# Evidence-led research brief workbook
No account or paid AI service is required. Work manually or use an approved tool.

## Fictional source packet — not real research
All organizations, documents, dates and measurements in this packet are invented for teaching.

### A — Drafting trial note
Author: fictional operations team. Version 1, 6 October 2026. Purpose: report drafting-stage effort.
Forty internal reports were processed manually and with AI assistance by a volunteer team. Across the same forty reports, mean drafting time was 30 minutes manually and 18 minutes with AI. These figures exclude review and corrections. No independent quality assessment was performed. Staff selected the reports; assignment was not randomized.

### B — Review-stage appendix
Author: fictional review lead. Version 2, 8 October 2026. Purpose: complete the stage-time record for the same forty reports and both workflows in A.
Mean review and correction time was 4 minutes manually and 12 minutes with AI assistance. All forty reports are included; there are no missing review records. These times exclude drafting. The appendix does not measure setup, source maintenance, licensing, or later downstream work. It provides no separate quality score.

### C — Trade newsletter summary
Author: fictional newsletter editor. 9 October 2026. Purpose: summarize the trial note.
The article says, "AI cuts report preparation time by 40%." It cites A as its only evidence and does not mention B. It reports no new experiment or measurements.

## 1. Question and evidence boundary
- Decision question and intended reader:
- Outcome, task, population and time period:
- Evidence needed and what would prevent a recommendation:
- Search terms, places searched and contrary evidence sought (for a real task):

## 2. Source register
| ID | Author/title | Version/date | Purpose | Passage/location | Inspected access | Original source or dependency | Limits |
|---|---|---|---|---|---|---|---|
| A | | | | | | | |
| B | | | | | | | |
| C | | | | | | | |

## 3. Claim comparison
| Proposed claim | Source IDs/passages | Metric and sample | Direct support / calculation / interpretation / unresolved | Correction or limitation |
|---|---|---|---|---|
| Three sources prove 40% faster total report preparation | | | | |

## 4. Arithmetic record
- Drafting comparison, with denominator:
- Manual measured-stage mean:
- AI-assisted measured-stage mean:
- Combined-stage difference and percentage:
- Why adding these means is justified in this packet:
- Excluded work and unresolved quality:
- What changes if B covers only 25 of the reports:

## 5. Brief — 120 words maximum
Finding with source IDs, limitation, and bounded next step:

## 6. Review
- Every number checked against the packet:
- Repeated reporting distinguished from independent evidence:
- Supported claim kept separate from generalization:
- Important missing evidence named beside the recommendation:
- One observation that would change the conclusion:

Compare your completed attempt with the worked answer in the guide.

Source guide: https://learn.aicompetence.org/knowledge/research-with-ai/