Job to be done
Create a comparable, auditable record of one AI-assisted workflow experiment.
Required inputs
- Defined job and owner
- Pre-AI baseline
- Data sensitivity boundary
- Fixed evaluation cases
- Success and stop criteria
Procedure
- Define the job, owner, affected people, data boundary, baseline, and stop criteria before running an AI system.
- Freeze representative evaluation cases and record what a satisfactory result requires.
- Record each approach, model or tool version, prompt or configuration fingerprint, cost, latency, and human intervention.
- Score every approach against the same cases and retain failures, overrides, and rejected results.
- Choose whether to stop, revise, pilot, or deploy; name the accountable human decision owner.
- Measure the real outcome after use rather than substituting an offline score for impact.
- Create a sanitized proof pack that excludes raw inputs, confidential artifacts, and unsupported causal claims.
Outputs
- Baseline record
- Experiment comparison
- Evaluation matrix
- Human decision log
- Follow-up outcome
- Sanitized proof pack
Publication gate
A result remains private or draft until every applicable gate passes.
- Method version is fixed
- Baseline and calculation are visible
- Negative results remain present
- Privacy review passes
- Claim scope matches the evidence
- A human approves publication
Limitations
- The protocol improves traceability; it does not prove causation.
- Results remain setting-specific until independent replications exist.
- High-stakes uses require qualified domain and legal review.
Sources and prior-art context
These sources inform the working synthesis. Their inclusion does not imply endorsement.
- NIST AI Risk Management Framework - National Institute of Standards and Technology
- Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile - National Institute of Standards and Technology
- Artificial Intelligence: An Accountability Framework for Federal Agencies and Other Entities - U.S. Government Accountability Office
Change history
- 2026-08-25 - Published the author-controlled working specification and explicit non-claims.