Partner data quality audit and restatement of reported results

VF = recounted / reported.

Restatement rule (conservative): the VF is applied to the partner's reported total when the VF is below 1; a figure is not inflated when the VF is above 1 as long as the missing documents (C's supplementary list) have not been sent and verified. For R2.1, the correct unit is the distinct person, so 0.40 is used for A.

R1.1: A 2,400 × 0.98 = 2,352; B 3,100 × 0.78 = 2,418; C 1,900 (kept, 1,995 if the supplementary list is validated). Restated total 6,670 households, i.e. 83.4% of the target (instead of 92.5%). With C at 1,995: 6,765, i.e. 84.6%.

R2.1: A 5,600 × 0.40 = 2,240; B 7,200 × 0.55 = 3,960; C 4,100 × 0.98 = 4,018. Restated total 10,218 people, i.e. 68.1% of the target (instead of 112.7%).

Limitations: the extrapolation assumes that the 3 sites drawn are representative of the 9; with 3 sites, precision is low, especially for B where variability between sites appears high. For B, the restated figure is a provisional ceiling, not a validated result: the warehouse issues (2,850 kits) and the anomalies in document 3 require an exhaustive verification. The R2.1 figure also remains uncertain as long as the de-duplication rule is not applied in the same way by all three partners.

Partner A: methodological error, with no indication of fraud. The attendances are real but the unit is wrong (attendances instead of people), because the reference sheet contains no definition. Responsibility is shared: the consortium did not define the indicator. Correction: recount in distinct people across all sheets, train the team.

Partner C: transmission weakness, with no indication of fraud. The supplementary list is validated by the site committee; it must be obtained, verified (households that genuinely arrived after the closure, no duplicates) and incorporated.

Partner B: serious indications of fraud ("red flags"): identical fingerprints in series, lists filled in by a single hand, a distribution recorded on a day when movements were suspended, a gap of 250 kits between warehouse issues (2,850) and reported distributions (3,100), VFs of 0.78 and 0.55. These are not yet evidence: what is missing is a full reconciliation of stock, delivery notes and lists, the vehicle trip logs for 14 August, an independent verification with a sample of households and an analysis of where the 2,850 issued kits went.

What I do: I document the findings factually, I secure the documents (copies, photos, retention of originals), I report immediately to the anti-fraud focal point (article 11) and I inform my line manager. I suspend validation of B's data in the ITT. What I do not do: I do not investigate myself, I do not question B's staff about the suspicions, I do not accuse anyone by name, I do not share the information beyond those who need to know, and I do not wait until the report is finished to report.

R2.1 reference sheet: title: "Number of distinct people who attended at least 3 of the 4 sessions of the hygiene promotion cycle". Definition: a person is counted only once, regardless of the number of sessions attended; the threshold of 3 sessions out of 4 gives meaning to exposure to the message. Unit: person. Calculation: number of distinct combinations of household card number and position in the household present at least 3 times in the attendance database, over the period. Disaggregation: sex, age group (under 18, 18 to 59, 60 and over), disability (Washington Group questions), status (displaced, host), district. Source: named attendance sheets entered in the shared database, with a unique identifier. Frequency: weekly data entry, monthly consolidation. Responsible: partner monitoring officer (data entry), consortium MEAL assistant (de-duplication and checks), MEAL manager (validation).

Control measures: (1) monthly rotating document-based verification on 3 sites per partner, with the VF published (accuracy, validity); (2) automatic de-duplication by unique identifier in the database (precision, validity); (3) unannounced spot checks during sessions and distributions, and verification calls to 5% of households (reliability, integrity); (4) monthly reconciliation of stock, issue vouchers and distribution lists (integrity); (5) reporting calendar with a deadline, and the supplementary list compulsorily sent within 7 days (timeliness, completeness). A joint training of monitoring officers and a methodological note annexed to the report can be added.

Example: "A data quality audit carried out in month 9 on 9 randomly selected sites led the consortium to restate its results. For indicator R2.1, two partners were counting attendances rather than distinct people: the validated total is estimated at 10,218 people (68% of the target), instead of the 16,900 previously reported. For indicator R1.1, the validated total is estimated at 6,670 households (83% of the target), instead of 7,400. These estimates are conservative and will be consolidated by an exhaustive verification in one district. The R2.1 indicator sheet has been redefined, a shared database with a unique identifier is in place and monthly document-based verifications are now carried out. The consortium maintains the target of 15,000 people and will present a catch-up plan in month 10."

Suspicion of fraud: it does not appear in the narrative report; it is the subject of a separate, confidential notification to the donor by the anti-fraud focal point, within the deadlines of the agreement.

To B's director: I listen to him, I note his explanation and tell him it will be taken into account: if lists were copied out, the originals or what remains of them and the stock records should make it possible to verify this. I do not discuss the suspicions or the procedure under way, and I remind him that article 14 reserves data validation for the lead organisation. I stress that the partnership relationship matters and that the aim is to present defensible figures, which also protects him.

To the country director: I refuse to send figures that we know are not validated: that would be a false declaration to the donor, contrary to the agreement and to our accountability commitments; if fraud is confirmed later, the organisation will have knowingly sent false figures. I propose a solution: restated figures, a methodological note, a separate report by the anti-fraud focal point, and a continuity plan in Montepuez (reinforced support to B if the suspicions are cleared, or another access modality if not). If he maintains his request, I record it in writing and refer it to the regional MEAL manager or the anti-fraud focal point, in line with internal policy.