gemini31propreview · Model identity and integrity report

Family

google

V3 sub-model reference

Available ✓

Checks

知識截止 · 能力探針 · 拒絕模板 · 拒絕溫度參數

Family-fingerprint coverage

17 weighted rules · 5 feature categories

Concrete reference answers, refusal templates, and individual signal weights are not published because disclosure would help spoofing endpoints evade detection.

235 runs · 115 hosts · 167 anomalies

Family-fingerprint consistency

  google         174  ███████████████████░░░░░░   74% (expected)
  anthropic       29  ███░░░░░░░░░░░░░░░░░░░░░░   12% (suspicious)
  unknown         26  ███░░░░░░░░░░░░░░░░░░░░░░   11% (suspicious)
  openai           6  █░░░░░░░░░░░░░░░░░░░░░░░░    3% (suspicious)

Observed V3 sub-model mapping

  → abstain                                113 (48%) — low confidence
  → google/gemini-3.1-pro-preview           64 (27%) ⚠ mismatch sim=0.86
  → unknown                                 35 (15%) ⚠ mismatch
  → google/gemini-3.1-flash-lite-preview    14 ( 6%) ⚠ mismatch sim=0.82
  → google/gemini-3.6-flash                  3 ( 1%) ⚠ mismatch sim=0.76
  → google/gemini-2.5-pro                    2 ( 1%) ⚠ mismatch sim=0.72
  → openai/gpt-5.4                           2 ( 1%) ⚠ mismatch sim=1.00
  → google/gemini-3.5-flash-lite             1 ( 0%) ⚠ mismatch sim=0.68
  → anthropic/claude-sonnet-4.6              1 ( 0%) ⚠ mismatch sim=0.70

Capability score distribution (auxiliary)

  0-49     ███████████████░░░░░░░░░░  144
  50-69    ██░░░░░░░░░░░░░░░░░░░░░░░   21
  70-89    ███████░░░░░░░░░░░░░░░░░░   67
  90-100   ░░░░░░░░░░░░░░░░░░░░░░░░░    3

BazaarLink Probe combines a Bayesian V2 family classifier across nine fingerprint categories with V3 sub-model references to verify endpoints claiming gemini31propreview. Low family consistency, a different V3 match, or divergence from selfClaim is flagged as anomalous. BazaarLink Probe home.