Monitoring, quality and evidence
You are the person between raw collection and anything anybody would act on. Field progress, data quality, evidence review and the handoff to analysis all run through you.
What you are responsible for
Whether what was collected is good enough to use, and whether the evidence behind a finding actually holds. You are not responsible for the statistical analysis itself, and you are the person best placed to notice when a result does not match what the field is telling you.
Watching the field
The monitoring dashboard shows what is being collected, where it is slowing, and what has been flagged for review. Progress and quality sit on the same screen because separating them is how a programme ends up with a full dataset nobody can use.
Reviewing quality
Quality flags are raised by scoring rather than by intuition, and they are a prompt to look rather than a verdict.
- Open the flagged responses
Each carries the reason it was flagged — an unusually short interview, a conflicting answer between entries, a value outside a plausible range.
- Decide, and record the decision
Accept it, or send it back for recollection. Either way the decision is recorded against the response, which is what makes the dataset defensible later.
- Look for the pattern, not the record
One flagged interview is noise. Fifteen from the same assignment in the same week is a field problem that no amount of record-by-record review will fix.
Evidence
Evidence is the link between what someone said and what a report claims. The register holds those links, and it is the thing that makes a finding checkable months later by somebody who was not there.
- A quotation stays attached to the response it came from. It is never separated from its source.
- A translation is never presented as the original. Where a translated quotation is published, the original wording is published beside it.
- A finding with no resolvable evidence does not publish. The export is refused rather than produced with a caveat.
Handing off to analysis
When collection for a wave is complete and quality review is done, the dataset goes to analysis. What matters at this point is that the methodology record is complete — a study that has not declared its sampling frame will be reported as not having declared one, rather than being given a default.
When something goes wrong
| What you see | What to do |
|---|---|
| Completion has stalled in one area | Look at the assignment and the channel before the enumerator. A route that does not reach the people you need looks identical to a team that has stopped working. |
| A response cannot be flagged or unflagged | Check your permissions. Quality actions require an explicit permission rather than a role name alone. |
| A report will not export | This is usually the evidence gate doing its job: something the report claims does not resolve to a record. The refusal names what is missing. |
| The numbers in two places disagree | They should not. Every product renders from the same approved publication at a named version. If two disagree, one is stale — report it rather than reconciling by hand. |
Limits worth knowing before you plan around them
What this role cannot do, and what the platform does not offer
- There is no per-enumerator quality score. The platform does not attribute responses to individual enumerators, so it cannot compute one and does not pretend to.
- There is no “returned for correction” queue against an enumerator identity, and no supervisor comment thread inside the app. Feedback to a field worker goes through you, not through the platform.
- Statistical summaries, cross-tabs and disaggregation on some surfaces are reported as not measured where no computation is wired. A blank there is honest rather than broken.