When the evidence stops.

AMR research often produces an uncomfortable result. After reviewing a national action plan, surveillance data, donor records, development-bank projects, government documents and scientific literature, the evidence chain simply stops.

A budget cannot be located. A project looks relevant to AMR, but the AMR-specific financing cannot be isolated. Surveillance data exists, but national representativeness is unresolved. A policy commitment is visible, but implementation evidence cannot be connected to it.

The temptation is to close the gap. It should usually be resisted.

That distinction decides whether AMR intelligence remains auditable or starts manufacturing certainty.

AMR has more evidence now. That does not make every claim provable.

The global AMR information environment is much stronger than it was a decade ago. WHO's Global Antimicrobial Resistance and Use Surveillance System, GLASS, gives countries standardized methods for collecting, analysing and sharing AMR and antimicrobial-use information. WHO's 2025 global antibiotic-resistance surveillance report drew on more than 23 million bacteriologically confirmed infections reported by 104 countries for 2023 and introduced a framework for assessing national data completeness.

National policy implementation is tracked through TrACSS, the Tracking AMR Country Self-Assessment Survey. The 2025 survey launch notes that TrACSS is a country-led annual assessment of multisectoral AMR national action-plan progress. In 2024, 186 countries submitted responses.

The WHO AMR Data Portal brings indicators from GLASS, TrACSS and secondary sources into one access point. It also separates national action-plan maturity into distinct stages, from no plan through approved plans, costed operational plans and financial provision in national plans and budgets.

These sources answer important questions. They do not all answer the same question.

Every source has a scope.

If TrACSS reports that a country has an approved AMR national action plan under implementation, that is meaningful evidence. It is evidence of an officially reported national status. It is not the same as an independent audit of every activity carried out under that status.

The same applies to surveillance. GLASS provides a standardized architecture, but WHO continues to work on data quality, completeness and representativeness. The fact that WHO introduced a national-data-completeness scoring framework in the 2025 report is a reminder: participation in a surveillance system and complete national observability are not identical.

Development-finance data adds another layer. The World Bank's Projects & Operations system and Documents & Reports repository disclose a large amount of project information. The World Bank project cycle also shows that documents appear at different stages, from identification and appraisal through implementation and completion.

Failing to locate an AMR activity in one project search does not prove that no relevant national health investment exists. It proves something narrower: the activity was not identified within the evidence scope searched.

The real problem appears in financing.

Imagine a country priority: strengthen microbiology laboratory capacity for AMR surveillance.

An analyst finds a 50 million dollar health-system programme containing laboratory equipment, workforce training and digital surveillance infrastructure. Is the laboratory priority financed? Possibly. Can the full 50 million dollars be counted as AMR financing? Almost certainly not without more evidence. Can the analyst conclude that no AMR laboratory financing exists because the programme title does not say "antimicrobial resistance"? Also no.

The correct state may remain unresolved. That is not analytical failure. It is information.

The financing may be relevant but unattributable. The project may be enabling rather than directly AMR-specific. Additional documents may clarify the relationship. Or the evidence may end there.

Four evidence states are better than binary answers.

Plan-to-proof analysis needs a place to put uncertainty. A binary field often cannot do the job.

The critical state is unknown. Traditional reporting often handles unknowns badly. They disappear, become zeros, or become assumptions because a dashboard wants every row populated.

Each transformation makes the page look more complete while making the analysis less reliable.

Provenance sets the limit of the claim.

The easiest way to overstate an AMR finding is to detach it from the evidence path that produced it.

Consider the claim: "Country X invested 80 million dollars in AMR." Without provenance, it looks precise. With provenance, the real questions appear. Was that the total project value? Was AMR one component among several? Was the figure approved financing, committed financing or expenditure? Was AMR stated explicitly, or inferred from laboratory, surveillance or infection-prevention activities?

The source is not just a citation at the bottom of the page. It defines what the claim is allowed to say.

A serious "not found" result should contain metadata.

"Not found" should not be empty. It should record what was searched, what was not searched and what would resolve the uncertainty.

That turns absence from a dead end into an auditable research result. When new evidence appears, the unknown can be updated without reconstructing the whole analytical process.

Coverage is not proof.

This distinction matters more as AMR datasets grow. WHO's AMR Data Portal integrates primary and secondary indicators. GLASS continues to expand standardized surveillance. TrACSS provides broad global participation. Development-finance systems contain large volumes of activity and project information.

That is progress. But larger evidence coverage does not remove the need to ask what each observation proves.

A country appearing in GLASS does not mean every national surveillance question is observable. A TrACSS implementation score does not independently verify every programme activity. A World Bank health project does not make its entire financing envelope AMR capital. An IATI activity missing from a search does not establish that no development financing exists.

The discipline lies in preserving those boundaries.

The standard should be reproducibility.

Health intelligence is under pressure to produce clean answers. Decision-makers want totals. Dashboards want populated fields. Cross-country comparisons want standardized indicators.

But when the underlying evidence is fragmented, forced completeness can create a false picture. A better standard is reproducibility.

Could another analyst see the same sources and understand why a claim was verified? Could they identify where inference entered? Could they see exactly where the evidence stopped? Could they distinguish a missing public record from evidence that an activity did not occur?

If yes, uncertainty becomes manageable. If no, even a precise-looking dashboard may be difficult to defend.

The larger implication.

AMR does not need analysts to pretend every evidence chain is complete. It needs systems that can show where the chain is complete and where it is not.

That means treating unknowns as first-class research objects. It means preserving provenance. It means separating observation from inference. And it means refusing to turn "not found" into "does not exist" because the latter fits more neatly into a database.

That is a stronger foundation for decisions about AMR policy, financing and implementation.