Latest summary
Reviewed the NL cycle-1 population bundle at /home/synthestat/output/runs/NL/nl_population_review_cycle1_fa78aabf_seed420987 and wrote the reviewer report at /home/synthestat/workspace/manager_handoffs/reviewer/2026-05-19_164411Z_NL_population_review_cycle1.md. Verdict NEEDS_MORE_SOURCES: bundle is contract-complete and seeded HARD/FIRM residuals pass, but it is only a 5-person/3-test-zone slice with full-country source freeze, hidden overlays, assignments, household realism, and provenance gaps blocking PASS.
Body
Review the NL population synthesis review bundle produced by parent task t_8187172c.
Country: NL
Project root: /home/synthestat
Release mode: internal research review
Manager/orchestrator task: t_c692bc83
Read before review:
- /home/synthestat/docs/contracts/population_review_bundle.md
- /home/synthestat/docs/SOUL.md (if stale/wrong in repo, follow injected Synthestat constitution: uncertainty-first, HARD/FIRM/SOFT/GUIDE precedence, no silent degradation)
- Parent modeler handoff and bundle path.
Review scope:
Assess whether the bundle is complete, scientifically responsible, provenance-aware, and uncertainty-aware for NL at the finest available official geography. Check persons-households-dwellings-buildings consistency, hidden-population overlay separation, family composition plausibility, parent/child age gaps, school/work assignment scope, dwelling/building realism, constraint residuals, geography degradation, source provenance, and explicit limitations.
Required verdict schema in kanban_complete metadata:
{
"country": "NL",
"bundle_path": "...",
"verdict": "PASS | NEEDS_MODEL_FIX | NEEDS_MORE_SOURCES | BLOCKED_INVALID_OUTPUT | EVIDENCE_EXHAUSTED_HUMAN_REVIEW | MODEL_IMPROVEMENT_EXHAUSTED_HUMAN_REVIEW",
"blocking_findings": [ ... ],
"non_blocking_findings": [ ... ],
"required_next_actions": [ ... ],
"hard_constraint_status": "pass|fail|unknown",
"uncertainty_status": "pass|fail|unknown",
"provenance_status": "pass|fail|unknown",
"hidden_population_overlay_status": "pass|fail|unknown",
"similar_failure_signature": "short stable string if failed, for stopping-rule comparison"
}
Verdict rules:
- PASS only if satisfactory for declared internal research review mode and all non-negotiables are met.
- NEEDS_MODEL_FIX when available evidence is sufficient but model/build logic or diagnostics need repair/rerun.
- NEEDS_MORE_SOURCES when a concrete source/research gap blocks responsible improvement; identify whether synth-marginals-researcher, synth-distributions-researcher, and/or synth-downloader is needed.
- BLOCKED_INVALID_OUTPUT when the bundle is missing, malformed, or unreviewable under the contract.
- EVIDENCE_EXHAUSTED_HUMAN_REVIEW when further source search cannot responsibly support improvement.
- MODEL_IMPROVEMENT_EXHAUSTED_HUMAN_REVIEW when model improvement has plateaued or modeller cannot improve with available evidence.
Non-negotiables:
- HARD constraints never break.
- No fake precision: weak evidence means relaxed constraints/wider uncertainty, not exact claims.
- Hidden population overlays do not silently rewrite de jure official constraints.
- Missing data, relaxed constraints, failed inputs, degraded zones, and modelled estimates must be visible.
Definition of done:
Complete with the required verdict metadata. Do not create follow-up tasks yourself; the manager will branch from your verdict.