Work sample / AI tooling
Evidence before output
A simplified model for using AI to understand complex source material while keeping claims traceable and decisions in human hands.
Recreated from a general learning and evidence framework. This is a process example, not an internal banking document or an automated decision system.
Artifact / Source-to-decision model
Five checks before a claim becomes guidance
01 / ScopeDefine the jobWhat decision or task must the answer support?
02 / SourceAssign a roleIs the material primary evidence, context, opinion, or only a lead?
03 / AssessCheck fitnessReview authority, evidence, currency, independence, and directness for this claim.
04 / ChallengeExpose uncertaintyFind contradictions, missing information, and interpretations presented too strongly.
05 / ApplyReview the decisionTrace consequential claims to sources; have a person confirm the intended use.
| Output label | What it means | Next action |
|---|---|---|
| Supported | The source directly addresses the claim. | Keep the citation and context with the output. |
| Interpretation | A plausible synthesis, not an explicit source fact. | Label the inference and ask for review. |
| Unresolved | Sources conflict or do not answer the question. | Do not turn it into a delivery instruction; identify the decision owner. |
The same pattern can support product documentation, learning, and story analysis. The tool helps organize evidence; it does not determine policy or replace domain review.