The case for SmartLARMS

Executive summary

Every legal instrument a project needs already exists. IFC Performance Standard 5 and World Bank ESS-5 state, in unambiguous terms, what a displaced person is owed, when, and on what evidentiary basis.[1][3] National expropriation law in every jurisdiction examined for this brief supplies the constitutional floor beneath it. The nine cases that follow did not fail because a standard was unclear or a law was silent. They failed — or, in two instructive cases, succeeded — because of whether a specific fact could be produced, attributed and dated at the moment it was needed.

That is not a policy problem. It is a systems problem, and it is the problem this paper sets out to demonstrate with the same rigour it asks the reader to apply to any claim about a project under their own authority: named sources, dated documents, and a standing invitation to check every figure against the record it came from.

Why this belongs on a minister's desk, not only a project manager's

Land acquisition and resettlement disputes are not a technical footnote to infrastructure delivery — they are, across the accountability mechanisms that review World Bank and IFC-financed projects, the largest or near-largest single category of complaint.[22][23] That composition is itself a finding: land disputes are not more frequent because complaint mechanisms are more accessible for them than for other harms. It is because displacement produces an identifiable harm to an identifiable person who knows exactly which authority caused it and precisely how to name them in a complaint.

For a ministry or a project sponsor, the exposure this creates is threefold, and each element compounds the others. First, disbursement risk: lenders increasingly tie payment certificates to verified compensation completion by section, so an unresolved resettlement gap does not merely generate criticism — it stops money moving and the works with it. Second, political risk: a complaint that reaches the World Bank's Inspection Panel or IFC's Compliance Advisor/Ombudsman becomes a matter of public record, reviewed by an institution the government does not control, on a timeline measured in years.[24] Third, fiscal risk: a court can and does overturn awards years after the fact — one case examined below saw a High Court award cut by 98% on appeal, and a separate one saw a government defend a KSh 1.93 billion payment against a petition alleging it was paid against the wrong parcel entirely.[13][10] Every one of these risks is a records risk before it is anything else, and every one of them is visible years in advance to whoever is actually keeping the record — which is the entire argument for keeping one properly from the start.

The legal floor every project already stands on

None of what follows depends on adopting a new standard. IFC PS-5 requires that displaced persons — across three eligibility categories, including those with no recognisable legal right to the land — be compensated at full replacement cost, calculated without deduction for depreciation, before possession is taken.[1] Guidance Note 5 elaborates the evidentiary expectation behind each element of that requirement: a documented cut-off date, a disclosed valuation method, and a completion audit conducted by a party independent of the implementing team where impacts are significant.[2] World Bank ESS-5 imposes the equivalent obligation on the borrower, through an agreed Environmental and Social Commitment Plan.[3] IFC's own Good Practice Handbook — the operational reference its safeguards specialists work from — treats a documented, individuated, time-stamped record as the ordinary expectation of a competent programme, not as an enhancement.[4]

National law in every jurisdiction this brief examines sits alongside that framework rather than beneath it. Uganda's Constitution conditions compulsory deprivation of property on prompt payment of fair and adequate compensation prior to the taking — a sequencing requirement, not merely a valuation one.[25] Kenya's Land Act sets out a mandatory inquiry into compensation at sections 107 through 133, and its own courts have held — in the SGR judgment examined as Exhibit 2 below — that "prompt full and just compensation" means an equivalent value ascertained through that inquiry, not a speculative one.[20] Tanzania's Valuation and Valuers Registration Act fixes the cut-off date to the date valuation itself commences, a rule the Tabora–Kigoma SGR resettlement plan applies section by section across a single corridor.[19] None of these three regimes is silent, vague, or under-developed. What every one of them assumes — and what none of them supplies — is the operational system capable of proving, per person, per parcel, per date, that the requirement was met. That is the gap this brief is about, and it is the same gap in every case that follows, regardless of which country's law or which lender's standard governed it.

The record: nine documented cases

Each case below is independently sourced and dated. Two — Exhibit 2 and Exhibit 8 — are included because a project or a claimant won a legal contest on the strength of its own records, which is the argument for this brief made from the winning side rather than the losing one.

Exhibit 1 — the audit that PS-5 required and nobody scheduled

Bujagali Hydropower Project, Uganda. The Compliance Advisor/Ombudsman's investigation into the Bujagali transmission line found that IFC "did not ensure the Resettlement Action Plan (RAP) for the transmission line met the requirements of PS5" — and specifically that the completion audit PS-5 mandates was never carried out, leaving no independent verification that compensation had reached full replacement cost.[6] The complaint that triggered the finding was filed in February 2015. CAO closed the case in April 2025 — ten years later — rating IFC's response on land compensation Unsatisfactory and recording no expectation of resolution.[7] A completion audit is, structurally, a query run against a record that should already exist. Where the record does not exist, the audit cannot be improvised under deadline — which is exactly what happened here, and exactly why the case took a decade to close without resolution.

Exhibit 2 — the case a government won because its paperwork was complete

Standard Gauge Railway, Kenya. A five-judge bench of the High Court dismissed a constitutional challenge to the SGR's land acquisition in March 2016 — not because the underlying grievances were unarguable, but because the National Land Commission and Kenya Railways produced the gazette notices, vesting notices and affidavits of service tying each disputed parcel to a named, served claimant. The court's holding is worth quoting directly: "it was for the Petitioner to demonstrate exactly when the alleged illegal entries or acts of trespass took place. The Petitioner did not and has not."[8] The claim failed on an evidentiary asymmetry, not on the substance of the grievance — and the asymmetry ran in the government's favour because its service record, alone among the case studies in this brief, actually held.

Exhibit 3 — two rates in one clause

Dar es Salaam Bus Rapid Transit, Phase 4, Tanzania. The disclosed, World Bank-financed Resettlement Action Plan states that the prevailing average bank deposit rate "is 8%," then instructs the valuer, in the following sentence, to "obtain the 7% of the value and add to the previous total" — inside the clause governing the disturbance allowance payable to every eligible project-affected person on the project.[9] Whichever figure was in fact applied cannot be settled by the governing document itself, and the inconsistency will be unanswerable the first time two neighbouring claimants compare their disturbance-allowance figures and find them computed differently for reasons the file cannot explain.

Exhibit 4 — a parcel reference one subdivision out of date

Nairobi Expressway, Kenya. A petition before the Environment and Land Court, filed in 2025, alleges that KSh 1.93 billion was paid in 2021 to compensate for land identified by an obsolete parcel reference — while the live, subdivided reference for the same holding was never touched by the road and remains under compensation-related restrictions to this day.[10][11] The claimant's remedy sought is instructive: not a renegotiated valuation, but an order compelling the National Land Commission and the Kenya National Highways Authority to produce the documents used to identify and value the parcel in the first place. The entire dispute is a question about which reference number was used — a fact that a correctly designed register makes trivial to answer and a spreadsheet makes trivial to lose.

Exhibit 5 — three surveys to count the same people

Ruzizi III Regional Hydropower Project, Rwanda and DRC. The World Bank's own Concept-stage safeguards disclosure records that social assessments conducted in 2012 and 2021 identified the presence of Batwa households in the project area but classified none of them as project-affected persons. A third survey, in 2022, finally identified roughly eighty Batwa households across both countries and confirmed that at least eleven are affected by land acquisition.[12] Nine years elapsed between the population first being noted and first being counted. What changed between the second and third survey was not the population's circumstances; it was whether the census instrument was built to ask the question that would have found them the first time.

Exhibit 6 — the breach that survived a valuation defeat

Karuma Hydropower Project, Uganda. The Court of Appeal, in January 2022, rejected a High Court's UGX 9.3 billion award to nine landowners on the ground that rock and mineral deposits are part of the land and cannot be valued as a separate asset — cutting the award by roughly 98% to UGX 203.2 million.[13][14] But the reduced sum was awarded as general damages for a distinct constitutional breach: possession of the land had preceded payment, contrary to Article 26(2)'s sequencing requirement. The valuation claim failed entirely; the sequencing claim succeeded anyway, on a fact — which date came first — that a register comparing possession date to funds-available date would have flagged automatically, long before either party reached a courtroom.

Exhibit 7 — one disputed parcel, an entire corridor frozen

Kampala–Jinja and Kampala–Entebbe Expressways, Uganda. By early 2021, of 4,535 project-affected persons evaluated for compensation on the Kampala–Entebbe corridor, 1,435 remained unpaid, concentrated almost entirely in one contested section.[15] The underlying cause reaches back to 2017, when a member of the Buganda royal family petitioned to halt compensation pending resolution of a title dispute over a single parcel, Block 273 Masajja — land the petition claimed formed part of a historic royal estate. The road authority responded by halting 80% of compensation across two separate expressway projects while the title question was litigated.[16] A single disputed claim, with no mechanism to segregate it from the undisputed majority, stalled payment to thousands of unrelated households for years.

Exhibit 8 — the right recognised because the record could name who held it

Lamu Port, Kenya (LAPSSET corridor). On 30 April 2018, the High Court held that traditional fishers working the waters off Lamu hold "penumbral property rights" under Articles 26, 28, 40 and 43 of the Constitution — rights a dredging programme for the new port had extinguished without compensation — and identified more than 4,700 such fishers as entitled to an award ultimately quantified at KSh 1.76 billion.[17] The judgment is remarkable precisely because no land was taken from any of the claimants: the court had to be persuaded, on the evidence put before it, that a defined population depended on a specific resource in a way the Constitution protected. Payment, once ordered, still took until 2024 to complete — six years after judgment — in part because verifying the eligible population against the number the court had accepted required a reconciliation exercise the case record itself does not suggest existed before the dispute arose.[18]

Exhibit 9 — a government that wrote its own gap analysis

Standard Gauge Railway, Tabora–Kigoma section, Tanzania. Not a failure — the exception this brief exists to generalise. The disclosed, AfDB-financed Resettlement Action Plan for this section contains its own table reconciling Tanzanian statute against AfDB Operational Safeguard 2, stating in terms that "PAPs under category C such as the encroachers to the land are not entitled to compensation under the Tanzanian laws" — and that the lender standard requires their inclusion regardless.[19] This is what the other eight exhibits are missing: a project stating, in writing, before payment begins, exactly which population national law would exclude and exactly which instrument requires their inclusion anyway. It is the single clearest piece of evidence in this entire brief that the problem it describes is solvable with discipline alone — before any software is involved. What software adds is making that discipline the default rather than the exception one well-run project happened to practise.

The pattern behind all nine

Read individually, these are nine unrelated disputes across five countries and three lenders' standards. Read together, they are one recurring failure wearing nine different faces: a fact that existed somewhere — in a valuer's field notes, a survey team's memory, a clerk's spreadsheet — and could not be produced, dated and attributed, at the exact moment an auditor, a judge or a claimant needed it. Not one of the nine required a new law, a new standard, or a policy nobody had thought of. Bujagali needed a scheduled audit. Musimba needed (and had) a service log. Dar es Salaam needed one stored number instead of two typed ones. Nairobi needed a cadastral check run at the moment of valuation, not carried forward from an earlier exercise. Ruzizi needed a census instrument that asked who uses land, not only who owns it. Karuma needed two dated fields compared automatically. Kampala–Entebbe needed a status field that could isolate one disputed parcel from four thousand undisputed ones. Lamu needed a reconciliation method decided before the court order arrived rather than after it. Tanzania's SGR shows what happens when the discipline exists anyway.

This is the case for treating a resettlement register as infrastructure in its own right, procured and resourced with the same seriousness as the physical works it accompanies — because in every exhibit above, the cost of not doing so was ultimately paid in cash, in schedule, or in a finding a government or a lender had to answer for in public. SmartLARMS exists to be that infrastructure: one linked, attributed, time-stamped register — census, valuation, entitlement, payment reconciled against disbursement files, and grievances — built offline-first for exactly the field conditions every one of these nine projects was operating under.

What nine cases cost, added up

A minister weighing whether a records system is worth procuring before a project begins, rather than after a dispute forces the question, is entitled to see the number attached to not having one. It cannot be computed precisely — several of the disputes above remain open, and a delay's true cost includes schedule effects this brief has no access to model — but the disclosed figures alone are not small. Karuma's award, before appeal, exposed the government to UGX 9.3 billion on a valuation theory the Court of Appeal ultimately rejected; defending that appeal, and the years of uncertainty before it was decided, was not free.[13] The Nairobi Expressway petition places KSh 1.93 billion in dispute over a single parcel identification error — an amount the National Land Commission and the road authority must now either defend in litigation or explain how it was recovered.[10] Lamu's fishers waited six years between judgment and full payment of a KSh 1.76 billion award, a gap the reporting attributes in part to the verification exercise the eligible population required — work that had to happen eventually and cost considerably less to do before the court order than after.[18] Bujagali's transmission line dispute consumed a decade of institutional attention across two organisations before closing without resolution.[7] None of these figures is a projection. Each is a disclosed sum a real institution had to account for, in public, because a record that should have existed at the time of payment did not exist when it was later tested.

What an appraisal review actually tests

For a lender's safeguards team, the question is rarely whether a borrower has heard of PS-5 or ESS-5 — every borrower seeking project finance has. The question an appraisal mission actually tests is narrower and more mechanical: can this specific record, for this specific household, be produced on request, and does it join up — census entry to valuation to entitlement determination to payment to acknowledgement of receipt — without a gap a reviewer has to take on trust. IFC's own handbook is explicit that this chain, not the plan document, is what a supervision mission actually samples.[4] A borrower whose system can answer that request in the meeting room moves through supervision in the time it takes to run the query. A borrower whose system requires two weeks and a special data-pull exercise to answer it has, in the reviewer's experience if not on paper, already failed the test — because the inability to produce the record quickly is itself evidence about how the record was kept.

This is also where disbursement conditionality bites hardest, and where the cost of a weak register compounds fastest. A condition requiring that compensation be verified complete on a given section before civil works begin there is only as fast as the verification method behind it. Where that method is a manual reconciliation between a census spreadsheet, a valuer's workbook and a finance system's disbursement file — three systems that were never designed to talk to each other — verification itself becomes the bottleneck holding up a payment certificate, independent of whether the underlying compensation was in fact complete. A borrower loses schedule time to prove a fact that was already true, because nothing in its system could say so quickly enough.

What a procurement decision should actually specify

Ministries and sponsors that have absorbed the argument this far typically ask the same practical question next: what should a terms of reference or a system procurement actually require, so that the next project does not become exhibit ten. Four requirements recur across every case above, and none of them is exotic. A stable, unique identifier per person and per parcel that survives the life of the project, so a record from year one can still be located and trusted in year eight. A change history on every entitlement determination and every valuation, showing who revised what and why, so a legitimate correction and an unexplained one are distinguishable rather than indistinguishable, which is the condition every completion audit in this brief was actually tested against.[4] Two valuation columns, not one blended figure, wherever national law and the applicable lender standard diverge — Karuma and Bujagali both turn on exactly this distinction going unrecorded.[2] And field capture that works without connectivity, because every project examined in this brief operates in exactly the rural and peri-urban conditions where a system requiring a live network connection is a system that will not be used at the point data is actually collected.

What this does not solve

A register cannot supply a policy that does not exist, survey a footprint nobody walked, or force a lender to enforce a milestone it has chosen not to enforce. Karuma's constitutional breach still required a court to award damages after the fact; a system of record makes the breach visible before it happens, which is a materially different — and more valuable — thing than curing it afterwards, but it is not the same as a government choosing to act on that visibility. Ruzizi's nine-year census failure is closed by a correctly designed instrument, not by storage capacity alone — the schema still has to ask the right question, which is a design discipline before it is a technical one. And Musimba's lesson cuts in a direction worth stating plainly rather than softening: the party with the better record wins the evidentiary contest, which is a reason to be that party and not, on its own, a guarantee that the underlying position being defended is the just one. The United Nations Guiding Principles on Business and Human Rights are explicit that access to effective remedy is a distinct obligation — not a by-product of good administration — and a register that makes a project's own conduct legible does not discharge that obligation on its own; it is the precondition for an institution actually meeting it.[26]

What changes, stage by stage

It is worth being concrete about what a properly instrumented project looks like at each point where one of the nine exhibits above went wrong, because the abstraction "a system of record" can otherwise sound like a general aspiration rather than a specific operational change.

At census, every household and every parcel receives a stable identifier the moment it is first recorded, and the instrument asks not only who owns the land but who uses it, farms it, rents it or depends on a resource attached to it — the question Ruzizi III's first two surveys did not ask and its third finally did.[12] At valuation, every asset carries two figures rather than one: the statutory value under national law, and the full replacement cost under the applicable lender standard, with the gap between them visible as a number rather than absorbed into a single negotiated figure nobody can later explain — the distinction whose absence decided both Karuma and Bujagali.[13][6] At disclosure, the cut-off date is recorded against the section it actually applies to, with the channel, language and audience it was communicated through, so that a later claimant is tested against a specific evidentiary record rather than a general assertion that notice was given — the exact fact Musimba's respondents could produce and Karuma's claimants relied on the court to test the other way.[8]

At payment, possession date and funds-available date sit on the same record, per parcel, so a sequencing breach of the kind Article 26(2) prohibits is visible as a flag before it becomes a damages award — not a fact reconstructed from memory once a claim has already been filed.[25] Where a specific claim is disputed, it carries its own status rather than blocking disbursement to every unrelated claimant on the same corridor — the design that would have prevented one title dispute over Block 273 Masajja from freezing four-fifths of compensation on two separate expressway projects.[16] And at closure, the completion audit is a report run against a record that has existed continuously since the project began, rather than a reconstruction exercise commissioned under pressure once an accountability mechanism has already opened a case.[7]

None of these six changes is exotic, expensive, or dependent on a capability that does not already exist in commercial software. What they require is that the six live in one linked register rather than in six separate systems a finance team, a valuer, a field enumerator and a grievance officer each maintain independently — because it is precisely in the gaps between those separate systems that every one of the nine exhibits above actually occurred.

The proposition this brief asks a reader to test

Every documented resettlement failure examined in this brief involved a standard that already existed, in writing, before the failure occurred. What was absent, in each case, was a system that made a specific fact impossible to lose between the person who recorded it and the person who later needed it — an auditor, a judge, a claimant, or a minister answering for the project in public. That is a testable claim against a public record, not an assertion asking to be taken on faith, and the sources behind every exhibit above are listed for exactly that purpose: so a reader with the standing to act on this brief can verify each one independently before doing so.

That is the gap SmartLARMS is built to close — and the reason to close it before the tenth case, not after it. A tenth case is not a hypothetical risk being managed conservatively; on the base rate this brief documents, across five countries and a single decade, it is closer to an actuarial certainty for any institution running a comparable pipeline of land-intensive projects. The question a minister, a sponsor or a lender is actually deciding is not whether a resettlement dispute of this kind will eventually reach their desk. It is whether, when it does, the record their institution can produce reads like Exhibit 2 and Exhibit 9 — or like the other seven.