Site policy
Evidence State
AI Experiment Log (AIEL) records the evidential standing of individual information obtained in an Experiment as an Evidence State. It distinguishes direct observation, confirmation with additional evidence, inference, pre-verification hypothesis, and explanations not supported under the tested conditions.
Evidence State describes the evidential standing of individual recorded information. It applies to a different target from Artifact Availability, which describes Artifact retention and publication, and Experiment Status, which describes the lifecycle state of the Experiment as a whole.
On public Experiment pages, Evidence State is used primarily in the Evidence summary and in explanatory prose to distinguish observation, verification, inference, hypothesis, and disproven explanations. Individual public Experiment pages link here instead of repeating the same definitions.
In the public Evidence summary, the items listed as Recorded content are those material to understanding the Experiment’s principal Results or Conclusion, together with their Evidence State. Setup actions and operational history that belong in the Experiment Log are not duplicated here unless they are themselves a material result. Design branches or future test items that have not been observed at runtime are not written as affirmative results in this table; they are classified according to their actual role, such as a genuine unresolved Evidence question, validation not performed, or an out-of-scope item. The first two table columns are Recorded content and Evidence State.
OBSERVED
A matter directly observed through an experiment, measurement, log, or operation result.
VERIFIED
A matter confirmed with additional evidence beyond the observation itself, such as a manufacturer response, specification, or independent verification.
INFERRED
An interpretation judged to be the most consistent with the observations and known facts. It is distinct from a directly confirmed fact.
HYPOTHESIS
A plausible explanation proposed before verification and treated as something to test during the investigation.
DISPROVEN
An explanation that investigation or testing did not support, at least under the conditions that were examined.
Relationship to Unresolved questions
A material question that the retained Evidence cannot answer, and that matters to interpreting the Experiment’s Results or Conclusion, is written as an Unresolved question rather than by introducing another Evidence State. Validation that was simply not performed, management/traceability work, and items outside the Experiment scope are treated separately rather than being labeled unresolved.
Evidence States do not replace explanatory prose. Each public Experiment page should first explain what was observed, what was confirmed, and what can only be inferred, then use the Evidence State as supporting classification.
Evidence State describes only the evidential standing of individual recorded information. See Artifact Availability for Artifact retention and publication and Experiment Status for the lifecycle state of the Experiment as a whole.