Zambia's infrastructure data: one of six public checks can run

The door is not the verdict

A door is marked Transparent Project. You expect to walk into a trophy room. You walk into an inspection room. Good, the inspector says, now: good by what test?

The tempting sentence arrives with the published record: the project is transparent, so the project is accountable, so the project, surely, is good. That sentence is easy to attack. Anyone can find one transparent project that failed and one secret project that worked, and the claim collapses. So this essay does not make it. It makes the harder, truer one. Without disclosure, you cannot test whether a project is good. With disclosure, you still have to run the test. A visible contract is not yet a knowable project.

To show what that means, the essay runs six inspection tests against a country's published infrastructure record, using CoST Zambia's field-level mapping of the ZPPA e-GP platform. One scope point governs everything below: these are findings about a mapping of a platform and a sample of records within it, not findings about one named project. Where a test describes a field, it describes the platform's field-level picture, not a verdict on any single road.

Four reasons a record cannot answer

Most writing about disclosure collapses every gap into one word: hidden. The mapping supports four distinct states, and they call for four different remedies.

An absent field is not evidence of withholding. In this mapping, most gaps are not restricted and not hidden: they are either collected but undisclosed, or simply never collected. Only the restricted state is a deliberate withholding, and it is the rarest.

The six tests

Each test resolves one way against the mapping. None of these is a verdict on a road. They are a verdict on what the record lets you check.

Test one: who owns the contractor? Not collected. The supplier is named and graded, but who owns or controls the firm is absent from every reviewed system. The remedy the mapping records is policy action, not a missing form, because no register currently captures it. Beneficial ownership is the clearest case in the set of information that publication alone cannot supply.

Test two: was procurement competitive? Sharp. Procuring entity, procurement method, number of bidders, cost estimate and contract price are published. This is the part the system is built to disclose, and it does.

Test three: did the price change? Not published. Contract variations are captured by the National Council for Construction under its monitoring mandate, but they never reach the public e-GP system. The change exists; the public record of it does not.

Test four: did the project finish on time? Not published. Progress updates, payment certificates and actual completion dates are collected inside government but are not integrated into any public system. You can see the project was awarded, not whether it landed on time.

Test five: was the asset maintained? Not collected. Maintenance schedules, cost logs and inspection data are not captured in any reviewed system. This is not withheld; it is simply not collected, so the test cannot even be built yet.

Test six: can anyone verify the record? Not published. The record's linked-release field is empty, and most implementation evidence sits in unstructured forms or behind the disclosure line. A record you cannot attach proof to, or link to others, can be read but not verified.

The tally: one sharp, three not published, two not collected, none restricted. One test can run. Five cannot. Visible is not the same as testable. The record is bright on the questions a procurement system likes to answer, and dim on the ones an inspector needs answered. The door was real. The trophy was not.

What the mapping found across the life of a project

Disclosure does not fail at random. It thins along the life of the project. Phase by phase, the mapping's coverage runs: identification 28 percent, preparation 12 percent, procurement 21 percent, implementation about 2 percent, completion 8 percent, maintenance 0 percent, decommissioning 0 percent. The pattern is structural, not accidental: in four of the seven phases the dominant cause is not restriction but collection that never reaches the public.

Maintenance and decommissioning score zero because no reviewed system collects them at all. Beneficial ownership sits in the same category. These are not failures of will; they are fields with no home yet.

The mapping's forward picture is that a set of OC4IDS fields can be published today, a further set becomes publishable within twelve months once format work is done, and another group needs policy action rather than formatting. The public summary of the mapping records the first two boundaries; it does not publish a count for the policy-action group.

Even a sharp record teaches you little alone

A record that passes every test can still teach you nothing on its own. The questions that matter most are portfolio questions: what share of contracts has no disclosed owner, is that share rising, which sectors disclose least, can the records even be linked. Those need every record, structured the same way, reachable in bulk. That is usability, and it is a separate build from disclosure.

Uganda's Government Procurement Portal makes the point. As of the November 2025 mapping prepared for the Africa Freedom of Information Centre and Uganda's PPDA, it met 17.6 percent of the OC4IDS requirements assessed, 26 of 148 data elements, and 4.2 percent of project-level disclosure paths. That is a historical baseline from before the portal's June 2026 upgrade, not a current reading. It could tell you who won which contract. It could not yet tell you whether the works were finished on time, within budget, or to the commitments made.

The close is a work order

A gap is not the record failing to try; it is the record showing you exactly where the next reform has to happen. Beneficial ownership needs policy change, not a database column. Variations and completion data need a disclosure switch flipped on data government already holds. Maintenance needs a capture workflow that does not yet exist.

The honest verdict is not "this project is good." It is "this project can now be tested." That is a smaller claim, and a far harder one to take away from you.

The data

The six inspection tests

Tally: 1 sharp, 3 not published, 2 not collected, 0 restricted.

Lifecycle disclosure coverage (Zambia mapping, true percentages)

Coverage is fields mapped against applicable OC4IDS slots per phase. In four of the seven phases the dominant cause is collection that has not reached the public.