01Survey intelligence platform
CensaNation reconstructs survey values from microdata, administrative registers, and macroeconomic indicators — then compares the reconstruction against what was recorded in the field. Where the two disagree, you get a ranked, explained, reviewable list. Your statisticians decide what happens next.
It does not replace your surveys or your statisticians
The platform recommends. Your statisticians decide. Every flag is a suggestion for analyst review, every action is logged, and nothing is published without your approval. Runs entirely within your own infrastructure — microdata never leaves your jurisdiction.
The output, in one view
Select a programme. The accent mark appears only where the reconstruction and the collected record disagree beyond the review threshold — that divergence is what the platform delivers for analyst review.
| Reported earnings band | Collected | Reconstructed | Gap | Flagged for review |
|---|---|---|---|---|
| Band 1 — lowest | 104.2 | 103.8 | -0.4 | Within threshold |
| Band 2 | 118.6 | 119.4 | +0.8 | Within threshold |
| Band 3 | 131.0 | 130.2 | -0.8 | Within threshold |
| Band 4 | 147.5 | 168.9 | +21.4 | Flagged for review |
| Band 5 | 162.3 | 161.1 | -1.2 | Within threshold |
| Band 6 — highest | 171.8 | 196.4 | +24.6 | Flagged for review |
What the pattern suggests
Divergence concentrates in the upper earnings bands — the pattern an office would expect from under-reporting at the top of the distribution, not from scattered enumerator error.
Rows are marked where the gap exceeds the review threshold your office sets. Nothing here is a correction — it is a question put to an analyst.
Every figure shown on this site is synthetic and for illustration only. It is not drawn from any survey, office, or country.
Reconstruct, not predict
When a model predicts a value nobody measured, you must trust it blindly. When a model reconstructs a value you already measured, you can check the model against your own record. The gap between the two is the product's output.
Three ways offices use it
The same reconstruct-and-compare approach applies whether the round closed years ago, is in the field this week, or has not started yet.
01Retrospective
Run against a round already published — a labour force round, for instance. The collected answers are known, so the exercise validates itself: the report shows how closely reconstruction reproduces your figures, and precisely where it does not.
02In-field
Run alongside active fieldwork — an agricultural production survey mid-season, say. Records diverging sharply from the independent estimate are flagged for supervisor review or callback while teams are still deployed and correction is still cheap.
03Between rounds
Where samples are thin, non-response left holes, or the next household expenditure round is years away, produce model-based estimates with explicit uncertainty intervals, reconciled against published national aggregates.
The scale of the portfolio
Statistical offices do not run one survey. They run a standing portfolio, repeatedly, for decades, at very large scale. Individual offices in large countries run well over a hundred distinct surveys a year.
These instruments interlock. Expenditure surveys supply the weights for the consumer price index. The census supplies the sampling frame for nearly everything else. Labour force data feeds the national accounts. An undetected error in one instrument propagates into several published statistics.
And every completed round already sits in the archive — decades of microdata whose quality has never been independently re-examined.
Coverage
Wherever an office runs a repeating survey alongside administrative data, the same reconstruct-and-compare approach applies.
Where offices start
The round is closed and published. Nothing is at risk, no released figure can be altered, and accuracy is checkable against data the office already holds.
A technical briefing walks your methodologists through the approach in detail. No data required to begin the conversation.