# Eight-claim citation review: complete fictional source packet

Packet ID: OM-CITATIONS-071 · Version: 2026-09-04-v1 · Review time: 2026-09-04.

All passages, organizations, product specifications, studies, and results below are invented teaching material. IDs are local locators, not published references or DOIs. No customer, model benchmark, or OpenMax feature was tested. English is the authored source text; Chinese and Japanese packets translate this same exercise. Quoted English in the negation exercise is intentionally retained.

## Source inventory and complete relevant passages

### S1 · Participation note · v1 · section Results

We invited 120 organizations. Thirty organizations responded. Of those 30 respondents, 18 reported using the feature. We did not establish feature use among the 90 organizations that did not respond. These figures describe the invited set and respondents, not a representative estimate of a wider market.

### S2 · Completion-time pilot · v1 · section Design and observations

Eight teams participated. The mean completion time was 10 minutes before the pilot and 8 minutes during the pilot. There was no control group. The task mix changed between the two periods. This record does not isolate the causal effect of AI and does not predict results for other teams.

### S3-v1 · Fictional product page limit · effective 2026-06-01 · Limit

The document page limit is 100 pages under version 1. This note describes the fictional product configuration effective June 1, 2026.

### S3-v2 · Fictional product page limit · effective 2026-09-01 · Limit

Version 2 replaces version 1 for requests from September 1, 2026 onward. The document page limit is now 60 pages. The earlier 100-page note remains a historical record, not the current specification.

### S4 · Citation verification glossary · v1 · Definition

Citation verification checks the identity of a referenced work and whether its content supports the associated statement. This glossary provides no measured success rate, evaluation dataset, or product performance result.

### S5 · Subgroup appendix · metadata only · Table A2 unavailable

Metadata identifies a document called Subgroup appendix, version 1, containing Table A2. No abstract, table, or body passage is supplied. The original claim mentions a 12-point subgroup improvement, but that number and its meaning cannot be checked from this packet. Do not invent the missing table.

### S6-v1 · Eligibility note · original version · Result

The original note records 40 qualifying observations out of 200 eligible observations, or 20%.

### S6-correction · Eligibility note · correction effective 2026-09-02 · Corrected result

The denominator in the original note is corrected from 200 to 160 eligible observations. The numerator remains 40. The corrected proportion is 25%. This notice corrects the denominator; it is not a retraction.

### S7 · Causal interpretation note · v1 · Limitation

Results do not establish causation.

## Claim inventory: preserve these original assertions

- C1: “60% of all 120 invited organizations use the feature.” Cited source: S1 Results.
- C2: “AI caused a 20% reduction in completion time.” Cited source: S2 Design and observations.
- C3: “The current document page limit is 100 pages.” Cited source: S3-v1 Limit; current review date is September 4, 2026.
- C4: “The glossary demonstrates that 90% of citations are verified successfully.” Cited source: S4 Definition.
- C5: “Table A2 shows a 12-point improvement in the subgroup.” Cited source: S5 Table A2, unavailable.
- C6: “The corrected proportion is 40/200, or 20%.” Cited source: S6-v1 Result; the correction is available to this review.
- C7: The source states, “Results establish causation.” Cited source: S7 Limitation.
- C8: “30 of 120 invited organizations responded, a 25% response rate.” Cited source: S1 Results.

## Decision key and revised candidates

| ID | Access to relevant body | Current evidence | Original support | Action | Candidate or next step |
|---|---|---|---|---|---|
| C1 | Available | S1-v1 Results | partial | rewrite | Of the 30 respondents, 18 reported feature use (60%); use among all invitees is unknown. |
| C2 | Available | S2-v1 Design and observations | partial | rewrite | Observed mean time fell from 10 to 8 minutes during the pilot; the record does not isolate AI's effect. |
| C3 | Available | S3-v2 Limit | contradicted | rewrite | As of September 4, the fictional product's current limit is 60 pages under version 2. |
| C4 | Available | S4-v1 Definition | background | remove | Remove the success-rate assertion. A new performance claim would need appropriate evaluation evidence. |
| C5 | Unavailable | S5 metadata only | unresolved | hold | Request authorized access to Table A2; do not approve the number or infer it is false. |
| C6 | Available | S6-correction Corrected result | contradicted | rewrite | The corrected proportion is 40/160, or 25%; the denominator was corrected. |
| C7 | Available | S7-v1 Limitation | contradicted | rewrite | The source states, “Results do not establish causation.” |
| C8 | Available | S1-v1 Results | direct | keep | 30 of 120 invited organizations responded, a 25% response rate. |

partial = narrower support; background = related topic but no proof of the assertion; contradicted = conflict with inspected evidence; unresolved = necessary evidence missing; direct = supports current wording. These labels do not evaluate overall study quality. Reviewer and release owner are unassigned in this teaching packet; candidates are not publication approvals.

## Reproduce the arithmetic, not an AI performance claim

- C1 respondent use: 18 / 30 × 100 = 60%.
- C8 response rate: 30 / 120 × 100 = 25%.
- C2 observed relative decrease: (10 − 8) / 10 × 100 = 20%; causal attribution remains unsupported.
- C6 old fraction: 40 / 200 × 100 = 20%; corrected fraction: 40 / 160 × 100 = 25%.
- Original support counts: direct 1 + partial 2 + background 1 + contradicted 3 + unresolved 1 = 8.
- Relevant-body coverage: 7 / 8 × 100 = 87.5%.
- Direct support across all original claims: 1 / 8 × 100 = 12.5%.
- Direct support among body-assessable claims: 1 / 7 × 100 ≈ 14.29%.
- Actions: rewrite 5 + keep 1 + remove 1 + hold 1 = 8. Six edited/retained candidates do not imply 75% original accuracy.

## Reopen conditions

If S1 changes, reopen C1 and C8. If S3 changes, distinguish historical from current claims before editing C3. If S5 becomes accessible, reassess C5 from the actual table and context rather than changing the label automatically. Every revised sentence needs a new review entry preserving the original. This packet neither authorizes uploads of real confidential documents nor substitutes for qualified review of high-stakes conclusions.