ae0j's picture
Publish benchmark dataset v1.0.1
f7fc448 verified
|
Raw History Blame Contribute Delete
2.34 kB

Codebook

The primary key is product_id + criterion_id. There is exactly one row for each product–criterion pair.

Field Type Meaning
product_id string Stable product slug.
product string Product display name.
version string Application version tested for this assessment.
platform string Platform used for this assessment.
plan string Recorded plan or access state.
test_date string Month, date, or date range of the published test.
criterion_id string Stable criterion slug.
scenario string Human-readable criterion label.
expected_result string Behavior used to evaluate the criterion.
observed_result string Evidence-bounded observation.
coverage enum tested, feature_sought, or not_tested.
outcome_code enum pass, mixed, failed, not_found, or not_tested.
outcome string Human-readable outcome label.
evidence_url URL/null Primary published evidence for a tested result.
additional_evidence_url URL/null Optional second evidence item.
details_url URL/null Scoped review section used for a Not found result.
repeat_count integer/null Exact comparable count when systematically recorded; blank here means not normalized.
limitations string Assessment-specific scope and interpretation constraints.

Outcome definitions

  • pass — Passed: The expected behavior worked in the published test.
  • mixed — Mixed: The behavior worked only partly, varied across runs, or passed with an important limitation.
  • failed — Failed in our test: The expected behavior did not work in the published test.
  • not_found — Not found: We did not find the feature in the tested plan or flow. This does not mean it is missing from other plans or versions.
  • not_tested — Not tested: We did not test this behavior strongly enough to publish a result.

Coverage versus outcome

  • tested: a behavioral result was published; outcome is pass, mixed, or failed.
  • feature_sought: the feature was sought in the tested plan/flow but not found; outcome is not_found.
  • not_tested: no normalized result is claimed; outcome is not_tested.

Do not convert not_found or not_tested to failed, and do not treat a blank repeat_count as one run.