Practical guide
AI management report verification checklist
A management report can look finished while still hiding broken lineage, misleading comparisons, or actions nobody owns. Use this checklist after the report has been drafted and before it is released. It is a verification artifact, not a report-writing template: the reviewer follows each item back to evidence, records exceptions, and decides whether the report is ready, ready with a visible.
A management report can look finished while still hiding broken lineage, misleading comparisons, or actions nobody owns. Use this checklist after the report has been drafted and before it is released. It is a verification artifact, not a report-writing template: the reviewer follows each item back to evidence, records exceptions, and decides whether the report is ready, ready with a visible limitation, or held for correction.
Use CLEAR as a release gate
Run the five sections in order: Context, Lineage, Exceptions, Actions, Release. Do not mark an item complete because the wording sounds plausible. Record the file, query, calculation, approval, or other evidence that lets a second reviewer repeat the check.
### C — Context and decision
Evidence to retain: the approved reporting brief, metric dictionary, reporting calendar, and the decision or meeting for which the report was prepared.
### L — Lineage and calculations
The U.S. Government Accountability Office’s data-reliability guide treats assessment as a decision-dependent process: the questions asked and the work performed should reflect how the data will be used. That is a useful discipline here. A report does not need perfect data for every purpose, but it needs evidence that the data are fit for this report’s stated decision. See GAO-20-283G, Assessing Data Reliability.
### E — Exceptions and uncertainty
### A — Actions and accountability
The Local Government Association’s performance, finance and risk reporting guidance emphasizes joining performance, finance, and risk information so leaders can see the relationship between delivery, resources, and uncertainty. Even outside local government, that is a useful test: an action without its resource or risk context is hard to govern.
### R — Release record
The release record should contain the report identifier, version, reporting period, review date, reviewer, disposition, open exceptions, and links to the retained evidence. GAO’s evidence-based policymaking practices also provide a broader reminder that evidence becomes useful when it is deliberately acquired, assessed, and applied to decisions—not merely attached to a document.
- [ ] The reporting period, audience, and decision the report supports are stated.
- [ ] The scope says which teams, products, regions, or cases are included and excluded.
- [ ] Every target or comparison uses a named baseline and compatible period.
- [ ] The report distinguishes observed results from forecasts, estimates, and management interpretation.
- [ ] Terms that could change the decision—such as “active”, “resolved”, or “at risk”—have an operational definition.
- [ ] Each decision-relevant number points to a source system, file, table, or query and an extraction time.
- [ ] Filters, joins, exclusions, currency conversions, and manual adjustments are recorded.
- [ ] Totals reconcile to their components, and percentages have the correct denominator.
- [ ] The reviewer has independently recalculated at least the headline measures and one sampled detail.
- [ ] Any copied chart still matches the current source and labels its units, period, and population.
- [ ] Source changes since the previous report are described rather than silently blended into a trend.
- [ ] Missing, late, duplicated, or disputed inputs are quantified where practical.
- [ ] A material anomaly is investigated or shown as unresolved; it is not smoothed away.
- [ ] Contradictory sources are presented with the chosen treatment and rationale.
- [ ] Limitations appear next to the affected conclusion, not only in a distant appendix.
- [ ] Claims using words such as “caused”, “improved”, or “on track” are supported by the design and evidence available.
- [ ] Residual uncertainty is translated into what the reader should and should not decide.
- [ ] Every recommended action names an owner, due date, and completion evidence.
- [ ] The action follows from the cited finding; it is not a generic recommendation.
- [ ] Dependencies and escalation triggers are explicit.
- [ ] The report separates a proposed action from an already approved commitment.
- [ ] Sensitive or access-restricted evidence is referenced safely rather than copied into an unrestricted report.
- [ ] A named preparer and independent reviewer have signed and dated the release record.
- [ ] The release version is immutable or otherwise distinguishable from working drafts.
- [ ] Corrections have an owner and a route for notifying prior recipients.
- [ ] The evidence pack has a retention location and access owner.
- [ ] The next review date is set for recurring reports.
- [ ] The release decision is recorded as release, release with limitation, or hold.
Sources: [1] [2] [3]
Worked test: two representative reports
### Example A: monthly service report
A draft says response time improved from 14 to 9 hours and recommends extending the new triage process. The reviewer traces both numbers. The earlier figure covers all tickets; the later figure excludes weekends and unresolved tickets. CLEAR produces this result:
| Section | Finding | Disposition | |---|---|---| | Context | “Response time” is undefined | Hold | | Lineage | Populations and time windows differ | Hold | | Exceptions | 18 unresolved tickets are excluded | Show limitation and recalculate | | Actions | Process owner named, but success evidence is absent | Add a four-week comparison using one definition | | Release | Headline conclusion changes after recalculation | Issue corrected draft |
The report is not rejected because the process failed. It is held because the comparison cannot yet support the claimed improvement. The corrected version uses one population, reports the unresolved-ticket count, and describes the result as an observed association rather than a causal finding.
### Example B: AI capability programme report
A second draft reports 82% completion, notes strong participant feedback, and asks the steering group to expand the programme. The source register confirms 41 of 50 invited people submitted the required assessment, but feedback came from only 12 volunteers. CLEAR permits the completion figure, labels the feedback as a small self-selected response, and changes the expansion request into a bounded next step: review assessment evidence quality and delivery cost before deciding whether to add another cohort.
The artifact retained with the report is compact:
text Report: AI capability programme — October Decision: Continue, change, or stop the next cohort Verified: 41/50 required submissions; denominator confirmed Exception: Feedback n=12, voluntary and not representative Action: Programme owner to review evidence quality and cost by 18 Nov Disposition: Release with visible limitation Reviewer/date: M. Chen / 9 Nov
Decide what happens after the checklist
Release when all material items have reproducible evidence and no open exception could change the decision. Release with a limitation when the uncertainty is visible, bounded, and does not undermine the intended use. Hold when lineage is broken, comparisons are incompatible, a material contradiction is unresolved, or the action depends on an unsupported claim.
Two Prune can supply assessment evidence about how participants performed in realistic AI-enabled work simulations. It does not certify the accuracy of a customer’s management report, determine legal compliance, or make employment decisions. The report owner remains responsible for source access, calculation review, approval, and the final management decision.
Sources
- 1.GAO-20-283G, Assessing Data Reliability · U.S. Government Accountability Office
- 2.Local Government Association’s performance, finance and risk reporting guidance · www.local.gov.uk
- 3.evidence-based policymaking practices · U.S. Government Accountability Office