The RAP completion audit: evidencing compliance you cannot reconstruct

A resettlement completion audit asks one question in many forms: did the people named in this plan actually receive what the plan promised, before they were displaced, and are they no worse off than they were? It is normally carried out by an independent party after implementation is substantially complete, and it is the point at which a project's records — not its intentions — determine the finding.

Projects fail completion audits for a narrower set of reasons than might be expected. Rarely because the resettlement was badly done. Frequently because it cannot be shown to have been well done.

What the audit tests

The scope follows the standard's requirements, and each item resolves to a documentary claim:[1][4]

• Completeness. Was everyone eligible identified — including tenants, employees and users of common property?
• Adequacy. Did compensation reach full replacement cost, and can the basis be shown per asset?
• Timing. Was compensation delivered before displacement, per household?
• Delivery. Did the money or the in-kind entitlement reach the named person, and is there proof?
• Grievances. Were they received, handled within the stated timeframes, and resolved?
• Restoration. Have livelihoods been restored relative to the baseline?
• Vulnerability. Were vulnerable households identified against criteria and given the additional assistance the plan specified?

Note the shape of these. Each is a statement about a specific person, a specific amount and a specific date. None can be satisfied by a narrative summary, and none can be honestly assembled after the fact.

Why reconstruction fails

The common approach when an audit is announced is to assemble the record retrospectively: pull the payment schedules, collect the acknowledgement forms, reconcile against the entitlement matrix, and produce a pack. This almost always surfaces the same three problems.

The chain does not join up

A completion audit needs an unbroken line from a person, to the assets recorded against them, to the valuation of those assets, to the entitlement derived from it, to the payment made, to the confirmation of receipt. In most projects these live in different systems — census in one dataset, valuations in a valuer's workbook, payments in the finance system, acknowledgements in a filing cabinet, grievances in a third spreadsheet. Joining them retrospectively is manual, partial, and produces exactly the unmatched records the audit is looking for.

Timing cannot be proven

"Compensation before displacement" is a per-household comparison of two dates. If payment dates were recorded but displacement dates were not — or were recorded at works-package level rather than per parcel — the project cannot demonstrate the sequence, even where the sequence was correct.

Revisions have no explanation

Entitlements change legitimately: a re-measurement, a grievance upheld, a correction. If the record holds only the final figure, an auditor sees a number that differs from the original schedule with nothing to explain it. Legitimate revisions and irregular ones look identical in a system that stores state rather than history.

A spreadsheet records what is true now. An audit asks what was true then, who changed it, and why. Those are different questions, and only one of them a spreadsheet can answer.

Reconciliation is the hard part

The requirement that consistently causes the most difficulty is proving delivery. Payment is usually executed by the project or a paying agent through banks, mobile money operators or cash, producing disbursement files in the payer's format — not the project's.

Reconciling those files back to entitlements per person is where the discrepancies live: payments with no matching entitlement, entitlements with no matching payment, amounts that differ, duplicates, and payments to names that do not match the register. Each is individually explicable and collectively fatal to an audit finding if nobody has looked before the auditor does.

Doing this reconciliation continuously rather than at the end changes the character of the exercise entirely. Discrepancies found in the month they occur are administrative corrections with people still in post who remember the case. The same discrepancies found two years later are findings.

The records that must exist from day one

Working backwards from the audit, these have to be created contemporaneously because they cannot be created later:

• Census entries with enumerator attribution and timestamp, geo-referenced.
• Valuations with method, evidence, date and valuer — and every revision retained with its reason.
• Entitlement determinations traceable to the matrix row that produced them.
• Payment records with date, amount, method, and evidence of receipt by the named person.
• Displacement dates per parcel, not per works package.
• Grievances with the full history: raised, handled, decided, communicated, accepted or escalated.
• Consultation and disclosure records with date, location, language and attendance.
• Monitoring data against the socioeconomic baseline at defined intervals.

Every one of these is a record about an event, created at the time of the event, attributed to a person. That is the whole design requirement, and it is why resettlement record-keeping is closer to an audit trail than to a database.

Why this receives more scrutiny now

The evidentiary bar has risen over the past decade. The 2015 ICIJ investigation into World Bank–financed projects estimated 3.4 million people physically or economically displaced between 2004 and 2013, and documented cases where the Bank's own rules had not been followed.[5] The sector-wide response has been a tightening of verification rather than a rewriting of principles — and increasingly, lenders tie disbursement to compliance milestones, which converts an audit finding into a financial event.

That is the practical reason to treat the completion audit as a design input at the start of a RAP rather than a reporting exercise at the end. The audit does not ask whether the resettlement was done well. It asks whether you can show it.