An indicator that survives an audit

Auditors do not break indicators by being clever. They break them by being literal. They take the definition in your logframe, recalculate the number from your own data, and see whether they land where you landed. When they do not, the finding is rarely fraud. It is usually a definition that was never precise enough to be recalculated. Here is what an indicator needs to survive that process.

A definition two people can execute

Number of people trained is not an indicator, it is a wish. Number of unique trainees who attended at least 80 percent of sessions of an accredited course is an indicator, because two people can execute it independently and land on the same figure. The words doing the work are unique and 80 percent. Every definition should name three things, and your data analyst should never have to ask what counts.

• The counting unit: people, households, trainings or completions. Pick one.
• The inclusion rule: what qualifies a record for the count.
• The exclusion rule: drop-outs, duplicates, dead records, and where they are excluded, in the count or in a footnote.

Verification, named in advance

The means of verification is not a formality at the end of the logframe row. It is the audit trail, chosen before the data exists. Attendance records is weak. Signed daily attendance forms, photographed and attached to the training record is strong, because it names the artefact and says where it lives. When the auditor asks how you know, you point at a thing, not at a colleague.

Disaggregation before the data

Deciding after the fact that you need sex, age band and district is how teams end up with dashboards full of unknown. Disaggregation has to be a field on the form, mandatory, from a short fixed list, and it has to be decided when the indicator is written, because it determines the form. By sex and by district is easy. By income band is only possible if the band was collected. The disaggregation you did not ask for at baseline is the one the auditor will ask for.

The trail

The test of a real audit trail is that you can walk it. Start at the figure on the dashboard: 4,112 trainees, 61 percent women, third quarter, illustrative numbers. You should be able to click through to the submissions behind it, with their dates, their enumerators and their attachments, and from a submission to the photographed attendance sheet behind it. No export, no spreadsheet in the middle, no the person who knows is on leave. A trail that takes an afternoon to assemble by hand is a trail nobody assembles until the auditor asks.

• The dashboard figure links to the indicator definition, its formula and its period.
• The indicator links to the submissions that feed it, each stamped with its form version.
• The submission links to its evidence: photographs, signatures, GPS.

Recalculate it yourself, first

The cheapest audit is the one you run before the auditor arrives. Once a quarter, recalculate your headline indicators from raw submissions with the definition in front of you, and compare against the dashboard. Teams that do this find the same three causes of a gap: a definition applied one way in the spreadsheet and another in the report, a filter that quietly excludes a form version, and a correction that never flowed through. All three are an afternoon's work to fix when you find them, and none of them are when an auditor does.

Change, logged

Indicators get corrected. A duplicate is removed, a misfiled submission is reassigned to the right district, and the quarter moves from 4,140 to 4,112. That is normal work. What matters is that the change is logged: who, when, why, with both numbers retained. An indicator whose history is invisible fails an audit even when the current figure is right, because the auditor cannot tell correction from manipulation. Corrections in the open are credibility.

None of this is hard when it is designed in, and all of it is painful when retrofitted. Define the indicator precisely, name the artefact, fix the disaggregation before the form is built, and keep the trail clickable. Do that, and the audit becomes what it should be: two weeks of mild anxiety instead of two months of archaeology.