Performance Is Not Proof: The Outcome Ledger MBG Needs for School Benefits

MBG Watch · 2026-08-25

The premise

A child who comes to school because food is waiting is telling us something important. So is a teacher who says children look more focused after eating. These are not trivial observations. Hunger can make school harder, and a reliable meal can make the school day more humane.

But performance signals are not proof.

As of 25 August 2026, Indonesian public sources already connect MBG with school benefits. Badan Gizi Nasional wrote in March 2025 that MBG contributes to stronger school attendance and concentration, and described attendance and concentration as medium-term benefits monitored under the programme. In August 2025, BGN’s head, Dadan Hindayana, said student attendance had risen from an average of 70–75 percent to at least 95 percent, and also cited one SPPG in Warung Kiara, Sukabumi, that had measured weight, thinness, height, and other nutrition-related changes after 1.5 years. The education ministry’s vocational directorate reported Sorong implementation in 294 schools with 59,803 beneficiaries, and quoted school leaders describing children as more focused, more active, more cheerful, and more regular in attendance.

Those are claims worth examining, not dismissing. They also show why MBG now needs a stricter public outcome ledger. Meals delivered, meals eaten, attendance, classroom energy, nutrition status, anemia, cognition, learning, wellbeing, and household food substitution are different layers. If they are collapsed into one public story, the programme can look successful before the child benefit has actually been verified.

This is the same measurement risk MBG Watch has already named in nutrition claims. Why MBG Has No Measurable Nutrition Outcomes argued that delivery architecture is not the same as nutrition impact. The Jayapura Stunting Claim asked whether promoted numbers were actually measuring what they claimed. Earlier Care Is Not Earlier Proof applied the same discipline to mothers, toddlers, and 3T routes. Seen Without Being Watched set the privacy boundary for beneficiary validation. The present question is the school version: how can MBG recognize real school-day improvements without converting every visible improvement into causal proof?

The answer is not a child-level dashboard. It is a layered ledger.

What the evidence supports

International evidence gives MBG a reason to measure school benefits seriously. It does not give MBG permission to skip measurement.

The World Food Programme’s 2021 focused review of school feeding in low- and lower-middle-income countries synthesized 20 publications, including 12 randomized experiments and quasi-experiments. WFP frames school feeding as a possible multi-benefit intervention: children may benefit, families may benefit, and communities may benefit. But the review’s structure matters: it separates modalities, outcome measures, and impact areas rather than treating school feeding as one automatic cause of all desirable outcomes.

Cochrane’s review of school feeding for disadvantaged schoolchildren gives a useful older benchmark for caution. In the higher-quality randomized trials from low-income countries, children fed at school gained an average of 0.39 kg more than controls over 19 months. Children attended school more frequently, translating to about 4 to 6 additional days per year per child. Educational and cognitive outcomes were more mixed: gains appeared in math achievement and some short-term cognitive tasks, but the review still called for better-designed studies and more robust outcome data.

This is the shape of the evidence MBG should take seriously:

• Attendance can plausibly move early, especially where children are hungry or families face food costs.
• Classroom energy and focus can plausibly improve in the short term, especially among children who would otherwise skip breakfast or eat insufficiently.
• Anthropometric and micronutrient outcomes need repeated, sample-based measurement over months or years, not anecdotes or single-site impressions.
• Learning outcomes require comparison groups, baseline data, and care with confounding. A child can attend more often and still not show durable learning gains if teaching quality, illness, fatigue, transport, or household stress remain binding constraints.
• Household effects are real but often invisible. A school meal may add nutrition, or it may partly replace food the household would otherwise have provided. Both can reduce family burden; only one necessarily increases the child’s total intake.

The Indonesian claims now in circulation fit mostly in the first two layers: attendance and observed classroom functioning. The Warung Kiara example begins to touch nutrition measurement, but BGN’s own press account presents it as one SPPG evaluation rather than a nationally representative result. That distinction should remain visible.

What the evidence does not support

The public record does not yet support a broad causal claim that MBG has improved learning outcomes, cognitive development, anemia, stunting, or long-term wellbeing at national scale.

It also does not support treating school enthusiasm as a complete outcome. A child may be excited because the meal is good, because food is scarce at home, because the school day has become more social, or because the programme is new. Each explanation matters, but each implies a different policy response.

Nor does attendance prove nutritional adequacy. A child can attend for the meal but not eat enough of it. A meal can be delivered but arrive late, be refused, be traded, be spoiled, or be nutritionally weak. A school can show improved attendance while absenteeism falls for reasons unrelated to MBG: calendar effects, new school management, local campaigns, changed recording practices, weather, transport, or a comparison year disrupted by another event.

This is why the August 2025 BGN statement deserves careful handling. A rise from 70–75 percent attendance to at least 95 percent would be large. It may point to a real change in some settings. But the public needs to know the denominator, geography, school level, baseline period, attendance definition, comparison group, and whether recording practices changed. Without those details, the number is a signal to investigate, not proof of national impact.

The same is true of teacher observations. Teachers are close to the child and should be heard. But asking teachers to become unpaid evaluators of a national programme is unfair and methodologically weak. Their observations should help design measurement; they should not be the whole measurement system.

The least-harm path

The least-harm path is to publish enough to make school-benefit claims auditable, while keeping children out of public surveillance.

For MBG, that means separating four records:

First, the service record: what was delivered, when, to whom, under which kitchen, route, menu, and safety conditions.

Second, the consumption record: how many intended beneficiaries were present, how many meals were safely available, how many were actually consumed or refused, and why non-consumption occurred. This can be aggregated by school, kitchen, route, week, and beneficiary category. It does not require publishing names.

Third, the school-day functioning record: attendance, reason-coded absence, late arrival where relevant, and sample-based attention or participation measures. These should be compared to pre-MBG baselines and to similar schools or periods not yet exposed to the same implementation intensity.

Fourth, the child-outcome record: sample-based nutrition and learning measures collected under controlled audit protocols, not displayed as public child profiles. Anthropometry, hemoglobin or micronutrient measures where ethically and practically justified, dietary diversity, short attention tasks, and learning assessments can all be useful. They should be sampled, consent-aware, privacy-preserving, and interpreted with uncertainty.

This approach protects three things at once. It protects children from being watched as proof objects. It protects teachers and parents from being turned into the unpaid measurement arm of the state. And it protects MBG itself from premature claims that later fail under scrutiny.

The minimum school-benefit outcome ledger

BGN could publish a minimum ledger at institution, kitchen, route, district, and time-period level. Sensitive records would remain available only for controlled audit. The public version should include:

1. Delivery denominator: intended beneficiaries by school or service point, day, beneficiary category, kitchen, menu, route, and delivery time.

2. Safe-availability denominator: meals arriving within the safe time-temperature window, rejected meals, late meals, missing meals, and service interruptions.

3. Safe-consumption denominator: beneficiaries present; meals accepted, partially eaten, refused, replaced, or discarded; and reason categories such as absence, fasting, illness, allergy, taste, portion size, lateness, spoilage concern, or household instruction. No child names in the public data.

4. Attendance baseline and comparison: pre-MBG attendance by comparable calendar period; current attendance; school-level comparison groups where possible; method notes for changes in attendance recording.

5. Reason-coded absence and non-service: illness, weather, transport, teacher absence, exam periods, school closure, kitchen disruption, route disruption, food-safety hold, or other local factors. This keeps MBG from claiming effects caused by something else.

6. Classroom functioning samples: periodic, light-touch attention or participation measures using sampling rather than continuous surveillance. The method should be published; raw child-level data should not be.

7. Nutrition samples: repeated anthropometry and, where justified, hemoglobin or micronutrient testing on ethically approved samples, with age, sex, geography, baseline, follow-up interval, and confidence intervals.

8. Learning samples: grade-appropriate literacy and numeracy checks, compared with baseline and credible controls. The public record should show aggregate results and uncertainty, not individual children.

9. Household substitution and burden check: sample surveys asking whether MBG added to total food intake, replaced home food, reduced household food spending, changed breakfast patterns, or shifted care burdens. This is necessary because family burden can fall even when child intake does not rise as much as expected.

10. Correction path: what happens when a school or kitchen shows poor consumption, repeated lateness, safety holds, unexplained attendance claims, or weak nutrition progress. Measurement without a correction path becomes theatre.

11. Privacy boundary: public data only at institution, kitchen, route, district, and time-period level; sensitive data under controlled audit access; no public child profiles; no biometrics by default; no family blame for non-consumption or absence.

12. Uncertainty notes: what can be inferred, what cannot yet be inferred, which groups are under-sampled, where baselines are weak, and which claims require later follow-up.

What I am uncertain about

I am uncertain whether the strongest Indonesian attendance claims are based on a national dataset, a set of observed locations, or programme communications assembled from selected examples. The public BGN statements reviewed here do not provide enough methodological detail to tell.

I am also uncertain how consistently MBG sites record meals eaten rather than meals delivered. If the system mainly records production and dispatch, then school-benefit claims will sit on a weak foundation even when kitchens are operating well.

Finally, I am uncertain whether existing Indonesian studies on concentration and motivation are strong enough to support causal claims. Some may be useful as local case evidence, but MBG’s national public record should not depend on small observational studies when the programme is large enough to support better evaluation.

The cautious conclusion is also the constructive one. MBG may well improve attendance, classroom readiness, and some child outcomes in some settings. But durable benefit has to be shown layer by layer. Performance signals can justify continued investigation. They cannot substitute for the outcome ledger children deserve.