Latest summary
Reviewed BG cycle-1 bundle and wrote reviewer handoff at /home/synthestat/workspace/manager_handoffs/reviewer/2026-05-19_164341Z_BG_population_review_cycle1.md. Verdict NEEDS_MORE_SOURCES: bundle is contract-complete and honest about seeded scope, but not passable as a country population due 2-zone/8-person fixture scope, missing live/full-source coverage, unavailable hidden/assignment layers, seeded building grounding, incomplete provenance, and comparable-country/modelled distribution dependence.
Body
Review the BG population synthesis review bundle produced by parent modeler task t_6c64d812.
Manager/orchestrator task: t_2a07ba7d
Project root: /home/synthestat
Country: BG
Release mode: internal research review
Read before work:
- /home/synthestat/docs/SOUL.md
- /home/synthestat/docs/contracts/population_review_bundle.md
- parent task handoff and artifact paths
Review objective:
Determine whether the BG population review bundle is valid and responsible under the Synthestat QA process. Do not fix the model yourself. Produce a verdict that the manager can branch on.
Required checks:
1. Contract completeness: all files required by docs/contracts/population_review_bundle.md exist, or unavailable.json files give explicit reasons.
2. Constraint precedence: HARD constraints are exact; FIRM/SOFT tolerances are declared and residuals are justified; GUIDE/INFORMATIONAL sources are not overclaimed as measurement.
3. Uncertainty: every modelled estimate, hidden-population overlay, weak geography estimate, and assignment layer has uncertainty bounds or explicit unavailability.
4. Provenance: source_provenance.json records source IDs/URLs/retrieval timestamps/reference periods/geography levels/quality flags for material inputs.
5. Hidden population handling: homelessness, refugee/asylum, Ukrainian displaced, Syrian refugee, undocumented/seasonal, student, and institutional overlays are included only where supported; exclusions or weak evidence are explicit; overlays do not silently rewrite de jure constraints.
6. Geography quality: finest available official geography is used where possible; degraded zones are listed and justified.
7. Household/dwelling/building realism: persons-households-dwellings-buildings links are coherent where available; unavailable layers are explicitly marked.
8. No silent degradation: missing sources, failed downloads, relaxed constraints, modelled estimates, and limitations are visible in machine-readable diagnostics and model_notes.md.
Verdict taxonomy (must use exactly one):
- PASS
- NEEDS_MODEL_FIX
- NEEDS_MORE_SOURCES
- BLOCKED_INVALID_OUTPUT
- EVIDENCE_EXHAUSTED_HUMAN_REVIEW
- MODEL_IMPROVEMENT_EXHAUSTED_HUMAN_REVIEW
Definition of done:
- Complete via kanban with metadata containing: verdict, bundle_path, blocking_findings, nonblocking_findings, required_next_actions, evidence_gaps, model_fix_requests, and whether this is materially similar to prior failed cycles.
- If verdict is NEEDS_MORE_SOURCES, identify which specialist lanes are needed: synth-marginals-researcher, synth-distributions-researcher, and/or synth-downloader, with concrete source questions.
- If verdict is PASS, name final review/delivery artifacts and any residual limitations for human review.
- If invalid, list the missing/invalid bundle files explicitly.