Correctness
    score: 36
    max_score: 36
    notes: ""

  D5: # Accessibility
    score: 15
    max_score: 15
    notes: ""

  D6: # Comprehension
    score: 42
    max_score: 69
    notes: ""


general_notes: "The public summary for Gemma 4 can be found <a href=\"https://storage.googleapis.com/transparencyreport/report-downloads/pdf-report-nn_2026-7-31_2026-7-31_en_v1.pdf\" target=\"_blank\"> on a GoogleAPIs storage</a>, linked from <a href=\"https://transparencyreport.google.com/\" target=\"_blank\"> Google's Transparency Report hub</a>. The summary matches the structure of the template, and differs only minimally from the training data summary of the Gemini 3 Pro family of models. Points are lost primarily for providing high-level information where more detailed information is required, for instance in the field where a list of large publicly available data sources should be provided where the provider answers instead that \"Our publicly available datasets include data across various sectors such as educational, government, legal, and research sectors comprising a wide variety of media types and languages\". We assessed the summary's score to be <span class=\"openness partial\">59.54%</span> with Grade <strong>C</strong> for <strong><i>transparency</i></strong>, and <span class=\"openness partial\">53.6%</span> with Grade <strong>C</strong> for <strong><i>usefulness</i></strong>."