How Postwar Recovery Is Measured | What Must Be True Before We Call It Progress?

What Must Be True Before We Call It Progress?

Recovery is measured by whether people regain dependable access to the things that make life possible—not simply by how much money is spent, how many buildings are repaired, or how many people cross a border. A credible assessment follows the whole journey from resources to functioning services, from services to usable opportunities, and from opportunities to improvements that last. It also asks who remains excluded, what the evidence cannot establish, and whether an intervention caused the improvement being claimed.

That is a more demanding task than counting reconstruction projects. It is also a more useful one. The purpose of measurement is not to produce an impressive account of recovery. It is to discover what still prevents recovery, early enough for someone to do something about it.

The chapters below develop a practical way to read that evidence. The framework is an educational synthesis, not an official recovery index or a certification of any country. International sources are identified in the notes. The district, households, classroom conversations and numerical exercises are explicitly fictional teaching examples; they are not reported interviews, actual programme results or descriptions of a current conflict. Legal, engineering, health and humanitarian decisions require the relevant competent institutions and local standards.

1. The school with a new roof

Imagine a school on the edge of a fictional river district. Its roof has been replaced. The windows are glazed. A delivery of desks has arrived. At the entrance, a notice announces the reopening. The reconstruction report records one school completed, six classrooms restored and a substantial number of children expected to benefit.

Every statement may be accurate.

Now move the point of observation. Instead of standing at the school gate, stand in the kitchen of a household across the river. A child lives there, but the bridge on the usual route remains closed. The alternative journey is longer. The available transport leaves before the caregiver can finish another responsibility. A place in the reopened school exists, but the household cannot reliably use it.

Move again, this time into a classroom. A teacher is present, although the class includes children who have missed very different amounts of learning. Some can read the textbook. Others can recognise individual words but cannot follow a paragraph. The room is full, which is good. Yet the same lesson cannot be understood in the same way by everyone sitting in it.

Move once more, into the school office. The headteacher knows the temporary transport subsidy will end before the next term. A replacement budget has not been confirmed. Attendance looks healthy this month. Its continuation is uncertain.

Has the school recovered?

The question is too large for the evidence supplied. The building has recovered a particular physical function. A teaching service has resumed. Some children have regained access. Learning recovery has not yet been established. The financial conditions supporting attendance may not last. These statements do not cancel one another. Together, they describe the situation more honestly than either celebration or dismissal.

This is the first discipline of recovery measurement: do not make one true observation carry several unproved conclusions.

A roof is evidence of a roof. It can also be one necessary condition for education. It is not evidence, by itself, of regular attendance, appropriate teaching, understanding, progression or future independence. When we call the whole system restored because one component has been repaired, we shorten the causal chain in our language. The unrepaired links do not disappear. They become harder to see.

The reverse mistake is equally damaging. It would be wrong to say that the roof does not matter because learning remains incomplete. Children should not have to wait for perfect teaching arrangements before gaining a dry room. A physical improvement can be real and worthwhile without being the final outcome. Measurement should protect the value of partial progress while preserving the visibility of unfinished work.

The OECD’s evaluation criteria distinguish several questions that a single completion count cannot answer: relevance, coherence, effectiveness, efficiency, impact and sustainability. The organisation presents these as complementary grounds for judgement, not as a method that automatically produces a causal answer. That distinction is important. A project can meet its construction objective and still fail to fit the needs of the people expected to use it. 1

For the school, a better report would connect several records. It would state what construction was verified, how many teaching days actually took place, who attended, what barriers kept other children away, what learning evidence was available, and which costs remained unfunded. It would not need to make the report enormous. It would need to keep the claims separate.

There is a practical consequence. If the problem is the roof, the next action is construction. If the problem is the bridge, the next action concerns access. If the problem is instruction pitched above the children’s present understanding, more desks will not solve it. If the problem is an expiring transport budget, a photograph of a full classroom can conceal a failure already approaching.

A useful measurement system therefore resembles a good diagnostic conversation. It does not ask only whether something is better. It asks what changed, for whom, through which mechanism, with what limits, and what observation would alter the next decision.

Return to the child across the river. Their experience is not a sentimental addition to the statistics. It reveals the missing connection between the restored asset and the intended benefit. The report did not necessarily contain a false number. It contained an unfinished explanation.

This is where the larger inquiry begins. Recovery becomes more measurable when we stop asking a single number to describe a whole life.

2. Recovery is not a rewind button

The word recovery invites an apparently reasonable comparison: before the war and after it. What was lost? What has been restored? How close are we to the old level?

That comparison matters. It is not enough.

In the fictional river district, suppose the prewar school served a community in which some children were already excluded. Restoring the previous attendance level could reproduce that exclusion. Suppose the old water system was unreliable, although its breakdowns attracted little public attention. Rebuilding it exactly would restore an inadequate service. A return to the past can be an improvement from catastrophe without being an acceptable destination.

There are therefore at least three different reference points. The first is the prewar condition: what people had, including its strengths and inequalities. The second is the condition at the start of an intervention: what could actually be observed when a particular recovery programme began. The third is the standard or objective the recovery effort intends to reach. Confusing these reference points creates misleading claims.

Consider a simplified housing example. Before conflict, 800 of 1,000 households occupied housing meeting an agreed local adequacy standard. At the beginning of recovery, only 400 did. Two years later, 700 do. The district has made a substantial improvement from the start of recovery. It has not regained its earlier total. And even its earlier total left 200 households below the standard.

These three statements answer three questions: how far the programme period has moved, how much of the loss remains, and how much unmet need still exists. None requires us to pretend that the prewar world was fair or satisfactory.

A further complication appears when the population changes. Suppose the district now contains 1,300 households because people have arrived from elsewhere. Seven hundred adequately housed households then represent a smaller share of current need than the original comparison suggests. A programme can restore many homes and still fall behind the demand placed on it. That does not automatically show incompetence. It shows why a recovery target cannot be interpreted without its population and date.

The distinction between restoring structures and rebuilding a workable future also appears in formal recovery planning. The World Bank describes Recovery and Peacebuilding Assessments, undertaken with the European Union and United Nations, as a shared platform for assessing, prioritising and sequencing recovery and peacebuilding activities. Such planning is not simply a list of objects to replace. It concerns how interventions fit together in a changing situation. 2

For measurement, the implication is to state the ambition before choosing the indicator. Does a housing programme aim to repair damaged owner-occupied dwellings? To reduce homelessness? To enable voluntary return? To improve tenure security? To increase access for people with disabilities? These goals overlap, but they are not identical. An indicator can be well measured and still be unsuitable for the claim attached to it.

The ambition should also distinguish recovery from improvement. A municipality may reasonably decide to rebuild a damaged administrative service with more accessible opening hours and better records than before. Part of the result is restoration. Part is reform. Calling everything restoration obscures what has changed and why its cost differs from simple replacement.

Equally, improvement should not become an excuse to delay essentials indefinitely. A temporary service may be necessary while a more durable arrangement is prepared. Measuring the temporary service honestly means identifying what it provides, what it does not, who depends on it, and what would make its replacement ready. The choice is not between perfect rebuilding and no rebuilding. It is between explicit stages and indefinite improvisation.

Targets also have moral boundaries. A national average can recover while a community remains unable to live without discrimination. Economic output can expand without restoring access to justice. These are not merely statistical imperfections to be smoothed away. They are reasons to resist treating recovery as a single upward curve.

A stronger question is: which human capabilities should no longer depend on emergency arrangements, and what evidence would demonstrate that change? A household should be able to secure water, pursue learning, obtain care, earn or receive a lawful income, and resolve ordinary problems through usable institutions. The exact standards and responsibilities differ by place. The need to state them does not.

The prewar world supplies a history, not a complete instruction manual. Good measurement remembers the losses without making yesterday’s inadequacies the definition of success.

3. Follow the chain from money to a changed life

A reconstruction announcement often begins with money. A household’s account of recovery rarely does. It begins with a practical difference: the pump works, the payment arrived, the bus returned, the school accepted the child, the office corrected the document.

Between the announcement and the difference lies a chain of events. Measurement must make enough of that chain visible to locate failure.

Start with resources. A programme has money, people, equipment, authority and time. These are inputs. It then uses them: procuring materials, repairing a building, training staff, testing a system or reaching households. These are activities. The immediate products might be repaired rooms, completed training, issued documents or transferred payments. These are outputs. Whether people can use those outputs to improve their circumstances is a further question about outcomes. Longer-term changes, including unintended effects, require another level of assessment.

The vocabulary is less important than the separation. A reader should be able to translate a report into ordinary language: what went in, what was done, what was produced, what changed for people, and what lasted.

In the fictional district, a records office receives new computers. Staff are trained. An application portal opens. Five thousand records are digitised. These are legitimate achievements. Now ask whether residents can correct an erroneous birth date without repeated travel, whether a person with no internet connection can still apply, whether the updated record is accepted by the school system, and whether the office can recover the data after an outage.

A computer count cannot answer those questions. Nor can the number of staff trained. Even the number of records digitised can mislead if errors were reproduced faster than they were corrected. The point is not that outputs are unimportant. It is that the relationship between an output and a benefit must be examined rather than assumed.

One useful practice is to write each connection as a sentence beginning with “because”. The portal should reduce repeated journeys because residents can submit a complete application remotely. That sentence exposes several assumptions: residents can access the portal; they understand the requirements; remote submission is legally and administratively accepted; missing evidence can be supplied; and the application does not still require the same number of visits later.

Once these assumptions are visible, measurement becomes more intelligent. Instead of collecting twenty superficial indicators, the programme might examine where applicants stop, which documents are repeatedly rejected, and how many journeys are required from first attempt to final correction. This does not require treating residents as data-generating machines. It requires asking the few questions that distinguish one explanation from another.

The same approach clarifies responsibility. A construction team can complete its work while a service fails because staffing was never arranged. Blaming the builder for every later failure is unfair. Declaring the whole programme successful because the builder fulfilled the contract is equally misleading. The programme-level assessment must connect the organisations without pretending their duties are interchangeable.

There is also a difference between an administrative event and an experienced event. A benefit can be authorised but unpaid. It can be paid into an account but inaccessible to the intended recipient. It can be collected but reduced by fees. It can arrive in full but too late to prevent a damaging household decision. Each transition needs a different observation.

In our example, the district decides to measure a document-correction service from the applicant’s first complete submission to confirmed acceptance of the corrected record by the next relevant institution. This is not a universal metric for all registries. It is a deliberately chosen end point for a particular service problem. Its strength is that it extends beyond the office’s internal definition of completion.

The office should still keep its internal processing measure. Otherwise it may be blamed for delays outside its control. The stronger arrangement reports both: the office’s processing time and the resident’s end-to-end time. The gap between them becomes a question for coordination rather than a reason to argue about whose number is true.

Not every chain is linear. Better records may improve benefit delivery, which helps households afford transport, which increases school attendance. A transport repair can change which school a family chooses, altering classroom demand elsewhere. Recovery involves feedback. The measurement chain is a simplification, but a useful simplification when its limits are stated.

A good report therefore does not merely claim that inputs became outcomes. It identifies the important transitions and supplies evidence for the ones on which the conclusion depends. Where a transition remains unverified, the report says so. That unfinished link is often the most valuable information in the document.

4. Count people before calculating success

A percentage looks precise because arithmetic is precise. The population beneath the arithmetic may be much less certain.

Suppose the fictional district reports that 80 per cent of eligible households received support. The numerator is 800 households. The denominator is 1,000. The calculation is correct. But where did the 1,000 come from? An old register? A new survey? Applications already accepted? A population estimate? Each answer changes the meaning of the result.

If the denominator contains only people who successfully registered, the measure is a payment-completion rate among registered households. It is not a measure of coverage among all households in need. Families who lacked documents, missed the announcement, could not travel or were turned away disappear before the calculation begins.

Imagine that there are actually 1,250 eligible households. The programme has reached 800, or 64 per cent of them. Both 80 per cent and 64 per cent can be honest numbers when labelled properly. The first describes performance after registration. The second describes estimated reach across the eligible population. The misleading act is to present the first as though it answers the second.

This problem becomes more difficult when people move. A household can be registered in the district where it fled, appear on a return list in its place of origin, and temporarily stay with relatives in a third area. Counting records is not the same as counting households. Deduplication matters, but it is not automatically solved by matching names. Two records can represent one household at different times; two similar names can represent different people.

The correct denominator depends on the question. For local water planning, current residents matter. For understanding the experience of those displaced from a village, people now living elsewhere may remain part of the relevant population. For evaluating a benefit programme, the legally or administratively eligible population matters. No one denominator can serve every purpose without qualification.

A useful report states its population in a full sentence. “Current residents in the surveyed part of the district during the reference month” is less glamorous than “the population”, but it is much clearer. “Registered programme households whose payment was due this quarter” may be exactly the right population for a payment audit. It should not silently become “all affected families”.

Household definitions also deserve attention. In our fictional district, two families share a dwelling after displacement. Do they prepare food together? Share a budget? Receive support as one unit or two? The answer can alter eligibility, overcrowding statistics and household income measures. The evaluator should use the programme’s stated definition for the relevant calculation and explain where it differs from another dataset.

There is a practical way to prevent confusion: maintain separate counts of known records, estimated people or households, and unresolved matching cases. Do not merge them into a reassuring total before the uncertainty is resolved. A count of 1,000 confirmed cases plus 70 cases awaiting reconciliation is more informative than a single total whose status nobody understands.

The United Nations’ Fundamental Principles of Official Statistics emphasise professional independence, confidentiality, transparency, relevance and equal access. These are not merely ideals about how statisticians should behave. They protect the conditions under which a population count can be examined without being rewritten to fit the preferred political story. 3

A missing denominator should not paralyse all action. A district may need to distribute urgent support before a full population estimate exists. The honest response is to use the best available estimate, state its date and limitations, and revise coverage as evidence improves. Acting under uncertainty and pretending uncertainty is absent are different choices.

An approximate denominator can also be represented as a range. If the eligible population is plausibly between 1,100 and 1,300 households, reaching 800 implies estimated coverage between about 61.5 and 72.7 per cent. That range does not tell us which value is correct. It tells us that reporting 80 per cent as population coverage would require evidence outside the assumed range.

This is a small example of how uncertainty can be made useful. The range changes the interpretation without requiring a false claim of exact knowledge.

Before accepting a recovery percentage, ask who is allowed into its denominator, who is missing, and whether the same definition was used last time. A society can become more visible to its institutions even when its first corrected statistics look worse. That is not necessarily deterioration. Sometimes it is the end of invisibility.

5. Three clocks and the problem of the baseline

Recovery does not run on one clock. An intervention has an implementation clock, households have a life clock, and measurement has an observation clock. Their misalignment explains many apparently contradictory reports.

A water repair may be completed in one week. A household may notice the benefit only when the service remains dependable long enough to stop storing emergency supplies. An evaluator may visit once, during a period when the repair happens to be working. Each event occurs at a different point in time.

A baseline is meant to give comparison a starting point. Yet a postwar baseline is often already the result of several changes. Some households have left. Emergency assistance has arrived. Roads have reopened informally. Prices have moved. People have adapted. The first survey is not a photograph of the moment conflict ended. It is a measurement made under particular conditions at a particular date.

In the fictional district, suppose a programme begins in January but its baseline survey occurs in March. Repairs have already started in two neighbourhoods. Calling March “before the programme” would be inaccurate. The survey may still be useful as an early observation, but it cannot support a clean before-and-after comparison for those areas.

One response is to reconstruct earlier conditions from records. That can help, provided reconstructed evidence remains distinguishable from direct observation. An attendance book, a service log and a resident’s recollection have different strengths and vulnerabilities. A reconstructed baseline should state its sources, the period they cover, and what could not be recovered.

Another response is to choose a narrower question. Rather than claiming to estimate the entire programme’s effect from January, the evaluator might examine whether service reliability improved between March and September, while acknowledging that some early gains preceded the first measurement. Narrowing a claim is often better than stretching the evidence to fit a larger ambition.

Seasonality introduces a second clock problem. A household may have better food access after harvest even without a new programme. Travel times may change with weather. School attendance follows terms and holidays. Comparing unlike periods can attribute an ordinary seasonal difference to recovery.

Our district therefore compares the same seasonal window where possible. When that is not possible, the report names the difference and explains how it could affect the result. This does not make every observation unusable. It prevents a calendar effect from becoming an unearned causal claim.

A third problem is the duration of observation. A service measured on one day can establish its condition on that day. It cannot prove continuous reliability over three months. A household asked about its experience over the previous month provides a broader account, but recollection may be imprecise. Logs provide repeated records, but their completeness must be checked. The appropriate evidence depends on whether the claim concerns a moment, an interval or a pattern.

Consider electricity in our fictional example. A photograph of a lit building shows that lighting was available at the moment of the photograph. A meter record can show consumption over time. A service log can describe interruptions. A household account can reveal whether outages occurred during the hours when power mattered most. None replaces all the others.

Timing also affects fairness. An intervention may arrive sooner in an accessible neighbourhood than in a remote one. An endline conducted on one common date then gives the first neighbourhood twelve months of exposure and the second only three. Comparing their outcomes without considering that difference can mistake unequal time to benefit for unequal effectiveness.

A useful measurement record therefore includes the date a service actually became usable, not merely the date a contract was signed. It can then distinguish programme age from calendar time. This is especially important when phased implementation makes a single “after” period misleading.

The end of observation should be chosen with equal care. Measuring immediately before temporary subsidies expire can present an optimistic picture of sustainability. Measuring during an unusual disruption can present an unduly pessimistic one. The answer is not to search for the most flattering date. It is to explain the observation window and, where feasible, return after the conditions supporting the initial result have changed.

A baseline does not make a study rigorous by existing. It becomes useful when its timing, sources, population and relationship to implementation are understood. The beginning of a spreadsheet is not necessarily the beginning of a causal story.

6. Measure the time people lose

Some of the largest gains in recovery are measured in hours that people no longer have to surrender.

In the fictional district, a restored water service reduces the need for repeated collection journeys. A functioning registry reduces visits to different offices. A dependable school timetable reduces the uncertainty around a caregiver’s working day. These changes may not appear immediately as higher income. They alter what people can attempt with their time.

Time is not simply another cost to be converted automatically into money. It is a condition of access. A clinic may charge little but require a journey that a person cannot combine with work or care. A benefit may be formally available but require several visits during office hours. For someone with flexible employment, the procedure is inconvenient. For someone paid only when present, it may be unaffordable.

A useful recovery measure can therefore follow an entire task. How long does it take to obtain the service from the first attempt, including travel, waiting, repeat visits and the time spent correcting errors? The question should be specific. A general estimate of “administrative burden” can hide very different experiences.

Suppose two fictional households each receive the same monthly transfer of 100 units. One receives it through an accessible channel with a fee of one unit. The other pays six units in transport and fees and loses half a day of paid work. The gross transfer is identical. The usable benefit is not. We do not need to assign a monetary price to every minute to recognise the difference.

Measurement should also distinguish average delay from predictable delay. A service that normally takes two days is different from one that takes two days on average but sometimes takes three weeks without explanation. People plan around uncertainty as well as duration. A caregiver may make contingency arrangements because the service is unreliable, even if its average performance looks acceptable.

This gives us two questions rather than one: how long does the process take, and how much can people rely on that expectation? The first concerns speed. The second concerns predictability. They may require different changes. More staff can reduce a queue. Clearer information can reduce repeated unsuccessful visits. Better coordination can reduce the long delays affecting exceptional cases.

For a simple synthetic example, imagine five applicants whose processing times are one, one, two, two and twenty-four days. The average is six days. The median is two. Neither is false. Reporting only the median conceals the person waiting twenty-four days; reporting only the average suggests an experience nobody except the long-waiting applicant came close to having. A small distribution or a count of cases exceeding a stated threshold makes the problem clearer.

Thresholds must be chosen for reasons. A promise to resolve an urgent document problem within a week may be connected to school admission or a payment deadline. The same threshold might be unnecessary for a less urgent record request. The measurement should not invent a universal standard merely because a round number fits a dashboard.

Repeated small burdens matter too. Saving ten minutes once differs from saving ten minutes every day. Yet multiplying minutes by a population can exaggerate precision when actual use is uncertain. A responsible estimate states the frequency, the number of people using the service, and the assumptions behind any annual total.

In our district, a researcher resists announcing that a repaired crossing has “saved a million hours”. Instead, the report first establishes who uses it, how journey times changed, which trips became possible, and whether people shifted from another route. Only then does it offer a bounded estimate of time saved. The calculation follows the evidence rather than providing an impressive number in search of a basis.

Time also reveals unequal recovery. A new digital process may reduce waiting for connected households while increasing the burden on those who must find help to use it. The aggregate average can improve while a smaller group faces a worse process. Examining task time by access needs, location or application type helps reveal the trade-off without requiring the public release of identifiable personal information.

The deeper benefit is not merely that people move faster through a system. It is that less of their life is consumed by negotiating its failures. A dependable service returns room for study, work, care, rest and choice. Those possibilities are part of recovery even before a national income statistic can see them.

7. Water: a pipe is not a service

Water makes the difference between construction and recovery unusually clear. A pipe can be installed correctly while the service remains unavailable when people need it. A source can be nearby while collection consumes substantial time. Water can arrive regularly without meeting the relevant quality requirements.

The WHO/UNICEF Joint Monitoring Programme defines a safely managed drinking-water service through a combination of conditions: an improved source, access on the premises, availability when needed, and freedom from the relevant contamination. The definition does not allow a strong result on one dimension to replace a missing condition on another. 4

For the fictional district, that distinction changes the evaluation question. Counting repaired connections tells the municipality about its assets. It does not establish whether households receive the intended service. A household-level assessment must connect the condition of the system to the experience of using it.

Consider a separate illustrative exercise involving 100 households. Eighty have a repaired connection. Seventy report water available when needed. Sixty have evidence meeting the stipulated quality condition during the observation period. These are not necessarily the same households. It would be wrong to declare that 60 households meet all conditions simply because 60 is the smallest of the three counts.

To determine the number meeting all three conditions in this exercise, we need their overlap. The household records might show that only 45 satisfy all three. Or they might show that all 60 in the quality group also have repaired connections and availability. The separate totals do not identify the intersection.

This is a recurring measurement mistake: treating several encouraging percentages as though they automatically belong to the same people. The issue is not advanced mathematics. It is whether the evidence follows a household through all the required conditions.

The district should therefore choose its reporting level deliberately. It can report the status of the infrastructure, the separate service dimensions, and, where the data genuinely permit it, the share of households meeting the combined definition. The combined measure is especially useful when the public promise requires all conditions together. It should not conceal which condition is causing failure.

Suppose the fictional assessment finds that the principal gap is reliability. More connections are not the immediate answer. The next inquiry concerns interruptions, demand, operation and maintenance. If quality is the main gap, the response requires qualified investigation and the applicable safety standards. The educational point is that different failure patterns require different professional responses; this chapter is not a guide to water testing or treatment.

The time window matters. An acceptable sample at one moment cannot prove permanent quality. A month of reasonable availability does not establish performance through a different season. Equally, one disruption should not automatically erase a sustained improvement. Reporting should specify what was measured, when, how often, and which claim those observations can support.

Households beyond the network deserve separate attention. A report focused only on connected households may improve continuously while the unconnected population remains outside its field of view. The denominator should match the stated purpose. A utility performance measure can reasonably concern its customers. A district recovery claim concerns everyone whose access the district is claiming to describe.

A useful public summary might therefore say: “The project restored eighty connections. In the separate household assessment, forty-five of one hundred assessed households met all three locally specified service conditions during the stated observation period. Reliability was the most common unresolved condition. Households outside the surveyed area are not included.”

That paragraph is less triumphant than “water restored to the community”. It is more actionable. It recognises the construction achievement, identifies the service gap, states the measured population, and prevents the reader from extending the result beyond its evidence.

There is another question the report should keep open: whether households trust the service enough to change their behaviour. Continued expenditure on alternative sources may reflect unresolved problems, earlier experience, habit, or reasonable caution. Behaviour alone does not establish water quality, but it can reveal where the official account and household confidence diverge. Follow-up can distinguish those explanations.

The strongest recovery measure does not force the household to choose between trusting a government announcement and trusting its own experience. It creates a record in which both the technical conditions and the practical experience can be examined. The repaired pipe matters because of what it enables. Measuring that enabled service is how the district discovers whether the repair has reached its purpose.

8. Education: count learning without losing the learner

A reopened school is a genuine public achievement. It is not a complete measure of educational recovery.

An education system has to connect access, participation, appropriate teaching and demonstrated learning. Measuring only one of these can give a misleading account of the others. A child can be enrolled without attending regularly. A child can attend regularly without understanding the lesson. A child can perform well on a familiar exercise without transferring the skill to a new text or problem.

UNESCO’s SDG 4 monitoring framework distinguishes learning proficiency, completion, out-of-school rates and other educational dimensions. Indicator 4.1.1 concerns minimum proficiency in reading and mathematics at specified stages, rather than simply the presence of a child in a classroom. These distinctions provide a useful reason to keep attendance and learning evidence separate. 5

Return to the fictional school with the new roof. The first task is not to administer an elaborate test because measurement sounds rigorous. It is to decide what the school needs to know in order to teach. Can the returning children read the materials now being used? Which mathematical ideas are available to them? Which instructions do they misunderstand? A short, well-chosen assessment can inform teaching more directly than a large score that gives no clue about the difficulty.

For public recovery measurement, however, an additional question arises: which children took the assessment? Suppose 200 children are on the relevant population list, 160 are enrolled, 140 attend on the assessment day and 100 demonstrate the selected skill. These are illustrative figures, not school results. The proportion demonstrating the skill among those assessed is about 71.4 per cent. The proportion with demonstrated evidence in the original population is 50 per cent.

We cannot infer that every unassessed child lacks the skill. Nor can we count them as proficient. The report should preserve both facts: 100 children demonstrated the skill, and evidence is missing for part of the population. The missingness matters because attendance may be related to the same barriers that affect learning.

A rising average can also conceal a change in who is present. If the least-supported children stop attending, the school’s assessed average may improve without any child learning more. This possibility does not discredit assessment. It tells us to keep track of participation and, where appropriate and ethically feasible, compare the same learners over time while reporting those who are no longer observed.

Assessment design must also avoid rewarding mere familiarity. If the same short passage is used repeatedly, improvement may partly reflect memory of that passage. A carefully chosen new task can test whether the relevant skill transfers. The new task must still be comparable enough to support the intended comparison; novelty alone does not create a fair measure.

Language introduces another boundary. A child may understand a mathematical idea but not the language used to present it. A reading assessment, by contrast, is partly about language itself. The evaluator needs to specify the construct being measured rather than assume that every low score describes the same deficiency. Instructional language, prior schooling and accessibility can change what a task actually tests.

None of this makes a single classroom teacher responsible for resolving the entire measurement problem. Recovery systems should avoid imposing a reporting burden that consumes the time needed for teaching. Some information belongs in routine classroom assessment. Other questions require periodic system-level assessment with appropriate sampling and expertise. Repeatedly collecting the same information for different programmes is not automatically better evidence.

The fictional school chooses three connected records. It tracks regular participation, uses brief assessments to identify instructional needs, and periodically checks whether learning gains appear on unfamiliar but comparable tasks. It also records why some children remain absent. These records serve different decisions. Attendance barriers may require transport or household support. Instructional gaps require a teaching response. Incomparable assessment tasks require a measurement correction.

A careful recovery statement might read: “Classes resumed for the stated number of teaching days, and regular participation increased. The assessed learners showed improvement on the defined tasks. The result does not describe children who were not assessed, and the school is investigating their access barriers.”

That is neither a weak statement nor an admission that nothing worked. It is a map of what the evidence establishes. The child across the river remains inside the educational question even when they are absent from the classroom data. Educational recovery becomes more honest when the easiest children to count are not allowed to stand for every child who needs to learn.

9. Health: readiness, care and the journey between them

A clinic can be open without being ready to provide a particular service. It can be ready without delivering that service well. It can deliver competent care to those who attend while remaining inaccessible to people who need it.

These distinctions are not rhetorical. WHO’s Harmonized Health Facility Assessment separates service availability, service readiness, quality and safety of care, and management and finance. The earlier SARA approach focused on availability and readiness; WHO explicitly notes that SARA alone does not attempt to measure the quality of services or resources. The assessment framework matters because a facility inventory and a quality-of-care judgement are different forms of evidence. 6 7

For a postwar recovery report, the danger is to treat a completed inventory as a verdict on healthcare. A building, a staff roster and an equipment list can establish important conditions. They cannot, by themselves, show that an individual received appropriate care, that a referral was completed, or that the person could afford to seek help when needed.

Imagine the fictional district repairs three clinics. A report records all three as operational. One has regular staff but intermittent opening hours. Another is physically accessible but difficult to reach by transport. The third is functioning well yet attracts patients from a neighbouring area, increasing demand beyond its planned capacity. Calling all three operational may be administratively defensible, but it obscures different problems requiring different responses.

The first measurement question should follow the service being promised. Is the report describing basic availability, readiness for a defined intervention, continuity of an existing treatment pathway, or the quality of care? A broad claim such as “healthcare restored” cannot be supported by a narrow measure unless the missing dimensions are acknowledged.

Now follow a fictional patient’s journey without pretending to offer clinical advice. The person recognises a need, finds information, reaches a facility, is assessed by qualified staff, receives the indicated service or referral, and can continue the necessary follow-up. The system can fail before the person appears in a clinic register, after they leave, or between facilities.

If measurement begins only at the clinic door, those unable to reach care remain unseen. If it ends at the first consultation, incomplete referrals remain unseen. A recovery assessment does not need to collect every private medical detail to recognise these gaps. It can use appropriately governed service and access measures, with qualified health teams determining the standards and safeguards.

A simple illustrative count shows the distinction. Suppose a facility records 300 consultations in a month. That might reflect improved access. It might also reflect repeated visits because a problem was not resolved, a temporary outbreak, demand diverted from a closed facility, or changes in record keeping. The count alone does not tell us which explanation is correct.

Similarly, a decline in consultations could indicate improved health, reduced population, unaffordable travel, fear, shorter opening hours or missing records. Direction is not interpretation. The evaluator must connect the number to the service context.

This is especially important when a report claims prevention. If an intervention is intended to avert deterioration, success may not appear as an increase in service volume. The appropriate outcome may be continuity, reduced interruption or fewer avoidable access failures. Evaluating prevention requires a credible account of what would otherwise have happened, which cannot be replaced by the observation that a bad event did not occur.

Health recovery also exposes the limits of average travel time. A median journey of twenty minutes can coexist with a remote group facing two hours. The average may matter for system planning, but the tail matters for people whose needs cannot be postponed. Public reporting should make consequential access gaps visible without releasing sensitive household locations.

The fictional district therefore asks its health professionals to define a small number of service-specific measures. It reports facilities available, readiness for the services being claimed, continuity where follow-up is required, and barriers experienced by people unable to obtain care. It does not construct an amateur clinical score or label a whole population healthy because buildings reopened.

The lesson extends beyond medicine. A system’s readiness is a necessary condition for many benefits, but it is not the benefit itself. Postwar measurement should honour the work of restoring readiness while remaining curious about what actually happens when a person needs the service. The ultimate question is not whether a clinic can be photographed as open. It is whether a person can depend on an appropriate, safe pathway through it.

10. Livelihoods: employment is not the whole economic story

When a recovery programme announces jobs created, the first question is what counts as a job. A second question follows immediately: what changed in the worker’s life?

The ILO distinguishes employment—work for pay or profit—from other forms of work, such as unpaid care, volunteering and production for own use. Its labour-force framework also distinguishes unemployment from being outside the labour force. People can contribute considerable work without appearing as employed, and people who want employment may not satisfy every condition used in the unemployment measure. 8

These distinctions matter after conflict because a headline employment figure can conceal changes in hours, earnings, security and unpaid responsibilities. The purpose is not to reject the official definitions. It is to understand what they do and do not answer.

In a fictional livelihoods programme, 120 people complete training. Eighty later report some paid work. Fifty remain in paid work at a subsequent follow-up. The programme should not describe all 120 as having gained lasting employment. It should also not ignore the eighty who found work because some arrangements did not persist. Completion, entry into work and continued employment are different stages.

The figures still leave questions. Did participants work more hours than before? Were earnings paid on time? Did transport and other work-related costs absorb most of the income? Did a temporary project end as scheduled? Did some participants move into education or care responsibilities rather than simply “fail” to remain employed? A binary employment indicator cannot describe every transition.

Take a separate illustrative household calculation. A person earns 300 units a month before a disruption and 360 after a recovery intervention. At first glance, income has risen by 20 per cent. But suppose the cost of a relevant household basket rises from 100 to 150 units over the same period. The earlier income purchased three baskets; the later income purchases 2.4. Nominal earnings improved while purchasing power, measured against that particular basket, declined.

This calculation does not establish a complete welfare index. Household needs differ, prices vary, and the chosen basket may be incomplete. It does establish why money amounts require a price context. A report should not call households economically better off solely because the currency number is larger.

Net earnings matter as well. A new job can require transport, tools, childcare or accommodation. Some of these expenses may be worthwhile investments; others may make the opportunity unusable for the intended worker. Measuring gross pay without the major costs can overstate the practical change.

Unpaid work complicates the picture further. In the fictional district, a household’s adult obtains paid employment while another member takes on additional care. The wage is a real gain. The redistributed burden is also real. An evaluation that asks only the employed person about earnings may miss the change in the household’s total workload and the effects on another person’s education or rest.

A livelihood should also be understood across time. A short contract can provide a valuable bridge without becoming a durable income source. A small business can survive its first months only because a grant absorbs early losses. Neither result is necessarily a failure. The report should state the stage of recovery and the conditions on which continuation depends.

For business support, revenue is not profit, and an enterprise registered is not an enterprise operating. A shop may reopen with stock but few customers. A manufacturer may receive machinery but lack working capital. A trader may report higher sales while replacing inventory at even higher prices. The useful measure follows the particular constraint the intervention was intended to change.

There is also displacement between businesses. A grant-supported shop may gain customers who previously bought from an unsupported shop nearby. The supported business’s gain is not automatically a net gain for the district. This does not imply that support should never be provided. It means that claims about the wider economy need evidence beyond the beneficiary’s own accounts.

The fictional programme therefore reports a sequence: participation, demonstrated skill, entry into paid work or enterprise activity, earnings and major costs, and continuation over a stated interval. It records significant reasons for leaving, where participants choose to provide them. It does not treat every exit as a failure or every temporary success as a permanent one.

Economic recovery is not merely the return of transactions. It is the return of usable ways to support a life. Employment measures are essential to that inquiry, but they become more informative when connected to time, purchasing power, cost, stability and the work that remains unpaid.

11. Return: crossing home is not the same as rebuilding a life

A bus reaches a town. People step down with their possessions. A return movement has occurred. What it means for durable recovery remains to be established.

The distinction is particularly clear in the IASC framework for internally displaced people. Durable solutions concern the resolution of displacement-related assistance and protection needs and the ability to exercise rights without displacement-related discrimination. The framework is not limited to the act of returning to a place of origin; it recognises other sustainable settlement possibilities. JIPS describes analysis across different dimensions and includes both displaced people and host communities in its approach. 9 10

This chapter uses that distinction as a measurement lesson. It does not collapse refugees and internally displaced people into one legal category, and it does not offer individual advice about whether a person should return. Those situations require the relevant legal and protection frameworks. The common analytical issue is that a movement count does not establish the quality or durability of life after movement.

In the fictional river district, suppose 500 households are recorded as returning during a year. The count should specify whether it concerns unique households or return events. A household that moves back, leaves again and returns later can generate more than one movement. Without that distinction, repeated instability can inflate a total presented as successful return.

The next question concerns continued residence, but residence alone is not enough either. A household may remain because it has no viable alternative. Another may choose to settle elsewhere and achieve a more stable life. A recovery score that rewards return regardless of choice risks confusing a preferred administrative outcome with the person’s own circumstances.

Follow a fictional household after arrival. Its dwelling exists, but occupancy is contested. Water is accessible but unreliable. The children have places in school, yet the family has no stable income. One adult has an unresolved identity record, preventing completion of another administrative task. These are separate barriers that can reinforce one another.

An assessment should avoid declaring the household “reintegrated” merely because it has crossed a geographic threshold. It should ask which relevant conditions have become dependable and which remain unresolved. This can include housing stability, access to services, documentation, livelihoods, safety and the ability to participate without discrimination, using the appropriate framework and safeguards.

The comparison group matters. Suppose returning households report worse service access than households who never left. That difference may identify displacement-related barriers. But suppose both groups have equally poor access. Equality at a low level does not establish satisfactory living conditions. A comparative measure should be read alongside an absolute standard or a clearly stated account of unmet need.

The reverse comparison also requires care. Returning households might receive temporary support that makes their measured consumption higher than that of unsupported neighbours. This can be a legitimate response to need, but the report should examine duration and local effects rather than interpreting the temporary difference as complete recovery. It should not conceal hardship among host or non-displaced households.

Longitudinal evidence is especially useful here because a life after movement unfolds over time. A household observed at arrival may look different six months later. The problem is that follow-up itself can be incomplete. People move again, change contact information, or choose not to participate. Those no longer reached by a survey should not silently be classified as stable, unsuccessful or dead. They remain unobserved for that follow-up unless other reliable evidence establishes their status.

A UNHCR-hosted catalogue for research on durable solutions among internally displaced people in Iraq illustrates the use of a multi-year study and a multidimensional framework rather than a single return count. The existence of such a design does not remove every problem of follow-up or causality; it shows that the relevant questions extend beyond arrival. 10

For the fictional district, a careful report might distinguish movements recorded, unique households verified, households observed at follow-up, and the conditions those households report or demonstrate. It would explain the status of people not reached again. It would preserve voluntary choice as a separate issue rather than awarding points for moving in the direction the programme prefers.

Recovery should restore the ability to make ordinary decisions about where and how to live. Counting return can help describe one part of that process. Calling return a durable solution requires evidence about the life available after the journey, not merely the direction in which the journey ended.

12. Justice: a closed case may still be an open problem

A court reports that it has resolved more cases this quarter than last. That could be excellent news. It could also be a number whose meaning depends on what happened to the cases.

Were decisions reached after a fair process? Were cases dismissed because records were missing? Did people withdraw because they could no longer afford to continue? Were straightforward cases selected while the oldest or most consequential disputes remained pending? An administrative closure is an event. Justice is a more demanding claim.

UNDP’s description of SDG indicator 16.3.3 focuses on the share of people who experienced a dispute and accessed a formal or informal dispute-resolution mechanism. The denominator therefore begins with people who had disputes, not merely those who already reached a court. That is a useful distinction between institutional workload and public access. Access itself, however, does not establish that every process or outcome was fair. 11

In the fictional district, consider a land dispute. A resident files a claim. The file is accepted. A hearing is held. A decision is issued. Yet the relevant record is never updated, and the decision is not implemented. From one office’s perspective, the case is complete. From the resident’s perspective, the original problem remains.

A recovery assessment should therefore decide where its claim ends. It may measure filing access, time to first hearing, age of pending cases, availability of legal assistance, reasoned decisions, review where applicable, or lawful implementation. These measures serve different questions. They should not be merged into a claim that “justice has been restored” without explaining the connection.

Speed has a real value. Prolonged uncertainty can block the use of property, delay family decisions and make rights difficult to exercise. But speed is not the only value. A procedure that becomes faster by preventing people from presenting evidence may improve a throughput measure while weakening fairness. A court that hears more complex cases may become slower on average while addressing problems previously ignored.

A synthetic example makes the issue concrete. Suppose a court begins a quarter with 1,000 pending cases, receives 200 new cases and closes 250. It ends with 950 pending cases, assuming no other adjustments. The backlog has fallen by fifty. The closure count is 250. Reporting only the closure count could suggest a larger reduction than occurred. Reporting only the remaining backlog could conceal the substantial work completed while new demand continued to arrive.

Now divide the pending cases by age. If the oldest cases remain untouched, a modest fall in the total may coexist with worsening delay for those who have already waited longest. The appropriate response is not necessarily to process every case strictly by age; urgency and legal requirements matter. The report should explain the prioritisation rule and show which groups of cases remain unresolved.

Case mix matters too. A routine administrative correction and a contested claim involving several parties are not equivalent units of work. A performance target based only on closures can reward the easiest cases. A better assessment describes the composition of work and avoids interpreting one raw count as a complete measure of institutional competence.

Public access has its own missing population. People who do not know where to bring a dispute, cannot afford travel, fear retaliation or lack required documentation may never appear in the court’s records. Household or community evidence can help identify these barriers when collected safely. The assessment should not pressure people to disclose sensitive disputes merely to fill an indicator.

The fictional district therefore uses both institutional and public-facing evidence. Court records show workload, delay and procedural stages. Appropriately protected feedback shows whether people can navigate the process and understand decisions. Follow-up on suitable cases checks whether an apparently completed outcome has become effective in practice. None of these sources alone is treated as the whole picture.

The distinction also prevents unfair blame. A court may issue a lawful decision while another institution fails to act. That failure belongs in the overall assessment without being misattributed. Recovery measurement is strongest when it can locate the broken link precisely enough to assign the next corrective task.

The goal is not a justice system with no disputes. A functioning society will continue to disagree about property, contracts, family matters and public decisions. Recovery becomes visible when those disagreements can enter a fair, understandable and usable process without requiring private power. A closed file is one possible sign of progress. A problem that has been resolved lawfully is the more important result.

13. Money spent is evidence of spending

A financial report can be accurate and still be mistaken for a report of social progress. The confusion often begins with verbs.

Money is announced, committed, contracted, disbursed, received, spent or accounted for. These words describe different events. A large announcement may concern several future years. A signed contract creates obligations but does not prove delivery. A payment can be made before the intended asset is usable. Expenditure can be properly documented while the resulting service remains incomplete.

For the fictional district, the first improvement is simply to stop using one financial total for all these stages. Suppose a recovery programme has an announced envelope of ten million units. Six million are covered by signed financing agreements. Four million have been disbursed to the programme. Three million have been paid to suppliers. Goods and works valued at two and a half million have been accepted under the relevant procedures. These illustrative figures are not contradictory. They describe a pipeline.

A report should not call the ten million “delivered to residents”. It should not call the two and a half million the entire financing available. Each figure has a legitimate use when its status and date are stated.

Now add outcomes. Some accepted work has restored a service. Some has produced assets awaiting another connection, staffing decision or operating budget. It would be wrong to assume that accepted expenditure has no value until every outcome is achieved. It would be equally wrong to treat acceptance as proof that every intended household benefits.

A financial audit and an outcome evaluation therefore answer different questions. The first can establish whether money was handled according to the relevant rules and supported by appropriate records. The second asks what difference the intervention made and whether the result is worth sustaining. Neither should pretend to replace the other.

This distinction protects accountability rather than weakening it. A project that is well managed financially but poorly conceived requires a different correction from a useful project damaged by fraud. In the first case, better procurement alone will not solve the design problem. In the second, a worthwhile objective does not excuse misuse of funds.

The district’s report should also separate price changes from physical progress. If the same quantity of work costs more because prices have risen, higher expenditure does not necessarily mean more work. If a lower price allows the same quality to be delivered for less, lower expenditure does not necessarily mean weaker recovery. The interpretation requires quantities, specifications, timing and relevant price assumptions.

Currency conversion adds another layer. A project’s cost can be reported in local currency or a reference currency. Changes in exchange rates can alter the reported value even when the local activity is unchanged. A comparison should state the currency basis and conversion dates rather than allow accounting presentation to create an apparent change in effort.

There is a similar problem with the expression “funding gap”. A gap can mean the difference between estimated needs and announced pledges, signed financing, cash received or resources available for a specific period. A report that moves between these definitions can make the gap shrink without any new spendable funds arriving. The calculation should identify both sides of the comparison.

Consider a school whose rebuilding is financed but whose operation is not. The capital budget may be complete while the service budget remains uncertain. A recovery report should include the costs that begin after construction: staff, utilities, supplies, maintenance and replacement where relevant. These are not administrative details outside the project’s meaning. They are conditions of its continued usefulness.

The report also needs to recognise domestic and household contributions. A programme may count external finance carefully while ignoring work, money or time supplied locally. That omission can exaggerate the donor’s share of the result and understate the burden placed on residents. Such contributions should not be romanticised or assumed freely available; where they are material, they belong in the account of what the intervention required.

The fictional district adopts a modest rule. Every headline financial figure receives a status, date, currency basis and purpose. Every major outcome claim identifies evidence beyond the financial transaction. A reader can then follow the movement from financing to delivery without being asked to treat the movement of money as the movement of life itself.

This is not hostility toward spending. Recovery needs resources. The distinction is that money has to become something useful through a series of decisions and actions. Counting expenditure tells us that part of the process occurred. Measuring recovery asks what that process enabled, whether the benefit reached people, and what will keep it working when the account is closed.

14. The cheapest repair is not always the least costly recovery

A district with limited resources must compare alternatives. Refusing to compare costs is not compassionate; it can leave fewer people helped. But the comparison has to concern the same kind of result.

A cost per building repaired can be useful for construction management. It does not necessarily identify the least costly way to restore a service. Two buildings may serve different populations, have different operating requirements or remain usable for different lengths of time. A lower construction cost can be offset by higher maintenance, shorter life or poor access.

Consider two explicitly fictional options for a small service facility. Option A costs 100,000 units to restore and 20,000 units a year to operate. Option B costs 140,000 units initially and 10,000 units a year to operate. Over five years, ignoring discounting, inflation, replacement, residual value and differences in service quality, the totals are 200,000 and 190,000 units respectively.

The simplified calculation does not prove that B is better. It shows that the cheaper initial repair is not automatically the cheaper five-year arrangement. A real decision would require the omitted factors, including whether either option can actually deliver the needed service. The purpose of a worked example is to expose the structure of the decision, not to supply a universal answer.

Now suppose Option B takes an extra year to become usable. The delayed service has a cost even if the final asset is more economical. Some of that cost might be estimated through temporary arrangements or unmet need. Some may be difficult to value monetarily. A careful comparison states the trade-off rather than hiding time behind a five-year total.

Conversely, an apparently rapid option may only move delay elsewhere. A temporary building opens quickly, but it requires repeated closures or unusually costly staffing arrangements. The initial schedule looks better while the service experience remains unstable. Timeliness should be measured at the point where people can use the intended benefit, not only where a contractor declares a milestone complete.

Cost-effectiveness also depends on the denominator. “Cost per beneficiary” sounds straightforward until one asks whether a beneficiary is a person registered, contacted, served once, served regularly or experiencing a verified improvement. Dividing by everyone expected to benefit can make an ineffective programme look cheap. Dividing only by successful cases can conceal how many people the programme failed to reach.

The fictional district therefore reports both reach and result. A programme costing 100,000 units may contact 1,000 households, provide the intended service to 700, and have verified evidence of the selected outcome for 500. The respective ratios are 100, about 143 and 200 units. They are different measures, not competing versions of the same measure.

The outcome measure still needs a time horizon and an uncertainty statement. A cost per verified outcome at six months should not be described as the cost of a permanent improvement. If the evidence comes from a sample, extrapolation should be justified and its uncertainty explained. Precision in division cannot repair uncertainty about how many outcomes actually occurred.

A further issue is substitution. A new programme may replace an existing service rather than add its full reported benefit. That replacement can be worthwhile if it improves quality, fairness or cost. But the comparison should concern the incremental difference: what changes relative to the feasible alternative, not relative to an imaginary world in which nobody would otherwise act.

Equity can justify choosing a more costly option. Reaching a remote or previously excluded group may cost more per person. That does not make the decision inefficient by definition. It means the objective includes access and fairness, not merely the largest number served for the lowest amount. The report should make that objective explicit so a value judgement is not disguised as arithmetic.

The same discipline applies to resilience. Backup capacity may look wasteful during ordinary operation but support continuity during a disruption. Its value cannot be assessed only by how often it is used in a quiet month. At the same time, invoking resilience should not make every expensive backup unquestionable. The relevant failure, likelihood, consequence and alternative should be examined to the extent the evidence permits.

A useful cost comparison therefore asks: compared with what, over what period, for which population, at what quality, with what risk and with whose costs included? Those questions do not eliminate judgement. They make judgement visible.

The least costly recovery is not necessarily the project with the smallest invoice. It is the arrangement that uses scarce resources well to produce the intended, usable and sufficiently durable benefit. Sometimes that requires a better asset. Sometimes it requires a simpler process, a repaired connection or a budget for the staff who make an existing asset work.

15. The average can improve while the people most in need fall behind

An average is a summary, not a promise to every person inside it.

Suppose the fictional district measures successful completion of a public-service task. In one period, 90 of 100 well-connected households complete it, while 20 of 100 households facing access barriers do so. Overall completion is 55 per cent. Later, the survey reaches 180 well-connected households and only 20 households facing barriers. Their completion rates are now 85 and 15 per cent respectively.

Both groups perform worse. Yet the combined rate is 78 per cent: 153 successful cases in the first group plus three in the second, divided by 200. The aggregate improves because the composition of the measured population changes toward the better-performing group.

All figures here are invented to demonstrate arithmetic. They show why a rising headline percentage cannot be interpreted without examining who was measured. The change could be genuine population movement, a sampling problem, a revised eligibility rule or selective loss to follow-up. The report must establish which explanation applies.

Even with a stable population, averages conceal the distribution. A reduction in average waiting time can coexist with a minority waiting longer. Higher median income can coexist with severe hardship among those without work. Better overall service coverage can leave the same remote settlements unserved year after year.

This is not an argument against averages. It is an argument for asking what decision the average is meant to support. A national planning estimate may reasonably use an aggregate. A decision about exclusion requires information about the people on the wrong side of the aggregate. The appropriate disaggregation should follow the problem rather than collect sensitive characteristics indiscriminately.

Location is often useful. So are age, disability-related access needs, household circumstances or programme stages, where lawful and relevant. But a group label does not automatically explain a difference. If a remote group experiences lower access, the cause might be transport, service distribution, cost, language, discrimination or a combination. The first task is to reveal the gap. The next is to investigate its mechanism.

Small groups raise a statistical and ethical challenge. Their estimates can be imprecise, while publishing detailed combinations of characteristics can make individuals identifiable. Hiding every small group would erase some of the people whose experience matters most. Publishing every detail can expose them. An appropriate response may combine protected analysis, broader public categories and carefully described qualitative evidence.

The relevant question is not how many categories fit on a chart. It is whether the analysis can identify consequential exclusion without creating avoidable harm.

Another trap is to compare only improvements. Suppose an underserved neighbourhood improves from 10 to 30 per cent coverage while another improves from 80 to 90. The first has gained more percentage points. It is still far less well served. Ranking programmes solely by the size of improvement can obscure remaining need; ranking solely by final level can undervalue difficult progress from a very low starting point.

A useful report shows both starting condition and current condition, then interprets the change in context. It can acknowledge that an intervention achieved a meaningful gain while identifying a large unresolved gap. Recovery does not become less real because it remains incomplete.

Targets can also create perverse comparisons between groups. An easy-to-reach area may meet its target quickly because the target was conservative. A harder area may miss an ambitious target despite helping more people. Performance against target is one piece of evidence about implementation, not a complete ranking of social value.

The district can improve its reporting with a simple structure. First, show the whole-population measure with its definition. Second, show the most relevant differences across groups or locations. Third, identify the largest remaining gaps and the limits of the evidence. Fourth, explain which decisions will respond to those gaps. This is more useful than adding a generic statement that nobody should be left behind while presenting only national totals.

Equity also concerns voice in measurement. The people easiest to interview are not necessarily those facing the greatest barriers. Public meetings can be dominated by those with time, confidence or influence. A survey conducted through an online channel can exclude people whose lack of connectivity is itself part of the problem. Inclusion has to shape how evidence is gathered, not merely how the results are described.

The fictional district’s most important discovery may therefore be that its apparent improvement partly reflects a shrinking view of the population. Correcting that view could lower the reported success rate. The correction is progress in knowledge, even though it makes the headline less flattering.

A recovery system should be able to survive that kind of honesty. Otherwise it will gradually learn to measure the people for whom it is already working and call the rest a data problem.

16. Recovery often depends on the weakest missing connection

Imagine four services in the fictional district: water, power, schooling and transport. Each has improved. The report presents four upward arrows. It is tempting to conclude that everyday life must be improving in the same proportion.

That conclusion may be right. It still needs evidence about connections.

A school can have power and a repaired building while pupils lack transport. A water system can have equipment while an operating budget is missing. A clinic can be staffed while its referral pathway is unusable. Improvements in separate components do not automatically combine into a functioning experience for the same person.

The mathematical issue is familiar from the water example: the overlap matters. If 80 per cent of households have one necessary condition and 80 per cent have another, we cannot assume that 80 per cent have both. Nor can we multiply the percentages without a reason to assume the relevant independence. In real service systems, access conditions may be strongly related.

A household with good transport may also be more likely to have stable power and nearby services. Another may face several barriers at once. The average condition in each sector can therefore look better than the combined experience of households who need all of them.

For a small synthetic exercise, consider ten households. Eight can reach a school. Eight can afford the necessary participation costs. Eight have the required documentation. It is possible for only four households to satisfy all three conditions: each of the six remaining households fails a different condition, with two failures assigned to each category. It is also possible for eight to satisfy all three if the same two households face every barrier.

The separate totals are identical in both arrangements. The combined reach is not. The lesson is to gather linked evidence only where it is necessary and safely governed, rather than infer a household-level result from unrelated sector totals.

Dependencies also affect sequencing. Suppose a repaired bridge would allow staff and supplies to reach both a clinic and a school. Its benefit extends beyond transport. If every sector evaluates only its own construction output, the bridge’s wider contribution may be undervalued. Conversely, a school project may appear unsuccessful because the bridge repair was delayed, even though the school construction itself was competent.

A recovery assessment should therefore identify the critical connections supporting its main outcomes. It need not map every institution in the country. It needs to know which dependencies could stop the promised benefit from becoming usable.

A practical method is to choose a small number of ordinary journeys: a child attending school, a household obtaining water, a resident correcting a document, a patient completing a referral. For each journey, ask what must be present, what can be substituted, and what cannot be bypassed safely or lawfully. The resulting map is a decision aid, not a technical instruction for operating the service.

The map should distinguish a bottleneck from a general weakness. A bottleneck constrains the outcome even when other inputs are available. If the only authorised signatory is absent, more clerks may not reduce delay. If a payment channel is inaccessible, faster eligibility decisions may not improve receipt. Correctly identifying the bottleneck prevents the programme from repeatedly investing in the parts it already knows how to improve.

Yet the weakest-link metaphor also has limits. Systems can have alternatives. A household may use another school, another source or another lawful service route. Some substitutions are costly but effective. Others transfer burdens to families. Measurement should record what substitutes actually do rather than assume either perfect dependence or unlimited adaptability.

Resilience is partly the ability to maintain essential function when one connection fails. In the fictional district, an offline application route might preserve access during a network outage. Its existence is not enough. Does staff knowledge survive? Are the records reconciled afterwards? Do residents know the route is available? A backup that cannot be used in practice is only a documented intention.

This is why exercises and retests can matter. A service may appear functional under ordinary conditions yet fail when a common disruption occurs. Qualified teams can test the relevant continuity arrangements without creating dangerous conditions or withholding essential services. The evaluator should report what was tested, under what circumstances, and which conclusion follows—not claim general invulnerability.

The most useful recovery meeting may therefore begin with a question that crosses departmental boundaries: where does a person still get stuck after each agency has reported its own task complete? The answer often lies between organisations, where nobody’s internal indicator describes the whole journey.

Recovery becomes more intelligible when we measure those connections. A set of successful projects is valuable. A life that can depend on them working together is the stronger outcome.

17. Improvement is not the same as impact

A programme starts. Conditions improve. Did the programme cause the improvement?

The sequence is encouraging, but it is not an answer. Other things may have changed: security, prices, migration, weather, private investment, household adaptation or another intervention. The task of causal evaluation is to distinguish the contribution of the programme from changes that might have occurred anyway.

The World Bank’s DIME guidance explains why a simple before-and-after comparison or a comparison of participants with non-participants can be misleading. Its discussion of difference-in-differences compares changes in a treatment group and a comparison group, with interpretation dependent on assumptions about how their outcomes would have evolved without treatment. Earlier parallel movement can help assess plausibility, but cannot prove the counterfactual. 12

Consider an original numerical exercise. Two otherwise broadly comparable areas each have a school-attendance rate of 50 per cent before a transport intervention. The intervention area later reaches 75 per cent. The comparison area reaches 65 per cent. The first area improved by twenty-five percentage points. The second improved by fifteen. The difference between those changes is ten percentage points.

That arithmetic is the beginning of an estimate, not the end of the argument. Interpreting the ten points as a programme effect requires a credible reason to believe that, without the intervention, the first area would have experienced the same change as the comparison area. The analyst must examine threats to that belief.

Suppose the intervention area also receives a major school-fee subsidy. The additional improvement may not belong entirely to transport. Suppose families with a strong preference for education move into that area. The population has changed. Suppose the comparison area’s school closes temporarily. Its weaker trend may exaggerate the apparent advantage of the intervention. Each possibility changes the meaning of the same arithmetic.

The appropriate response is not to discard comparison. It is to design it carefully and explain the assumptions. Multiple earlier observations, knowledge of implementation, records of other changes and sensitivity checks can strengthen an argument. No technique makes contextual understanding unnecessary.

Random assignment can sometimes create a stronger basis for comparison, but it is not a licence to withhold urgent rights or essential protection. Ethical review, genuine uncertainty about alternatives, informed participation where relevant, safety and operational feasibility must guide the design. This article does not propose randomising life-saving support to create a cleaner statistic.

Even in an ethically appropriate trial, measurement can go wrong. The assigned intervention may not be delivered as intended. People may move between groups. The intervention may affect nearby comparison areas. Follow-up may miss different kinds of people in each group. Random assignment addresses a particular problem of comparison; it does not solve every problem of data quality, interpretation or generalisation.

Often a full causal design is not available. A programme may begin during an emergency, the population may be moving, and access may be constrained. The evaluator can still assemble a disciplined account of contribution. Was the intervention delivered? Did the expected intermediate change occur? Did the timing fit? Were alternative explanations examined? Did outcomes differ where the relevant mechanism was absent? Are accounts from different sources consistent with the proposed pathway?

Such an account can be useful without pretending to produce a precise causal estimate. The report should distinguish “consistent with a contribution”, “strong evidence of an effect under the stated design”, and “observed improvement with attribution unresolved”. These are different strengths of claim, not a ladder on which every project must automatically reach the top.

The decision being made also matters. A municipality choosing whether to repair an obviously failed pump does not need to wait for an elaborate impact evaluation to establish that broken equipment cannot provide its intended function. But a claim that a national repair programme caused lasting social trust requires much more evidence. The burden of proof should match the breadth and consequence of the claim.

This distinction helps allocate evaluation effort. Spend the strongest causal work on important uncertainties: expensive policies being considered for expansion, competing approaches with plausible differences, or interventions whose effects are easily confused with broader trends. Routine verification of delivery remains necessary, but it is not the same research task.

The fictional district’s transport programme therefore publishes two statements. First, attendance increased in the intervention area during the observation period. Second, the comparison-based analysis estimates an additional change under stated assumptions, with limitations concerning other local developments and follow-up. It does not turn every additional pupil present into a proven programme effect.

A reader should be able to ask what would have happened without the intervention and find an actual answer about the method, even if that answer is that the evidence cannot identify the counterfactual precisely. “The situation improved after we arrived” is a timeline. It becomes an impact claim only when the explanation survives a serious comparison with other possibilities.

18. A real lesson from Sierra Leone: one programme, different results

A useful recovery evaluation should be able to find success on one dimension and no demonstrated success on another without forcing the entire programme into a single verdict.

Research by Katherine Casey, Rachel Glennerster and Edward Miguel provides a concrete example. Their published evaluation of a community-driven development programme in Sierra Leone used random assignment and a pre-analysis plan. The authors reported positive short-run effects on local public goods and economic outcomes, but no evidence of sustained effects on collective action, decision making or the involvement of marginalised groups in the measures they studied. They also showed how selective use of indicators could have supported misleadingly different interpretations. These are findings about that intervention and study, not a verdict on every community programme. 13

The measurement lesson is precise. Material delivery, economic change and institutional transformation are not interchangeable outcomes. An intervention can produce useful public goods without durably changing who holds influence. A report that measures only the first cannot simply claim the second. A report that finds no lasting institutional effect should not erase the material gains that were observed.

To see why, return to our fictional district. A community project funds the repair of a meeting hall and requires broad participation in planning. Attendance at the project meetings increases. The hall is repaired. The grant is accounted for. These are facts the programme could verify.

Now ask a different question: did people who previously lacked influence gain a lasting role in local decisions? The required meetings may have allowed them to speak during the grant process. Whether their influence persists after the grant ends is an additional empirical question. Attendance is evidence of presence. It is not automatically evidence of power.

An evaluator might examine how later decisions are made, who proposes agenda items, whether concerns receive responses, and whether participation continues outside the funded process. The appropriate methods need contextual and ethical care. The conceptual point is that durable institutional change must be measured where institutions continue to operate, not inferred entirely from behaviour under temporary programme rules.

A pre-analysis plan addresses another problem: the temptation to decide what success means after seeing the results. Imagine a fictional project with thirty measured outcomes. Some rise, some fall and some change little. A report that selects only the largest improvements can make almost any intervention look coherent after the fact. Selecting only the disappointments can produce an equally misleading condemnation.

The answer is not to ban unexpected findings. Unexpected findings can be valuable. It is to distinguish the outcomes and hypotheses specified in advance from exploratory discoveries made later. Readers can then see which claims were planned tests and which require further investigation.

This distinction is especially important when outcomes are numerous and related. A programme may measure several versions of participation, trust, income, satisfaction and service use. The more opportunities there are to find an encouraging difference, the more important it becomes to explain the family of measures and the analysis choices. A technically correct calculation can still be selectively presented.

The fictional district therefore writes a short measurement plan before the next programme begins. It identifies the main public problem, the central outcomes, the expected pathway, the comparison strategy where feasible, and the most important risks. It records what would count as evidence against the intended mechanism. It also leaves room for exploratory work, labelled as such.

This plan need not freeze the programme in place. Conditions may change, and the programme may need to adapt. The discipline is to document revisions when they occur, rather than quietly replace an inconvenient objective with one that has already been achieved. Learning is compatible with a record of changed expectations. It is not compatible with rewriting history.

There is a human consequence to preserving mixed findings. Staff who delivered a useful service deserve recognition for that achievement. Communities deserve honesty about what did not change. Funders need to know whether the programme should be maintained, altered, expanded or stopped. A single ceremonial label—success or failure—can serve all three groups badly.

The Sierra Leone study is therefore valuable not because it supplies a universal formula for peacebuilding, but because it demonstrates the importance of separating kinds of result and constraining selective interpretation. Our fictional extension makes the same point in practical terms: building a hall, filling a meeting and changing local authority are three different achievements.

The better question is not whether an intervention can be made to look successful. It is which part of its promise the evidence supports, which part it does not, and what that difference teaches the next decision.

19. The households the survey could not reach

A survey can be carefully administered to the wrong slice of the population.

In the fictional district, an evaluation team interviews households near the main road. The interviews are consistent, the questions are clear, and the responses are entered accurately. Yet the remote settlements most affected by transport barriers are absent. The quality of data collection within the reached group does not make the group representative of everyone.

This is a coverage problem before it is an arithmetic problem. A larger sample drawn from the same restricted frame may reduce random variation while leaving the central bias untouched. Ten thousand responses from connected households do not automatically describe households without connectivity.

The first question should therefore be how the population was represented in the sampling frame: the list, map or other structure from which people were selected. Did it include temporary accommodation, new settlements and households outside existing registers? Were areas excluded for safety or access reasons? Were those exclusions stated publicly at an appropriate level of detail?

The second question concerns participation after selection. A household may be selected but unavailable, unwilling to answer or impossible to locate. These situations should not be treated as equivalent. The reason for non-response may be related to the outcome being measured. People with unstable housing may be harder to reach. People who distrust the programme may decline. People benefiting most may be more willing to remain in contact.

Consider a separate synthetic follow-up. One hundred households entered a programme. At a later assessment, seventy are reached, and fifty report the selected positive outcome. Among reached households, the rate is about 71.4 per cent. Across the original hundred, the number known to have the outcome is fifty. Thirty households are unobserved.

Without additional assumptions or evidence, the overall outcome rate could range from 50 per cent, if none of the unobserved households has the outcome, to 80 per cent, if all thirty do. The bounds are deliberately simple. They are not a confidence interval and do not account for inaccurate responses. They demonstrate how much the missing observations alone could matter.

The report should not choose the favourable end of the range because the programme seems promising. It should not automatically choose the unfavourable end because caution sounds rigorous. It should report what is known, explain what is missing, and investigate whether further evidence can narrow the uncertainty without imposing unreasonable risk or burden.

Statistical weighting can help when the design and information support it. It cannot magically recover the experience of a group that is absent in ways the available data cannot describe. A report should explain what the weights adjust for rather than present weighted results as though the sampling problem has disappeared.

There is also a distinction between sampling uncertainty and systematic error. Sampling uncertainty concerns variation arising from observing a sample rather than the whole population under a specified design. Systematic error can arise from an incomplete frame, misleading questions, inaccurate records or selective non-response. A narrow confidence interval does not protect a result from a systematic problem built into the measurement.

For a general reader, the practical question is simpler: does the stated uncertainty describe all the important ways the number could be wrong, or only the random variation the statistical model includes? A technically correct interval can be misunderstood when its limited scope is not explained.

Small samples should not be dismissed automatically. A few carefully investigated cases can reveal a mechanism that routine statistics missed, such as a document requirement excluding a particular group. They cannot establish how common the problem is without further evidence. Mechanism discovery and prevalence estimation are different tasks. A strong evaluation can use both while keeping their claims separate.

The fictional district therefore designs a mixed approach. A suitable sample provides an estimate for a defined population. Targeted, protected inquiry examines groups likely to be missed and investigates specific barriers. Administrative records provide repeated operational information. None is treated as a substitute for every other source.

When safety prevents coverage, the report states the boundary. It may say that the evidence describes accessible areas and cannot support a district-wide estimate. That limitation is not a minor footnote if the inaccessible areas may have very different conditions. It belongs near the headline conclusion.

The most consequential missing data may be the data about people who do not trust the process enough to participate. Their absence cannot be interpreted simply as indifference. It may itself be a reason to reconsider how the evaluation approaches communities.

A survey becomes more credible when it can describe the edge of its own vision. Recovery measurement should not make the people hardest to reach disappear a second time—first from services, then from the evidence used to declare those services restored.

20. Three sources are not always three independent witnesses

Triangulation sounds reassuring: compare several sources and see whether they agree. The method is useful only when we understand how the sources were produced.

In the fictional district, a donor report states that a service reached 5,000 people. A ministry presentation repeats the figure. A newspaper quotes the presentation. Three documents contain the same number. There may still be only one underlying source, and it may be an estimate prepared before delivery.

Repetition is not independent corroboration.

A useful evidence trail therefore asks where a claim first entered the record and how later documents used it. Was the figure directly observed, reported by an implementing organisation, estimated from planned capacity, or copied from another source? Agreement between documents can reflect a common origin rather than separate verification.

The same issue appears in digital material. Several photographs may show the same facility on the same day. They can provide different views of that event without establishing performance over time. Several resident accounts may derive from one rumour. Conversely, two apparently contradictory records may concern different dates or different populations and both be accurate.

The evaluator should begin by separating the claim into components. “The clinic is open every weekday” contains a claim about a facility, a schedule and repeated operation. A photograph may support the facility’s existence and its condition at one moment. A staff roster describes intended coverage. Service logs describe recorded activity. User accounts can describe actual access. Each source has a role, but none should be promoted beyond what it observes.

Disagreement is therefore not automatically evidence of deception. Suppose logs show that a records office was open while residents report being unable to obtain a service. The office may have been physically open but unable to process that application type. The disagreement reveals a definition problem. Clarifying “open for what?” can be more useful than deciding which side is lying.

There can also be honest differences in time. A household describes repeated failure last month, while a site visit confirms a recent repair. The report should not use today’s functioning condition to erase last month’s experience. Nor should it use last month’s failure to deny today’s improvement. A timeline can reconcile apparently competing accounts.

Evidence should be graded by relevance as well as reliability. An exceptionally reliable record of expenditure may be weak evidence for learning outcomes. A carefully verified building image may be strong evidence of physical damage but weak evidence of population movement. The question is not whether the source is generally impressive. It is whether the source can support this particular claim.

For the fictional district, an evidence table might contain a claim, its source, observation date, population or location, method, known limitations and the decision it informs. Such a table need not be published in full where information is sensitive. Its purpose is to prevent the final narrative from losing the distinctions present in the underlying records.

Conflicting evidence should trigger a specific next question. If administrative records show payments completed but households report non-receipt, investigate the stages between payment instruction and accessible receipt. If school attendance records improve while learning does not, examine instruction, assessment comparability and the composition of learners. If complaints increase after a new reporting channel opens, examine whether access to reporting improved before interpreting the increase as worsening service.

This is a more productive use of contradiction than merely placing opposing claims side by side. The objective is to identify what observation could distinguish the plausible explanations. Sometimes the answer will require a small targeted check. Sometimes the evidence will remain inconclusive. Both outcomes are better than manufacturing agreement through selective omission.

Independent review matters, but independence should be described honestly. A colleague rereading the report may catch errors without being independent of the programme. An external evaluator may still rely on data collected by the implementer. A financial auditor may not have assessed outcomes. These forms of scrutiny can all add value, but they should not be described as equivalent guarantees.

The district’s final report therefore distinguishes internally checked figures, externally verified records, sampled observations and self-reported experiences. It explains where several sources share the same origin. It does not label the entire assessment independently validated because one component received an external check.

The purpose of triangulation is not to surround a preferred conclusion with enough documents to make it look inevitable. It is to discover which parts of the conclusion survive different ways of observing the world. A disagreement that changes the next question may be more valuable than ten reports repeating the same unexamined total.

21. What satellites and artificial intelligence can—and cannot—see

A satellite image can reveal a pattern unavailable from a single street. It cannot tell the whole story of the people beneath it.

Remote evidence can be particularly useful when direct access is limited. It can help identify visible changes, compare areas over time, and direct attention toward places requiring further assessment. The mistake is to treat a remotely observed proxy as though it directly measures the social outcome of interest.

Night-time light provides a clear example. World Bank researchers have used it as a proxy for economic activity. In a published analysis of Afghanistan, they specifically separated the influence of foreign military installations when interpreting changes in the country’s night-time brightness. The example illustrates why the meaning of a light signal depends on what produces it; it does not justify equating brightness with household welfare. 14

In our fictional district, newly installed streetlights increase measured brightness. That is evidence consistent with the operation of those lights during observation. It may also indicate improved public infrastructure. It does not establish that household incomes increased, that all homes have reliable power, or that residents feel safer using the street.

Conversely, a decrease in brightness could reflect the departure of a brightly lit compound rather than a proportional fall in every part of the local economy. The interpretation requires knowledge of the source and context of the signal. An attractive map does not remove that requirement.

The same caution applies to visible construction. A roof can be identified from imagery while the building remains empty. A road can appear repaired while an administrative restriction or unaffordable transport prevents use. Physical access, legal access and economic access are different layers. Remote observation may reveal one while missing the others.

Artificial intelligence can assist with classification, text comparison, anomaly detection and the organisation of large datasets. Those tasks can save effort. The output still needs to be understood as a model result produced from particular inputs, not as an independent fact merely because software generated it.

Consider an illustrative image-classification system trained to recognise damaged buildings. Its performance may differ between dense urban areas and scattered rural structures, between roof types, or between the imagery used for training and the imagery available after a new event. A single overall accuracy figure can hide errors concentrated in the areas most important to the assessment.

The relevant evaluation should therefore examine error types. A false indication of damage may direct resources toward a building that does not need that response. A missed damaged building may leave a household outside the assistance process. The relative consequences depend on how the output is used. A model suitable for flagging cases for human review may be unsuitable for making final eligibility decisions.

The district should also preserve the distinction between observed and inferred data. If a model estimates conditions for an inaccessible area, the public map should not display those estimates as though enumerators visited every household. The source, method and uncertainty belong with the result. A smooth visual surface can conceal a very uneven evidence base.

Language models create an additional risk because they can produce a fluent summary from incomplete records. A sentence may sound more certain than the documents supporting it. The safest analytical use is to keep summaries linked to their sources, check consequential claims, and prevent the model from filling gaps with plausible narrative. Fluency is a property of the writing, not a measure of evidential strength.

Automated deduplication needs similar restraint. Two records with similar names may refer to one person, or to different people. Merging them can alter eligibility, payment counts or household histories. The appropriate process depends on the consequences of error, the quality of identifying information and the availability of review. A high-confidence match score is not a legal determination of identity.

These cautions do not imply that technology should be avoided. They imply that the task should be defined clearly. Use remote imagery to identify visible changes. Use models to prioritise review where their limitations are understood. Use text tools to organise evidence without silently changing its meaning. Keep decisions affecting rights or safety within the relevant accountable professional and legal processes.

The fictional district treats remote and automated outputs as layers in the evidence system. It tests a sample of classifications where safe, reports uncertainty by relevant conditions, records changes to the model, and checks whether errors fall unevenly across communities. It does not call the resulting map a complete picture of recovery.

Technology becomes most useful when it expands what people can investigate without shrinking their curiosity about what remains unseen. A satellite can help locate a repaired roof. It cannot decide whether the child across the river has regained an education.

22. Listening is a method, not a decorative quotation

A well-chosen account from a resident can reveal something a large dataset missed. It can also be misused to make a report feel human without changing its conclusions.

The difference lies in the question being asked and the claim being made.

In the fictional district, a resident describes visiting three offices to correct a record. This account can identify a possible failure in the service journey. It does not establish how often the failure occurs across the district. The evaluator should investigate the mechanism, compare it with relevant records and, where necessary, gather broader evidence. The account matters because it changes the inquiry, not because it provides an emotional opening paragraph.

A public meeting offers another example. The room is full, participants speak, and the programme reports strong consultation. But who could attend? Who understood the language? Who felt able to disagree? Who was absent because of work, care, distance or fear? A meeting count cannot establish inclusive influence.

Listening should therefore be designed around the people and decisions involved. Different groups may need different opportunities to contribute. A person may be willing to describe an access barrier privately but not in a public room. Another may prefer a trusted intermediary. The process should not require people to disclose sensitive experiences in order to prove that consultation occurred.

Consent also has a practical meaning. A participant needs to understand what the conversation is for, how the information may be used, and whether declining will affect assistance or another relationship. Where there is dependency, the assurance that participation is voluntary should be supported by the actual process, not merely read aloud as a formula.

This chapter offers general principles for respectful evaluation, not a specialist protocol for interviewing survivors of violence. Sensitive research requires qualified design, safeguarding and the relevant ethical and legal review. An evaluator should not attempt to obtain a dramatic account simply because a report would benefit from a powerful story.

The choice of question affects the evidence. “How did this programme help you?” assumes help occurred. “What changed, if anything, after the programme began?” allows a wider answer. “Are you satisfied?” can be useful for a defined purpose, but it may reveal expectations, gratitude or fear as well as service quality. A more specific question about what happened during the last attempt to obtain a service may be easier to interpret.

Even a specific account has limits. Memory can be uncertain, dates can be approximate, and people can interpret the same event differently. These limitations do not make testimony worthless. They make it important to distinguish direct experience, inference and information heard from someone else. The evaluator should preserve uncertainty rather than tidying it into a more cinematic narrative.

Listening can also reveal successful adaptation that a programme did not create. A household may solve a transport problem through relatives, neighbours or private arrangements. The recovery report should recognise that agency without automatically attributing it to the intervention. People are not passive recipients waiting for institutions to act; the evaluation must leave room for what they do themselves.

There is a corresponding danger in romanticising adaptation. A family’s ability to cope with a broken service does not prove that the service is adequate. Coping may consume savings, children’s study time or a caregiver’s rest. Asking how the household managed, and what that management cost, can reveal the hidden burden behind an apparently successful outcome.

Qualitative analysis should look for patterns and differences, not merely select the most quotable lines. How do accounts vary by location or stage of the process? Which barriers recur? Which cases contradict the main explanation? Are there plausible reasons why some people experience the service differently? These questions turn listening into analysis.

The fictional district records a finding that no dashboard had shown: applicants with a particular document problem are being redirected between offices because neither office recognises responsibility for the next step. A small number of interviews uncovered the mechanism. Administrative review then establishes how the process works and how many cases are affected. The qualitative evidence and the operational evidence strengthen one another without making identical claims.

The evaluation should also close the conversation. People who give time to an assessment should not encounter only silence afterwards. Where appropriate and safe, findings and planned responses should be communicated in an accessible form. This does not require promising that every request will be granted. It requires showing that information entered a decision process rather than disappearing into a report.

Listening becomes a serious method when it can change what the programme believes about itself. A resident’s account is not there to decorate the numbers. It can reveal the question the numbers were never designed to answer.

23. The evidence must not create a new danger

Better measurement is not achieved by collecting everything that might someday be useful.

A postwar dataset can contain names, addresses, family relationships, displacement histories, disability information, financial circumstances and accounts of disputes. Each field may appear reasonable in isolation. Combined, they can reveal far more about a person than the original purpose requires.

OCHA’s revised Data Responsibility Guidelines describe a minimum approach to safe, ethical and effective management of operational data and align that work with wider humanitarian and United Nations data-protection guidance. The importance for recovery measurement is straightforward: data quality and data responsibility belong together. A highly accurate dataset can still be handled in a way that creates harm. 15

The fictional district begins with a simple test for each proposed field: which decision requires this information? If nobody can answer, the field should not be collected merely because it is easy to add to a form. A measurement system should have a purpose narrow enough to explain what it does not need.

The next question is who needs access. A service planner may need aggregate information about unmet need in an area. A caseworker may need identifiable information to resolve an individual problem. A public report usually requires neither the complete case file nor a map precise enough to identify a vulnerable household. Different purposes require different access arrangements.

Removing names does not necessarily make data anonymous. A small combination of location, age, household structure and an unusual event can identify someone to people who know the community. The risk depends on context. A public dataset should therefore be assessed for what can be inferred from combinations, not merely checked for an obvious name column.

Linking records can improve measurement and service delivery, but it also expands consequences. A social-support register, a dispute record and a movement history may each have been collected for a separate purpose. Combining them can produce a new profile that participants did not expect. The lawful basis, necessity, safeguards and accountability for linkage should be established rather than assumed.

The district also distinguishes assistance from evaluation participation. A household should not be led to believe that a favourable answer is necessary to continue receiving support. Separating case decisions from feedback collection can help, but the details depend on the programme and the applicable rules. What matters is whether the participant can reasonably understand and trust the distinction.

There is a practical issue of retention. Data needed for an immediate operational decision may not need to remain identifiable indefinitely. Other records may have legal or accountability retention requirements. The policy should specify the purpose, retention period, access and deletion or archival arrangements, with the appropriate professional and legal input. “Keep everything forever” is not a complete preservation strategy.

Security also concerns ordinary handling. A well-protected central database can be undermined by an uncontrolled spreadsheet export, a lost device or an emailed attachment sent too widely. Procedures must follow the data through its actual use. Training is useful only when staff have workable tools and clear responsibilities.

The evaluator should avoid exposing sensitive operational details through public evidence. A report can explain that some areas were inaccessible without publishing information that creates new risks for staff or residents. It can describe a service gap at an appropriate geographic level. Transparency should reveal methods and limits without assuming that every raw record belongs in public.

There is also a right to correction where the relevant framework provides it. An inaccurate record can affect eligibility, referrals or public decisions. A responsible system needs a way to receive and resolve corrections, and to propagate them to authorised linked systems where appropriate. Otherwise the same error can survive in several places after the original record has been fixed.

In the fictional district, an evaluator proposes publishing a household-level map of combined deprivation. The map is analytically interesting. After review, the team recognises that small areas and unusual household characteristics could make families identifiable. It publishes a less granular map, retains protected analysis for authorised planning, and explains the limitations of the public version.

The result is not less honest. It is a different form of disclosure suited to the risk. Readers can still see where unmet need is concentrated and how estimates were produced without receiving information they do not need about individual households.

The purpose of recovery measurement is to improve people’s ability to live safely and use institutions. Evidence practices that undermine that purpose are not redeemed by statistical sophistication. The best dataset is not the largest one. It is the one that supplies the necessary knowledge while keeping its collection and use proportionate to the people whose lives it describes.

24. When a target becomes a performance

An indicator does more than describe behaviour. Once rewards, reputation or funding depend on it, the indicator can influence behaviour.

In the fictional district, the records office is given a target for applications closed each week. Staff work faster. The closure total rises. Then residents begin reporting that complicated applications are being rejected early because rejection counts as closure and consumes less time than resolution.

The target has improved the measured output while weakening the intended service.

This does not mean targets should be abandoned. Clear expectations can support responsibility and expose delay. The design problem is to keep the measured event close enough to the public purpose that improving the number usually requires improving the service.

A single indicator is rarely sufficient for a consequential decision. Closure volume can be paired with reasons for closure, repeat applications, overturned decisions and the age of unresolved cases. These measures do not make manipulation impossible. They make some forms of distortion more visible.

There is a trade-off. Adding dozens of indicators can create reporting overload, conflicting incentives and new opportunities for selective presentation. The answer is not infinite measurement. It is a small, coherent set connected to the service’s main failure risks.

For example, the fictional records office tracks completion time for straightforward cases, the age and reasons for delayed complex cases, and whether residents can use the resulting document in the next relevant process. It separately records rejected applications and the reasons for rejection. A fall in delay is no longer allowed to hide inside an increase in exclusions.

Targets can distort selection before delivery begins. A programme rewarded for employment outcomes may recruit participants already most likely to find work. Its measured success rises while people facing greater barriers receive less support. The report should examine who entered the programme, who did not, and whether the selection matches the stated purpose.

A construction programme can face a similar incentive. It may choose the easiest repair sites to meet a completion target quickly. That can be sensible when rapid restoration is a legitimate objective. It becomes misleading when the result is presented as evidence that the most urgent needs were addressed. The target should state the priority it serves.

Timing offers another opportunity for distortion. A site can be prepared for a scheduled visit. A service can temporarily receive extra staff during inspection. Such preparation may demonstrate what the system can do under support, but not what it routinely does. Repeated observations, ordinary operating records and appropriate independent checks can help distinguish demonstration from sustained performance.

Satisfaction scores are also vulnerable to context. People may express gratitude for a modest service because expectations are low. They may fear criticising an institution on which they depend. Alternatively, expectations may rise as recovery progresses, producing lower satisfaction even while objective service conditions improve. Satisfaction is useful evidence about experience, not a substitute for every other measure of quality.

The fictional district notices another pattern: reports always compare current performance with the worst month in the available record. The comparison is not fabricated, but it is selected to maximise improvement. A more transparent report uses a justified reference period, shows a longer trend where available, and explains unusual events rather than choosing the most convenient starting point.

Programme adaptation introduces a subtler risk. If the original target becomes impossible or irrelevant, revising it may be responsible. But the change should be recorded with its date and reason. Quietly replacing an unmet target with a different achieved one turns adaptation into retrospective self-approval.

The same principle applies to missing data. An unmeasured outcome should not be assigned a favourable colour because staff believe the project is working. Nor should it automatically receive a zero that conflates ignorance with failure. “Not yet assessed” is a legitimate status when accompanied by a plan or an explanation of why assessment is not feasible.

A healthy reporting culture allows staff to surface bad news without assuming that every problem proves misconduct. Otherwise the organisation teaches its people to protect the indicator rather than improve the service. Accountability should distinguish honest uncertainty, implementation difficulty, avoidable error and deliberate misrepresentation. These require different responses.

The report itself can help create that culture. It can state what improved, what did not, what remains unknown, and what action follows. It can preserve earlier versions rather than erase inconvenient changes. It can include evidence that challenges the main explanation.

Measurement becomes a performance when the organisation’s first concern is how the result looks. It becomes a learning system when an unfavourable result can change the work. The district should not aim to make every indicator green. It should aim to make each important indicator truthful enough that the next decision becomes better.

25. A district review that refuses a false score

We can now bring the pieces together in one original worked review. Every number in this chapter is synthetic. The exercise is not a model of any real district and does not supply international standards. Its purpose is to show how a set of individually correct figures can be turned into a more useful account of recovery.

Imagine the river district at the end of a reporting year. The review covers a defined cohort of 1,000 households and a separately defined cohort of 600 school-age children. The cohorts are kept explicit for the exercise; a real review would need to address arrivals, departures, deaths, changes in household composition and the relevant population definitions.

The programme’s first draft contains a confident summary: water restored, schools reopened, health facilities operational, transfers delivered and the justice backlog reduced. None of these phrases is necessarily invented. Each is broader than the evidence presently assembled.

The review team replaces the summary with a table that keeps the observations and limits together.

Area	Synthetic observation	What the observation supports	What it does not yet establish
Water	900 of 1,000 cohort households assessed; 630 meet the locally specified combined service conditions	70% of assessed households, and 63% of the full cohort, have verified evidence of meeting the conditions	Conditions for the 100 unassessed households; performance outside the observation period
School participation	450 of 600 cohort children attend regularly under the stated definition	Regular participation is documented for 75% of the cohort	Learning, inclusion of every absent child, or continued participation after temporary support ends
Learning	420 children assessed; 252 demonstrate the selected skill on a defined task	60% of assessed children demonstrate the skill	Skills of the 180 unassessed children; improvement without a comparable earlier assessment
Health facilities	Four facilities open; three meet the specified readiness requirements for a defined service package	Availability and the observed readiness status	Quality of care, access for all residents or health outcomes
Transfers	1,000 payments due; provider records 900 paid; full usable receipt confirmed for 840	Provider completion for 90%; confirmed full receipt for 84%	Receipt status for 60 provider-paid cases not yet verified; completion of 100 failed cases
Justice workload	1,000 pending at start; 200 new cases; 250 closures; 950 pending at end	Net backlog reduction of 50 under the stated accounting	Fairness of every closure, implementation of decisions, or improvement for the oldest cases

The table is not the final evaluation. It is the beginning of a disciplined conversation.

Start with water. Of the 900 assessed households, 630 meet the combined conditions and 270 do not. One hundred are unassessed. The known share meeting the conditions in the full cohort is 63 per cent. If every unassessed household meets them, the full-cohort share would be 73 per cent. If none does, it remains 63 per cent. These are missing-data bounds, not statistical confidence intervals or estimates of future performance.

The immediate question is why 270 assessed households fail the combined conditions and why 100 were not assessed. The first is a service question; the second is an evidence question. They may overlap, but the review should not assume they are the same. An inaccessible area might have poor service, but that remains a hypothesis until supported.

Now consider schooling. Regular participation is documented for 450 children. That is an important result. The learning assessment reaches 420, of whom 252 demonstrate the selected skill. The 60 per cent proficiency figure concerns assessed children. Across the full cohort, 42 per cent have demonstrated evidence. The skill status of the rest cannot be inferred simply from absence on assessment day.

The district also lacks a comparable learning baseline. It can describe current assessed performance. It cannot honestly claim a measured learning gain from these data alone. The correct next step is not to invent a baseline from memory. It is to use the current evidence for teaching, examine access barriers, and establish a suitable comparison for future measurement.

The health observation is similarly limited. Four facilities are open, and three meet the specified readiness requirements for the chosen package. It would be wrong to average these observations into a claim that healthcare is “87.5 per cent recovered”. The two fractions concern different conditions, and neither establishes quality or access. The facility failing readiness needs a qualified service response; the broader access and quality questions require the appropriate evidence.

Transfers show why financial and household records need reconciliation. Assume each of the 1,000 households is due 100 units in this synthetic payment cycle. The total due is 100,000 units. The provider marks 900 payments as paid, representing 90,000 units. Full usable receipt is confirmed for 840 households, representing 84,000 units. Sixty provider-paid cases remain unverified, while 100 payments failed in processing.

The 6,000 units associated with the sixty unverified cases should not automatically be called lost, stolen or received. Their status requires investigation. The 10,000 units associated with failed payments are a distinct operational problem. Combining both into a generic 16 per cent “leakage” figure would convert uncertainty and processing failure into a claim the evidence does not establish.

The justice arithmetic is exact within the exercise: 1,000 plus 200 minus 250 equals 950. The backlog has fallen by fifty cases. Suppose, however, that cases older than the review’s stated age threshold increase from 300 to 320. Total pending work declines while the long-waiting group grows. A report that celebrates only the falling total misses the bottleneck affecting those cases.

What should the district do first?

There is no answer in the table alone. A priority decision needs information about severity, urgency, affected people, feasible responses, cost and responsibility. The water gap might have serious consequences requiring prompt qualified attention. Failed transfers might be straightforward to correct and essential to households. The old cases might concern urgent rights. The review must connect the measures to real decisions rather than rank incomparable percentages.

One temptation is to assign weights to the six areas and publish a single recovery score. That may be useful for a carefully justified purpose, but it introduces choices that must be visible. How much does one point of school participation compensate for a water-access failure? Should missing health-quality evidence reduce a score, remain unscored or block a particular conclusion? Who decided? A mathematical average cannot settle these value questions.

For this fictional review, the team chooses not to produce a single score. It publishes separate domain statements, highlights material risks, and marks unassessed questions explicitly. It also assigns a responsible institution and a next observation to each unresolved issue. This is a design choice for the example, not a claim that every composite index is invalid.

The revised summary reads differently: “The district has restored substantial physical and administrative capacity. Verified service access remains uneven. Regular school participation is documented for three quarters of the defined child cohort, but learning improvement has not yet been measured comparably. Transfer processing has advanced further than confirmed household receipt. The total court backlog has declined, while older pending cases have increased. Several consequential evidence gaps remain.”

That paragraph does not tell residents that recovery has failed. It tells them where progress is established and where further work is necessary. It also gives managers a reason to act rather than merely defend the original announcement.

The review’s most important achievement is not a new table. It is the refusal to convert six partial truths into one unsupported conclusion. When a district can do that, measurement begins to serve recovery instead of serving the report.

26. Turn a finding into a correction—and then test the correction

A finding that changes nothing is only a finding.

A recovery review becomes useful when it identifies an action, gives that action an accountable home, and specifies what evidence will show whether the response worked. This is not the same as demanding a new project for every disappointing indicator. Sometimes the necessary change is a clarification, a corrected record, a repaired connection or a different allocation of existing effort.

Take the fictional transfer programme. The review distinguishes 100 failed payment cases from sixty provider-paid cases whose receipt is unverified. The two groups should not receive one undifferentiated instruction to “improve delivery”. The failed cases require investigation of processing and eligibility or channel problems. The unverified cases require appropriate reconciliation and, where safe and proportionate, confirmation of what happened after the provider’s recorded payment.

A useful action note contains the problem, the decision, the responsible institution, the resources or authority required, the next check and the condition for further escalation. It should also state what the evidence does not justify. For the sixty unverified cases, the note should avoid alleging wrongdoing before the facts support that conclusion.

The next check must match the original problem. Sending another instruction to a payment provider is an activity. A provider reporting that a file was processed is an output. Confirming that the intended household can use the full authorised amount addresses the public-facing problem. The review should not allow the corrective action to stop at the same incomplete measurement that produced the original confusion.

The same reasoning applies to the school. If absence is concentrated among children facing transport barriers, another delivery of desks does not test the relevant explanation. A suitably designed access response should be followed by evidence about participation among the affected children. If participation improves but learning does not, the next inquiry changes. It should not keep repeating the first repair simply because that repair was visible and successful at its own stage.

Correction is therefore a sequence of hypotheses and observations. “We believe this barrier is preventing the outcome. We will address it through this lawful and feasible action. We will observe whether the expected intermediate change occurs. If it does not, we will reconsider the explanation.” This is ordinary disciplined reasoning, not a guarantee that a complex system can be controlled perfectly.

The timetable should reflect the mechanism. A clerical correction may be observable quickly. A learning intervention needs sufficient teaching and a fair assessment window. A maintenance change may need to be observed through the conditions that previously produced failure. Imposing the same reporting interval on every problem can reward what changes quickly and neglect what matters over a longer period.

A hypothetical review schedule might include an early operational check and a later durability check. These are examples of programme design, not reminders or work being scheduled in this conversation. The early check asks whether the correction was implemented and whether the immediate barrier changed. The later check asks whether the benefit continued when temporary attention or support receded.

There is also a cost to verification. Repeated surveys consume staff and residents’ time. The review should choose the least burdensome evidence adequate for the decision, not the most elaborate study it can imagine. Existing records may answer an operational question. A targeted inquiry may resolve a discrepancy. A major expansion decision may justify stronger independent evaluation.

The OECD’s guidance on evaluation in conflict and fragile settings is explicitly concerned with tailoring learning and accountability to such contexts and connecting evaluation to programme design and management. The practical lesson for this guide is that evaluation should inform decisions under real constraints, rather than become a detached ritual conducted after the important choices have already been made. 16

Corrective decisions should themselves leave a record. What was changed? Why? On whose authority? Which assumptions were revised? If a later outcome improves, the record helps distinguish the original intervention from the adaptation. If the outcome worsens, it helps identify what was tried without requiring staff to reconstruct the sequence from memory.

The record should also preserve dissent. A local team may believe the proposed correction misunderstands the barrier. Their explanation deserves examination before implementation, not merely a footnote after failure. Disagreement can expose a missing condition, an unrealistic assumption or an unintended consequence.

For the fictional district, the review meeting ends with fewer claims but clearer tasks. The transfer discrepancy is divided into two investigations. The school’s access barrier receives a specific response. The court’s older cases are examined under lawful prioritisation rules. The water evidence gap is separated from the service gap. Each action has a next observation capable of changing the decision.

That is what it means for measurement to become a learning loop. The point is not to prove the programme was right the first time. It is to make the next version less wrong, more useful and more accountable to the people it is meant to serve.

27. Measure what remains after the project leaves

A project can be successful during the period in which it receives exceptional attention. Sustainability asks a harder question: what remains when that attention changes?

The question is not whether every external contribution disappears. Some services legitimately depend on continuing public finance, partnerships or long-term support. Nor is sustainability a demand that vulnerable people stop needing assistance. The relevant issue is whether the benefits have a credible basis for continuation under the arrangement intended to support them.

In the fictional district, a clinic functions because a temporary programme pays staff, maintains equipment and supplies administrative support. Its current operation is real. A handover plan that transfers only the building does not transfer the capability. The receiving institution needs lawful authority, people, funding, knowledge, records and workable arrangements for maintenance and supply.

The same applies to a digital registry. Delivering the software does not establish who will manage access, correct errors, pay hosting costs, preserve backups or train replacement staff. A system can be technically complete and institutionally homeless.

A useful sustainability assessment therefore follows the operating arrangement rather than merely asking whether a memorandum was signed. Who is responsible for each necessary function? What resources are committed? Which skills exist in more than one person? Which contracts or procedures continue? What happens when a routine problem occurs?

The evidence should distinguish confirmed arrangements from plans. A proposed budget line is not an approved appropriation. An approved budget is not necessarily cash available on the required date. A trained officer is not evidence that the institution can retain or replace that capability. These distinctions should not become reasons to reject every handover. They identify what still needs to be made credible.

Duration matters. A service observed for a short period after handover can demonstrate early continuity. It cannot prove indefinite sustainability. A report should use language such as “continued through the stated follow-up period” or “continuation is supported by the following arrangements” rather than treating the future as already observed.

The relevant follow-up period depends on the service. Some weaknesses appear at the next payroll cycle. Others appear during a different season, a procurement renewal or staff turnover. A review should ask when the supporting arrangement is likely to be tested, not simply choose the nearest convenient anniversary.

Households also have a role in judging durability. Do they make fewer emergency arrangements? Can they plan work or schooling with greater confidence? Do they still expect to pay an unofficial intermediary when the ordinary process fails? These observations cannot replace administrative evidence, but they can reveal whether formal continuity has become practical dependability.

A service may remain available while transferring hidden costs to residents. A temporary subsidy ends, and the fee rises. A local organisation takes over but relies on unpaid work that is not sustainable for its members. A facility stays open because staff purchase supplies from their own income. The report should not label the service self-sustaining while ignoring the burdens holding it together.

There is a converse mistake: assuming that outside support proves failure. Some public goods require continuing finance and coordination. The test is whether the arrangement is legitimate, sufficiently reliable and suited to the intended benefit—not whether it can be described as independent of everyone else. No complex public system operates without dependencies.

Sustainability also includes the ability to detect and correct deterioration. A service that can identify a problem, explain responsibility and obtain a response may be more durable than one performing slightly better today but lacking any repair process. The measurement should therefore examine whether ordinary failures can be managed, not only whether the last observation was favourable.

The fictional school illustrates the point. Its transport subsidy is ending. The district does not claim sustainability because the classrooms are currently full. It checks whether a replacement arrangement is authorised, whether families can use it, and whether attendance remains stable after the transition. If the arrangement fails, the earlier attendance gain was still real. The sustainability claim must change.

Handover should therefore be treated as a transfer of responsibilities and capabilities with a subsequent test, not a ceremonial endpoint. The final report can state what has been transferred, what remains supported externally, what risks persist and when the next meaningful observation will occur.

Recovery is durable not when nothing can go wrong, but when ordinary institutions and communities can respond without rebuilding the entire emergency apparatus. Measuring that capacity requires looking past the project’s closing date. The end of a contract is an administrative fact. The continuation of a usable life is the more demanding evidence.

28. The report a resident should be able to read

After all the methods, calculations and caveats, a recovery report still has to communicate. A document that is technically careful but impossible to understand can fail the people whose lives it describes.

The answer is not to remove every qualification. It is to organise the account around the questions a reasonable reader would ask.

What changed? Who experienced the change? How was it measured? Compared with what? What remains difficult? What is uncertain? What will happen next because of the findings? A clear report should make these questions answerable without requiring the reader to reconstruct the whole project archive.

The opening should state the strongest conclusion the evidence supports, not the strongest conclusion the organisation hoped to reach. If construction is complete but service outcomes remain unmeasured, say that. If access improved for one group but not another, say that. If a causal estimate depends on assumptions, explain the important ones in ordinary language and provide the technical detail separately.

For the fictional district, compare two possible statements. The first says: “The recovery programme has transformed the district, restoring essential services and rebuilding community confidence.” The second says: “The programme restored specified facilities and improved verified access to several services. Some households still face barriers, evidence of learning improvement is incomplete, and continued operation depends on arrangements now being tested.”

The second statement leaves less room for celebration. It leaves more room for responsible action. It does not confuse the absence of a final answer with the absence of progress.

A resident should also be able to see the date behind the result. “Current” is not a date. Data collected six months ago may remain the best available evidence, but the report should not make it sound like a live observation. The publication date and the observation period belong together, especially where conditions can change quickly.

Numbers should carry their units and populations. Households are not people. Return events are not unique returnees. A percentage of applicants is not a percentage of all eligible residents. A model estimate is not a direct count. These distinctions can be made clearly without overwhelming the reader.

The report should show important limits near the claims they qualify. Hiding exclusions in an appendix while presenting an unqualified headline is not transparent communication. A short scope sentence can do more for honesty than several pages of generic methodological language.

It should also distinguish an observed failure from an unknown outcome. A service assessed and found unusable requires one response. A service not yet assessed requires another. The public deserves to know which situation applies. An unexplained blank or a reassuring colour should not stand in for that distinction.

There is room for narrative, but narrative must remain accountable to evidence. A clearly labelled fictional household can explain a mechanism, as the river district has done throughout this article. A real resident’s account requires appropriate consent, context and accurate representation. Neither should be used to imply a prevalence or a causal result it cannot establish.

The best ending names what the findings will change. It can be modest: a payment discrepancy will be reconciled; an access barrier will be investigated; a service definition will be corrected; a follow-up will test continuity after temporary support ends. A report becomes more credible when its authors are willing to let evidence alter the work.

Return, finally, to the school with the new roof. Imagine that the bridge becomes usable through the appropriate professional process. The child can now reach class. The teacher adjusts instruction after learning what returning pupils can understand. Attendance becomes more regular. A later assessment, designed to support a fair comparison, shows improvement on the selected skills. The district secures a workable operating arrangement and checks that participation continues.

These are fictional developments, not claims about a real programme. They show how the meaning of recovery becomes stronger as evidence follows the whole journey. The roof did not become less important when we asked about the bridge. The bridge did not become less important when we asked about learning. Each improvement found its place in a larger explanation.

Even then, the district should not declare every problem solved. Another household may remain excluded. Another service may be fragile. Recovery is uneven, and a trustworthy system should remain able to detect that unevenness. The purpose of measurement is not to discover the moment after which curiosity can stop.

The deeper achievement is a change in what public institutions can honestly say. They can identify a need, act, observe what happened, recognise limits, correct errors and explain their decisions. They can distinguish an intention from a delivery, a delivery from a usable benefit, and a benefit today from a credible prospect of continuation.

That capacity matters beyond any single indicator. It makes public promises testable. It allows residents to challenge an account without being told that every criticism denies progress. It allows staff to acknowledge a problem without erasing the work already done. It allows recovery to remain a shared inquiry rather than a competition over whose story sounds complete.

A completed project tells us that a defined task has ended. A recovering society needs to know whether ordinary life can proceed more dependably because of it. The distance between those two statements is where measurement does its most important work.

---

Sources and further reading

The sources below anchor the factual distinctions and published examples in the text. The fictional district, calculations, proposed review arrangements and explanatory scenarios are original teaching material. They are not findings endorsed by the cited institutions. Sources were checked on 15 September 2026; a source’s publication date is not the date of every observation it discusses.

1. OECD — Evaluation Criteria. The six criteria and their purpose as grounds for evaluative judgement, rather than a complete methodology. Read the source.

2. World Bank — Recovery and Peacebuilding Assessments: a tool to prevent conflict and promote peace. Asbjorn Wee, 27 January 2020. The joint World Bank, European Union and United Nations approach to assessment, prioritisation and sequencing. Read the source.

3. United Nations Statistics Division — Fundamental Principles of Official Statistics. Professional independence, confidentiality, transparency and the public role of official statistics. Read the source.

4. WHO/UNICEF Joint Monitoring Programme — Drinking water. The service ladder and the conditions defining safely managed drinking water. The original household-overlap exercise in Chapter 7 is not the JMP’s country-estimation method. Read the source.

5. UNESCO Institute for Statistics — SDG 4 Indicators. The distinctions between learning proficiency, completion, out-of-school rates and other education indicators. Read the source.

6. World Health Organization — Harmonized Health Facility Assessment: Introduction. The broader facility-assessment framework and its modules. Read the source.

7. World Health Organization — Service Availability and Readiness Assessment. Scope, indicators and the limitations of availability/readiness measurement alone. Read the source.

8. ILOSTAT — Work and employment are not synonyms. The distinctions between employment, other forms of work and labour-force status. Read the source.

9. Joint IDP Profiling Service — Durable Solutions. Multidimensional analysis drawing on the IASC framework, including displaced people and host communities. Read the source. See also JIPS’s introduction to the IASC framework.

10. UNHCR Microdata Library — Iraq: Access to Durable Solutions Among IDPs in Iraq, 2016–2021. A catalogue description of longitudinal, multidimensional research; no unpublished records or individual data were used for this article. Read the source.

11. UNDP — Support to reporting on global SDG 16 indicators under targets 16.3, 16.6 and 16.7. The population-based access-to-dispute-resolution measure, including indicator 16.3.3. Read the source.

12. World Bank DIME Wiki — Difference-in-Differences. The comparison logic and identifying assumptions. The attendance calculation in Chapter 17 is invented for teaching. Read the source.

13. Katherine Casey, Rachel Glennerster and Edward Miguel — Reshaping Institutions: Evidence on Aid Impacts Using a Preanalysis Plan. Quarterly Journal of Economics 127(4), 2012, pp. 1755–1812. Findings and interpretation were checked against the authors’ institutional publication abstract; this article does not reproduce the full paper or claim to reanalyse its data. Stanford publication record. Research-centre summary.

14. World Bank — Afghanistan’s economic twilight: Using nighttime lights to decode the dynamics of the economy after 2021. Silvia Redaelli, Oscar Eduardo Barriga Cabanillas and Walker Kosmidou-Bradley, 4 December 2024. A bounded example of interpreting night-time lights by the activity producing them; not a current assessment of the country. Read the source.

15. OCHA Centre for Humanitarian Data — Revised OCHA Data Responsibility Guidelines. 20 February 2025. Safe, ethical and effective management of operational data. Read the source.

16. OECD — Evaluating Peacebuilding Activities in Settings of Conflict and Fragility: Improving Learning for Results. 8 November 2012. Publication overview on context-sensitive evaluation, learning and programme management. The full manual was not used as a source for unverified procedural details. Read the source.

Continue through the eduKate War library

For the wider civilisational frame, start with What is Civilisation?. For the assessment of assets and losses, read How War Damage Assessment Works. For the state’s financial processes, read How Postwar Public Financial Management Works.

The preceding four articles examine justice-system recovery, civil-service reform, social protection and land administration. Each addresses a particular system. This article asks how evidence can show whether improvements in those systems become usable, equitable and durable changes in everyday life.