PRISMWORKS | AI WORKFLOW EVALUATION CASE
Companion: https://prismworks.ai/resources/evaluating-ai-workflows/
Version: 2026-11-02 | Copy one record per case. No submission required.
Use only authorized data and store records under your information policy.

CASE DEFINITION
Case ID / title / owner:
Purpose and consequence of failure:
Case type: normal / incomplete / conflicting / restricted / interrupted /
adversarial / other.
Inputs, source references, and relevant versions:
Requester identity and permitted scope:
Starting application state and required dependencies:
Expected completed outcome:
Expected escalation or abstention:
Disallowed actions:

ACCEPTANCE
Observable criteria, one per line:
For each criterion: authoritative observation, grading method, and reviewer.
Hard release blockers:
Acceptable quality, review effort, and latency for this case:
Ambiguous cases and who resolves them:

EXECUTION RECORD
Workflow and configuration version:
Model identifier, prompts, retrieval settings, and tool versions:
Evaluation harness, rubric, and grader versions:
Trial identifier and time:
Observed tool actions and final system state:
Source evidence references:
Outcome: accepted / rejected / unresolved
Criterion-by-criterion results:
Staff review and correction effort:
Latency and operating cost:
Unexpected behavior, missing evidence, or dependency failures:

REVIEW AND RELEASE
Human reviewer and decision:
Defect or follow-up action, owner, and review date:
Repeated-trial results and variation:
Held-out or development case:
Related regression cases:
Release decision owner and evidence reference:
Retention and access rules for this record:

A plausible narrative is not evidence that a system action completed.
An unresolved observation is not a confirmed failure or a successful result.
Select the case set and release thresholds for the actual consequence;
this template does not define a universal readiness score.