Synthestat · Estonia · population QA
EE 1:1 population synthesis QA cycle
Country-specific layer for synthetic people in households, dwellings, real building stock where available, hidden-population overlays, and work/school assignment evidence.
Board: synthestat-population-qa · Tenant: synthestat · Country: EE · Task workflow status: ready · Artifact completion: review_bundle_metrics_partial
Ideal-country quality criteria: impossible 1:1 benchmark
This is the common gold-standard benchmark for an ideal country. It is intentionally impossible to fully satisfy: complete success would mean a 1:1 replica of the real population where every person, household, dwelling, attribute, and assignment is exactly represented. The QA page uses it as an asymptote and gap taxonomy, not as a release promise.
Apply this same rubric to this country’s latest run, then report which needs are measured, constrained, modelled, unavailable, or blocked.
| Need | Unachievable ideal | QA evidence we require instead | Why perfection cannot be achieved |
|---|---|---|---|
| Complete de jure resident coverage | Every real resident represented exactly once in the right country, municipality, small area, household, and dwelling. | Synthetic person count equals official population at all enforced geographies; no unexplained duplicate, missing, or out-of-universe people. | A true 1:1 resident list is a confidential population register and changes continuously; Synthestat can only match official aggregates and declared source universes. |
| Complete attribute truth | Each synthetic person has the same age, sex, household role, education, occupation, industry, origin, health proxy, income proxy, and lifecycle state as the corresponding real person. | Published marginal and cross-tab constraints pass within HARD/FIRM/SOFT tolerances; modelled fields carry uncertainty and measured/constrained/modelled provenance. | Official releases do not expose a complete individual joint distribution, and many attributes are survey-derived, lagged, suppressed, or unavailable at fine geography. |
| Perfect household and family structure | Every household contains the exact real members and relationships, including multi-generation, partnership, child, shared, institutional, and edge-case arrangements. | Household totals, household-type distributions, age/sex/role consistency, fertility/child constraints, and structural invariants pass with explicit residuals. | Household membership is sensitive microdata; public sources usually expose only aggregate household/family tables and partial cross-tabs. |
| Exact dwelling and building grounding | Every household is assigned to its real dwelling and building with exact occupancy, vacancy, dwelling type, floor area, tenure, and address-level geography. | Dwelling/building capacity checks pass; vacancy/second-home/institutional dwellings are represented or explicitly unavailable; building links have source provenance. | Many countries lack open address-level registers; dwelling occupancy is confidential and time-varying. |
| Complete de facto and hidden-population overlays | Homeless, undocumented, refugees, students away from home, seasonal, institutional, tourists, and daytime populations are all represented with exact location and timing. | Overlay layers use interval estimates, source-specific quality flags, and never silently modify de jure HARD constraints. | Hidden populations are partly unobserved by definition; ethical/privacy constraints forbid exact person-level labels. |
| Exact school, workplace, facility, and mobility assignment | Every person is assigned to the real school, workplace, care provider, commute, and daily activity chain they use. | Assignment layers use official registers/OD flows where available; modelled assignments are flagged and validated only against aggregate flows/capacities. | Operational assignments are usually protected registers or dynamic behavioural data; Phase 1 must not imply they are known. |
| Full joint-distribution realism | The full multivariate joint distribution is identical to reality across all attributes, households, geography, and rare subgroups. | High-priority marginals/cross-tabs pass; sparse zones and prior-dominated attributes are clearly marked with quality tiers and credible intervals. | The joint distribution is non-identifiable from published marginals; IPF/BN/hierarchical pooling choose plausible distributions, not truth. |
| Zero uncertainty and zero lag | All values are current today and known without error. | Every output records reference period, retrieval timestamp, lag, confidence, uncertainty bounds, and degradation decisions. | Official statistics are lagged, revised, sampled, suppressed, and harmonized after collection. |
| Privacy-safe yet maximally detailed release | The system releases maximum useful detail while creating zero re-identification risk. | Release mode, k-anonymity/cell safeguards, perturbation/aggregation policy, and sensitive-field treatment are explicit. | Fine-area synthetic microdata can still create structurally unique records; synthetic does not mean anonymous. |
| Perfect reproducibility and auditability | Any user can trace every output record to exact source snapshots, transformations, constraints, relaxations, seeds, and code versions. | Run manifests, source provenance, checksums, frozen extracts, seeds, versioned crosswalks, validation reports, and relaxation logs are complete. | This is approachable but never final: source portals, classifications, geography, and code keep changing, so audits must be continuously renewed. |
Population artifact output status (separate from task status)
| Artifact completion | Row count source | People | Target population | National coverage | Absolute shortfall | Households | Dwellings | Houses/buildings | Max marginal deviation | HARD status | Run |
|---|---|---|---|---|---|---|---|---|---|---|---|
| review_bundle_metrics_partial | parquet_metadata_review_bundle | — | — | — | — | 561,655 | — | — | — | — | ee_population_private_household_national_2021_cycle2_seed420987 |
This table describes emitted population artifacts only. It is intentionally independent from the Kanban task workflow status below: a country can have all tasks done while its artifact is still only a seeded slice, or a passing review bundle can still lack national target completion. Deviation is the maximum absolute relative error across collected HARD/FIRM/SOFT marginal constraints in the latest review bundle. GUIDE/INFORMATIONAL priors are excluded. National target/coverage are read from build_manifest.json when available and override any visual impression of completion.
Kanban task workflow status (not artifact completion)
ready
1
done
7
These cards count board tasks only. They do not certify that the country-level population artifact is nationally complete or reviewer-approved.
Datasets and distributions
Lists come from the latest run bundle: source_provenance.json, distribution_diagnostics.json, and build_manifest.json.
Summary
| Datasets used | 0 |
|---|---|
| Distributions available | 0 |
| Constraints/distributions used in synthesis | 3 |
| Constraint types | — |
| Dataset variants | — |
| Finest-geography status | — |
Source gaps
- No source gaps listed.
Datasets used
| Dataset/source ID |
|---|
| None listed yet. |
Best source by distribution family
| Distribution family | Dataset/source ID |
|---|---|
| None listed yet. | |
Available distributions / priors in registry
| Spec | Label | Type | Geo | Status | Variant | Confidence | Data URI |
|---|---|---|---|---|---|---|---|
| None listed yet. | |||||||
Constraints/distributions used in synthesis manifest
| Constraint or distribution ID |
|---|
EE_STAT_RL21707_private_households_total |
EE_STAT_RL21707_private_household_members_total |
EE_STAT_RL21703_private_household_age_sex |
Current country tasks
| ID | Title | Assignee | Status | Created | Latest summary |
|---|---|---|---|---|---|
| t_8026f58d | EE final human/internal review note for private-household national PASS | synth-reviewer | ready | 2026-05-19 22:03:45 CEST | |
| t_b967c410 | EE reviewer pass: inspect cycle-2 national private-household model fix | synth-reviewer | done | 2026-05-19 20:59:14 CEST | Reviewed EE cycle-2 national private-household bundle at /home/synthestat/output/runs/EE/ee_population_private_household_national_2021_cycle2_seed420987 and returned PASS for the declared private-household scope. Independent parquet QA verified 1,317,666 persons, 561,655 households, 561,655 dwelling |
| t_1bf7ff30 | EE model fix: repair invalid child-alone household assignment in national private-household bundle | synth-modeler | done | 2026-05-19 20:51:16 CEST | Built the EE cycle-2 national private-household review bundle at /home/synthestat/output/runs/EE/ee_population_private_household_national_2021_cycle2_seed420987. The model now assigns one adult reference per private household before filling remaining slots, preserving exact private-household counts |
| t_8f902059 | EE population QA reviewer: inspect national-scale non-fixture rerun | synth-reviewer | done | 2026-05-19 20:17:33 CEST | EE review complete: the rerun is a real national-scale private-household artifact (1,317,666 persons; 561,655 households), not the 8-person fixture, and it matches declared private-household HARD counts. Verdict is NEEDS_MODEL_FIX because household assignment is structurally invalid (174,570 child-w |
| t_4eb48eb2 | EE population QA modeler: national-scale non-fixture synthesis rerun | synth-modeler | done | 2026-05-19 20:17:32 CEST | Built EE non-fixture national-scale private-household review bundle at /home/synthestat/output/runs/EE/ee_population_private_household_national_2021_seed420987. It explicitly rejects the old 8-person /output/EE fixture as national, matches 2021 Statistics Estonia private-household HARD controls exac |
| t_fdbf1017 | EE population QA downloader: freeze exact official payloads for national-scale rerun | synth-downloader | done | 2026-05-19 20:17:31 CEST | Froze the EE national-scale population QA official payload bundle: 48/48 approved Statistics Estonia PxWeb tables and 2/2 Maa-amet WFS layers completed with manifests, checksums, row/feature counts, schema sketches, and downloader-local catalogue handoffs. Updated downloader latest.md and appended t |
| t_9140b2f8 | EE population QA distribution closure: joint priors for non-fixture synthesis | synth-distributions-researcher | done | 2026-05-19 20:17:30 CEST | Completed EE population QA distribution closure with verdict DISTRIBUTION_READY_FOR_MODEL_FIX. Wrote dated EE findings and extraction specs, preserved/updated the shared latest evidence board with an EE section, and appended a manager update. |
| t_bcf523c3 | EE population QA source closure: exact national marginals for non-fixture synthesis | synth-marginals-researcher | done | 2026-05-19 20:17:29 CEST | Completed EE population QA source closure and wrote dated sources/needs handoffs plus latest.md under /home/synthestat/workspace/manager_handoffs/marginals/. Estonia is SOURCE_READY_FOR_MODEL_FIX after downloader freezes official Statistics Estonia PXWeb/Maa-amet WFS payloads; national target is 1,3 |
Process
Manager kickoff
synth-manager creates and controls the country loop.
Model build
synth-modeler generates the review bundle: people, households, dwellings/buildings or unavailable markers, overlays, assignments, manifests, residuals, diagnostics, uncertainty, provenance.
Reviewer gate
synth-reviewer audits constraints, marginals, household/family realism, hidden populations, dwelling/building grounding, work/school assignment, uncertainty, provenance, and privacy.
Branch
PASS finalizes; NEEDS_MODEL_FIX routes back to modeler; NEEDS_MORE_SOURCES routes to marginal/distribution researchers then downloader; exhausted evidence/model plateau stops for human decision.
Quality gates and stop conditions
- PASS: satisfactory for declared country evidence tier and internal review mode.
- NEEDS_MODEL_FIX: model logic, bundle, uncertainty, household/dwelling/assignment issue.
- NEEDS_MORE_SOURCES: missing marginal or joint/conditional evidence; researchers then downloader.
- EVIDENCE_EXHAUSTED_HUMAN_REVIEW: source search cannot responsibly improve the output.
- MODEL_IMPROVEMENT_EXHAUSTED_HUMAN_REVIEW: modeler cannot materially improve or diagnostics plateau.