GR model fix: add complete geography tiers and uncertainty metadata for national candidate
done synth-modeler
Task metadata
| id | t_677ca342 |
|---|---|
| title | GR model fix: add complete geography tiers and uncertainty metadata for national candidate |
| assignee | synth-modeler |
| status | done |
| tenant | synthestat |
| priority | 55 |
| workspace_kind | dir |
| workspace_path | /home/synthestat |
| created_by | synth-reviewer |
| created_at | 2026-05-19 21:15:27 CEST |
| started_at | 2026-05-19 21:16:45 CEST |
| completed_at | 2026-05-19 21:23:02 CEST |
Latest summary
Patched the GR national candidate bundle in place for the reviewer contract blockers: geography tiers now cover EL plus all 13 emitted ELSTAT region codes, parquet evidence_depth uses contract vocabulary, uncertainty intervals/bounds are explicit, and diagnostics expose zero A01/A06 max residuals plus the 13,303 unassigned-dwelling semantics. Verification preserved all row counts and assignment status counts; spawned reviewer re-check t_a6630369.
Body
Reviewer verdict for GR national candidate bundle /home/synthestat/output/runs/GR/gr_population_national_candidate_20260519T185320Z_35e33441_seed420987 is NEEDS_MODEL_FIX, not because it is a fixture (it is national-scale), but because metadata/diagnostic contract gaps block PASS_FOR_INTERNAL_REVIEW. Required fixes: 1. geography_quality_tiers.json must cover every emitted geography/zone, not only EL. Current persons/households emit 13 regional location_code values: 111,112,121,122,231,232,241,242,243,351,461,462,471; dwellings emit EL. Assign A/B/C tier and reason per emitted zone or explicitly document the national-only dwelling zone mapping. 2. evidence_depth values must use the Synthestat contract vocabulary (measured, constrained, modelled, unavailable) or include a documented machine-readable mapping. Current values include partially_constrained and aggregate_residual_only. 3. uncertainty_summary.json/model_notes must provide explicit uncertainty bounds/intervals for model-driven quantities and overlays, not just textual notes. Applies to private/non-private split, relationship roles, education/activity/occupation/industry/origin, residual overlays, and dwelling assignment residual. 4. Diagnostics should make fit status machine-routable: include max residuals for A01 region x age x sex and A06 region x household-size controls (review independently verified both are zero), plus explicit unassigned-dwelling residual semantics. Reviewer-verified positives to preserve: actual parquet metadata counts are 10,482,487 persons, 4,332,447 households, 6,596,761 dwellings, 4,332,447 assignment rows, 3 overlay rows; A01 national/region age-sex residuals are zero; A06 region household-size residuals are zero; source checksums validate; hidden overlays are aggregate and not misused as person labels.
Parents
[ "t_d4f4d8b6" ]
Children
[ "t_a6630369" ]
Runs
| ID | Profile | Status | Outcome | Started | Ended | Summary/error |
|---|---|---|---|---|---|---|
| 158 | synth-modeler | done | completed | 2026-05-19 21:16:45 CEST | 2026-05-19 21:23:02 CEST | Patched the GR national candidate bundle in place for the reviewer contract blockers: geography tiers now cover EL plus all 13 emitted ELSTAT region codes, parquet evidence_depth uses contract vocabulary, uncertainty intervals/bounds are explicit, and diagnostics expose zero A01/A06 max residuals plus the 13,303 unassigned-dwelling semantics. Verification preserved all row counts and assignment status counts; spawned reviewer re-check t_a6630369. |
Events
| Time | Kind | Payload |
|---|---|---|
| 2026-05-19 21:15:27 CEST | created | {
"assignee": "synth-modeler",
"status": "todo",
"parents": [
"t_d4f4d8b6"
],
"tenant": "synthestat",
"skills": [
"synthestat-infra05-validation-framework"
]
} |
| 2026-05-19 21:15:54 CEST | promoted | null |
| 2026-05-19 21:16:45 CEST | claimed | {
"lock": "vmi3188806:1706016",
"expires": 1779219105,
"run_id": 158
} |
| 2026-05-19 21:16:45 CEST | spawned | {
"pid": 1709049
} |
| 2026-05-19 21:23:02 CEST | completed | {
"result_len": 0,
"summary": "Patched the GR national candidate bundle in place for the reviewer contract blockers: geography tiers now cover EL plus all 13 emitted ELSTAT region codes, parquet evidence_depth uses contract vocabulary, uncertainty intervals/bounds are explicit, and diagnostics expose zero A01/A06 max residuals plus the 13,303 unassigned-dwelling semantics. Verification preserved all row counts and assignment st",
"verified_cards": [
"t_a6630369"
]
} |
Comments
No comments yet.