Data quality methodology

The pipeline

Every record follows the same route, whether it arrives from an automated connector or an institutional submission.

1. Retrieve from an authorised official source, preferring an API or structured download over reading a web page.
2. Record the origin — the organisation, the system, the source URL, the original identifier, the retrieval time and, where the source reports it, the time the source last modified the record.
3. Validate required fields, formats and ranges. A record that fails validation is held, not published.
4. Map to ZEPRA's canonical structure, aligned with OCDS for contracting processes and OC4IDS for project-level information.
5. Normalise institutions, locations, currencies, sectors and dates against ZEPRA's reference data.
6. Match and deduplicate. A record is linked automatically only above a confidence threshold. Below it, the record goes to a data steward.
7. Publish with the source and last-updated time shown, and the trust label the evidence actually supports.
8. Retain the history. Field-level changes are recorded against the record permanently.

Trust labels

These labels are the platform's most important claim, so each one means a specific thing.

Label	What it means
Source verified	Matches an official published record that ZEPRA has checked against the source.
Institution submitted	Submitted directly by the responsible institution through the government workspace.
ZEPRA reviewed	Checked by a ZEPRA data steward for completeness and internal consistency.
Imported automatically	Retrieved from an official source by an automated connector. Not yet reviewed by a person.
Awaiting verification	Held in the register pending confirmation against an official source.
Incomplete record	Key fields are missing. Read the record with that limitation in mind.
Source unavailable	The originating source could not be reached at the last attempt. Figures may be out of date.

A connector that retrieves a record successfully produces Imported automatically. It does not produce Source verified. A scraper that worked has demonstrated that a page was reachable, not that its contents are correct.

Completeness scoring

Each project record is scored against the fields a complete record is expected to carry, weighted by how much each matters. The score describes the data, not the project: a low score usually means the institution's information has not yet been shared with ZEPRA in full. The fields missing from a record are listed on its page.

Delivery assessment

Delivery status, procurement status and management attention are three separate things and are never collapsed into one.

• Delivery stage says where a project has reached.
• Procurement stage says where a tender has reached.
• Attention status is a judgement a person recorded, with a stated reason and a date.

“Not assessed” is not “on track.” Most records carry no assessment, and presenting them as healthy would be a fabrication. Where ZEPRA observes that a completion date has passed, it says so as an observation drawn from the recorded dates and leaves the assessment to the institution.

Physical completion

Physical completion is only ever the figure the responsible institution reported, carried with the basis on which it was assessed and the date it describes. ZEPRA does not derive physical progress from expenditure. Financial completion is shown separately and is labelled as a financial measure, because money spent is not work done.

Location

A project is plotted as a point only where the responsible institution published site coordinates. Where only an administrative area is known, the area is shaded and the record states the precision it actually has. ZEPRA does not place a marker at an administrative centroid and present it as a project site.

Corrections and disputes

A correction is recorded against the record with its reason and its author; the previous value stays in the change history. A record whose accuracy is contested is marked as disputed on its own page while it is examined, rather than being quietly removed. Anyone can request a correction.

What ZEPRA will not do

• Publish a figure no source reported. An absent value is shown as an em dash, never as zero and never as an estimate.
• Infer a project's location, budget or progress from a similar project.
• Present a demonstration record as a government record.
• Overwrite a corrected or disputed record without recording what changed.
• Bypass authentication, anti-bot controls or access restrictions on a source, or use data contrary to a source's terms.