Skip to main content

Plan: Ingest kommune-level 10th-grade absence (fravær) statistics

Ingests Udir's GSK_fravaer ("Fravær i grunnskole") report — median and average days/hours of documented absence, plus participant count, for 10th-grade pupils only, per kommune, for the latest school year. Atlas's fourth Udir source, and [Q39]'s grunnskole half. [Q39]'s videregående half (VGO_fravaer) is investigated in Phase 1 below and deliberately deferred — its geography does not resolve to kommune at all, a genuinely new problem for Atlas, not a parsing gap.

IMPLEMENTATION RULES: Before implementing this plan, read and follow:

Status: Completed​

Goal: Add udir-fravar as a served Atlas source — kommune-level 10th-grade absence, Atlas's first annual-absence axis (previously only fhi-vgs-gjennomforing's 3-year completion rate existed as a dropout-adjacent signal). Plugs Report #3 (Youth Outcomes) per [Q39]'s own framing.

Last Updated: 2026-10-03

Investigation: INVESTIGATE-new-norwegian-public-sources.md §4 [Q39], next in the Tier-1 Udir sequence after PLAN-016-udir-nasjonale-prover.md

Prerequisites: None. udir is already a valid publishers.yaml provider, topics.yaml already has education validated, and this source reuses the same statistikkportalen.udir.no client udir-gsi/udir-elevundersokelsen-mobbing/udir-nasjonale-prover already proved out — no new infrastructure for the GSK half shipped by this plan.

Phase 1: Confirm the real shape — DONE (verified 2026-10-03)​

Checked live rather than trusting [Q39]'s description or assuming either half's shape from a sibling Udir source. Mixed news this time: the grunnskole half matches udir-gsi's shape cleanly; the videregående half does not resolve to kommune at all — a genuinely new problem, not previously seen in any Udir source this session.

Tasks​

  • 1.1 Find the live reports. Confirmed — listed all 136 Rapportside entries (GET /rest/v1/Rapportside), the same technique PLAN-016 used. [Q39] names one source but the live listing shows it is genuinely two separate reports, not one report with a filter: GSK_fravaer ("Fravær i grunnskole", basePath rest/v1/Statistikk/GSK/FravaerG/1/1) and VGO_fravaer ("Fravær i videregående skole", basePath rest/v1/Statistikk/VGO/ResultatV/1/3). No "(gammel)" duplicate exists for either — confirmed by re-scanning the full 136-entry list for any other frav-matching name.
  • 1.2 GSK_fravaer's EnhetID IS the row hierarchy, matching udir-gsi/ udir-nasjonale-prover — NOT udir-elevundersokelsen-mobbing's column-dimension shape. Confirmed live: metadata.rowHierarchy is ["Nasjonalt","Fylke","Kommune","Enhet"], and radSti=-12.*.* returns 371 rows in one call — 15 fylke (2-segment id) + 356 kommune (3-segment id, including Svalbard 2100). Same depth-filter-by-segment-count technique as udir-gsi/udir-nasjonale-prover applies directly.
  • 1.3 VGO_fravaer's EnhetID hierarchy has NO kommune level at all — a genuinely new shape, not previously encountered. Confirmed live: metadata.rowHierarchy is ["Nasjonalt","Fylke","Enhet"], and radSti=-12.* returns exactly 15 rows (fylke only) — the next depth down is individual schools (org numbers, e.g. "982759846"), not kommuner. This is not a parsing gap: Norway's videregående skoler are organised and owned by fylkeskommune, not kommune, and Udir's own EnhetID tree reflects that directly. No kommune-level videregående fravær exists in this API at all — the only path to a kommune figure would be aggregating individual schools by their own postal/registered kommune, the same dim_school crosswalk work PLAN-010-udir-gsi.md [Q2] already deferred for GSI.
  • 1.4 v1 scope decision: ship GSK_fravaer (grunnskole, kommune-resolved) only. Defer VGO_fravaer explicitly, named as a deviation from [Q39]'s own framing, not silently dropped. Reasons: (a) VGO_fravaer's real resolution is fylke or school, not kommune — publishing it at fylke grain would be Atlas's first fylke-only-resolution indicator relation, confirmed by reading atlas-data/dbt/check-kommune-marts-exclude-sentinels.sh, which treats every indicators__/mart_indicators__ relation carrying a kommune_nr column as kommune-grain by construction — no precedent exists for a relation whose grain is fylke; (b) publishing it at school grain repeats the exact dim_school deferral already made twice (PLAN-010's [Q2], this investigation's own [Q33]) rather than resolving it. This is a cross-cutting decision (extends investigation §"D. New geographic resolutions", which only anticipated resolution finer than kommune, not coarser) and does not belong buried inside one source's Phase 2 — flagged as its own open question below ([Q4]) rather than decided unilaterally here.
  • 1.5 Confirm GSK_fravaer's own scope is 10th grade only, not all grunnskole grades. Confirmed, structurally, not just by prose — GSK_fravaer's own gyldigeFiltre list (EierformID, EnhetID, KjoennID, TidID, VisAntallPersoner, VisMaaltall) carries no TrinnID filter at all; the report's own description text states why: "Tabellen viser medianen for antall dager og timer fravær for elever på 10. trinn, slik det erført på vitnemålet" (the table shows median days/hours of absence for 10th-grade pupils, as recorded on the diploma). This measure is structurally defined as 10th-grade-only upstream, not a filter Atlas chose to pin.
  • 1.6 Confirm the measure columns. Confirmed — requesting both measure-toggle filters at once (VisAntallPersoner(1)_VisMaaltall(1)) returns 5 columns in a single call: Median dager, Median timer, Snitt dager, Snitt timer, Antall elever — more than filterVerdier's own VisMaaltall value list suggested (it names only "Snitt timer" as the id:1 option's label; the real response carries all four day/hour statistics plus the participant count under that one flag). Comma-decimal ("9,0") and ASCII-space-thousands ("4 277") cell shapes, same parseCell conventions udir-nasjonale-prover already built — reused directly, no new parsing logic needed.
  • 1.7 Confirm the suppression marker and sentinel codes for GSK_fravaer. Confirmed — literal * (same convention as every Atlas source this session), verified on Modalen (fully suppressed, all 5 measures). Svalbard (2100) present at the same tree depth as genuine kommuner, with real data (26 pupils) — same precedent as every prior Udir source, resolves through classify_region_code's existing svalbard branch, no new macro code needed. Utsira entirely absent from the row set (not suppressed — the "real absence, not suppression" shape udir-elevundersokelsen-mobbing/udir-nasjonale-prover already found, likely too small a 10th- grade cohort to report at all that year).
  • 1.8 Confirm Utlandet/2599 for GSK_fravaer. Confirmed present AND carrying real, non-suppressed data every year — same -13("Utlandet")→-34("25")→-476("2599", "Utlandet, uspesifisert") chain udir-nasjonale-prover found, and the same "carries real data" shape, not an empty one. ⚠️ A real correction, caught at the start of Phase 2, not Phase 1: this plan originally reported radSti=-13.*.* as returning 0 rows for this report. That manual test had an explicit, conflicting EnhetID(-12) filter alongside the -13.*.* radSti anchor — an invalid combination, not a real empty response. Re-tested without the conflicting filter (confirmed live, 65 real pupils for 2024-25) and the real data appeared immediately; the real ingest module (which never sets EnhetID in the filter string) confirmed this again independently on its first real run — Utlandet returns 5 real measure rows for every one of the 11 backfilled years. v1 decision unchanged in substance: query the -13.*.* anchor every run and include it — it is real, non-suppressed, upstream-published data, same discipline as udir-nasjonale-prover's own [Q1]/1.5 decision.
  • 1.9 Confirm available years. Confirmed — GSK_fravaer's TidID lists 11 school years (2014-15 through 2024-25); VGO_fravaer's lists 12 (2013-14 through 2024-25) — far more history than udir-nasjonale-prover's 4. ⚠️ A real correction, caught during Phase 2 start, not Phase 1: this plan originally recommended "latest year only, matching every other annual Udir source's convention" — that claim was wrong, found by actually reading udir-gsi's own index.ts rather than trusting its README's prose. udir-gsi does NOT defer backfill — its main loop (for (const year of years)) issues one data call per discovered year and ingests all 12, confirmed live (GSK/GSI/1/8/filterVerdier really does return 12 TidID entries, and the shipped row count, 18,816, is consistent with multi-year, not one). The real pattern across Udir sources is backfill fully when the call cost is cheap, defer only when it is not — udir-elevundersokelsen-mobbing deferred because of its ~702-call/year cost (a real, stated reason), not because of a universal "latest year only" convention; whether udir-nasjonale-prover's own "deferred by convention" claim holds up to the same scrutiny is not re-litigated here, but not inherited either. Revised recommendation: backfill all 11 years for GSK_fravaer — this report's call cost is cheap (2 radSti calls per year, same shape as udir-gsi), so there is no cost reason to withhold the real history, and udir-gsi's own actual behavior is the closer precedent than any sibling's prose claim.
  • 1.10 Confirm licence. NLOD, same portal (udir.no/om-udir/data/) already confirmed explicitly for udir-gsi/udir-elevundersokelsen-mobbing and inherited (not re-fetched a fourth time) for udir-nasjonale-prover — inherited again here for the identical reason: same portal, same organisation, no surface difference to re-check.

Validation​

✅ Confirmed 2026-10-03. Real API calls made directly against statistikkportalen.udir.no for both reports, the full filter-value discovery for both, and real data (Agder's kommuner, Modalen, Bygland, Beiarn, Svalbard, the Utlandet anchor, and VGO's fylke-level rows) — not inferred from any sibling Udir source's shape.

Open Questions​

  • [Q1] Scope — GSK_fravaer only, or wait and ship both halves together? Recommendation: ship GSK_fravaer now, defer VGO_fravaer. The two reports are genuinely separate endpoints with genuinely different geography (kommune vs. fylke/school) — there is no shared "fravær" dimension table to design once for both, unlike udir-nasjonale-prover's two report versions (which shared kommune geography and only differed in valid grades). Waiting on a fylke-resolution or dim_school decision to ship the grunnskole half, which is ready now, would be the same mistake PLAN-014/015/016 explicitly avoided — scope conservatively, ship what's real, name the rest as a deviation.
  • [Q2] Source naming. Should this source still be named udir-fravar even though it ships only the grunnskole half? Recommendation: yes, with the manifest description stating "10th grade, grunnskole only" explicitly — udir-sluttet-vgs/[Q40] is already a separate planned source for the videregående dropout signal, so there's no risk of the name implying broader scope than what's shipped; a future VGO_fravaer ingest (if the fylke/school-grain question resolves) would need its own source id anyway given the geography difference, not become a second half of this one.
  • [Q3] Measure scope — all 5 columns, or just the headline median? Median dager/Median timer are what the report's own description foregrounds (and what the diploma itself records); Snitt dager/Snitt timer/Antall elever are the same call's free extra columns. Recommendation: all 5 — no extra cost (same single call per kommune), and Antall elever is needed context for interpreting the other four (a median over 11 pupils reads differently than one over 400).
  • [Q3b] Historical backfill — RESOLVED during Phase 2 start, 2026-10-03. See 1.9's correction: back fill all 11 available years, not latest-only. udir-gsi's own real behavior (checked by reading its code, not its prose) is cheap-backfill-fully, and GSK_fravaer's call cost matches udir-gsi's shape exactly.
  • [Q4] The fylke/school-grain VGO_fravaer decision — cross-cutting, not this plan's to make. Flagging for 1PRIORITY.md/the investigation doc rather than deciding unilaterally: does Atlas want to (a) support a fylke-only-resolution indicator relation as a new first-class grain (extends investigation §D), (b) build dim_school + a school→kommune crosswalk and aggregate (same work already deferred twice), or (c) not ship VGO_fravaer at all and treat [Q39]'s videregående half as answered by udir-sluttet-vgs/[Q40] instead (which is VGS-specific dropout, a related but not identical signal)? Not resolved here.
  • [Q5] Presentation sensitivity. Per-kommune 10th-grade absence is a sharp, small-number-prone signal (confirmed: Modalen's entire row suppressed, Utsira absent) — same presentation_policy: 'sensitive' consideration flagged for every prior NAV/Udir source this session. Not resolved here; a Phase 2/3 decision.

Phase 2: Ingest module + raw table — DONE (verified 2026-10-03)​

Tasks​

  • 2.1 Created atlas-data/ingest/src/sources/udir-fravar/: manifest.yml (source_id: udir-fravar, provider: udir, periodicity: P1Y, license: NLOD, description stating "10th grade, grunnskole only — see [Q2]"), index.ts, parse.ts, fetch_retry.ts (copied), README.md, __tests__/ with real captured fixtures covering: an ordinary kommune (Agder's Arendal), a fully-suppressed row (Modalen), Svalbard (2100), and the Utlandet anchor's real data (2599). 17 tests, all passing.
  • 2.2 parse.ts — reused udir-gsi's depth-by-segment-count row filter directly (confirmed the same shape in Phase 1.2), issuing 2 data calls per year (radSti=-12.*.* + radSti=-13.*.*, same anchor pair udir-nasjonale-prover uses — both carry real data, see 1.8's correction), looped over all 11 discovered years (see 1.9's correction — 22 calls total, confirmed live in 6.1s wall time, still cheap, matching udir-gsi's own real backfill-fully behavior rather than a "latest year only" convention that turned out not to be universal). All 5 measure columns parsed per row; same parseCell comma-decimal/space-thousands handling as udir-nasjonale-prover.
  • 2.3 Migration 066_raw_udir_fravar.sql: raw.udir_fravar(region_code, year, measure, value, loaded_at) — one row per region/year/measure. No grade column, as planned (see 1.5) — this source is structurally 10th-grade-only.
  • 2.4 Dagster registration — raw_other.py/schedules.py, annual cadence, existing weekly-polled job pattern (same group as the other three Udir sources), no new job.
  • 2.5 Added ingest:udir-fravar npm script FIRST, verified via check-every-source-has-an-ingest-script.sh (✓ all 54 ingest source directories covered) and the real npm run ingest:udir-fravar invocation before any other Phase 2 work, per this session's standing discipline since husbanken-bostotte's urb-agents#1807.

Validation​

✅ Confirmed 2026-10-03. Real run against the live API against local scratch Postgres: 22 calls, 21,290 rows written, zero rows silently dropped. Svalbard (2100, real data, present in 6 of 11 years), the Utlandet anchor's real data (2599, present in all 11 years), and Modalen's fully-suppressed row all confirmed against real loaded data, not fixtures alone.

⚠️ A genuinely important correction, caught during this exact validation step: Phase 1's own 1.8 finding that the Utlandet anchor returns zero rows was WRONG — see 1.8/1.9's corrected text above. The real ingest module's first live run (which never sets EnhetID in its filter string, unlike the flawed manual Phase 1 test) returned 5 real measure rows for 2599 on every single one of the 11 years, immediately. Caught before this plan's Phase 3 work began, not after.

Phase 3: dbt staging and marts — DONE (verified 2026-10-03)​

Tasks​

  • 3.1 Added raw.udir_fravar to models/indicators/sources.yml.
  • 3.2 indicators__udir_fravar.sql — kommune_nr/region_kind via region_code_to_kommune_nr/classify_region_code. Confirmed against real loaded data: region_code='2100' → region_kind='svalbard', kommune_nr NULL (30 rows, present 6 of 11 years); region_code='2599' → region_kind='unspecified_within_fylke', kommune_nr NULL (55 rows, present every year) — both exactly as predicted, no macro change needed. Clean contents_code slugs for the 5 measures (median_dager/median_timer/snitt_dager/ snitt_timer/antall_elever), same CASE WHEN convention udir-nasjonale-prover used.
  • 3.3 Documented columns in models/indicators/schema.yml; mart_indicators__udir_fravar.sql api passthrough + marts/api/schema.yml entry. mart_atlas_inventory.sql depends_on + relation list updated; generate_api_v1.py's SCHEMA_COMMENT bumped (50→51 total, udir (3)→udir (4)).
  • 3.4 dbt build against real loaded data — clean (14/14 for the targeted build; mart_atlas_inventory and its full upstream dependency tree also rebuilt clean, confirming indicators__udir_fravar reports row_count=21290, is_empty=false, contributing_sources={udir-fravar}). Full 17-script atlas-data/dbt/check-*.sh suite run — all 17 pass clean (updated counts: 91 wrappers, 54 sources, 51 fact relations). Full ingest test suite (391 tests, 27 files) and tsc --noEmit both pass clean. website/npm run build completes clean, no broken links.

Validation​

✅ Confirmed 2026-10-03. Real local Postgres, not an empty schema. 2100/2599 resolution confirmed with a direct query against the built marts.indicators__udir_fravar table, same discipline as every prior sentinel this session — not assumed from the macro's existing branches alone.

⚠️ Two pre-existing, unrelated test failures observed during the full dbt build run (extracted_columns_are_still_populated, raw_sources_were_refreshed_recently) — both are local-scratch-Postgres staleness (most sources in this environment were never loaded, and the Brreg-enriched jsonb columns were populated once on 2026-09-24 and have gone stale since). Neither names udir-fravar or udir_fravar anywhere in their output; isolated by excluding both and confirming mart_atlas_inventory and its full dependency tree then build 100% clean (PASS=1639, WARN=1 pre-existing dim_postnummer relationship, ERROR=0). Not this plan's defect to fix — a standing local-environment gap, same class of thing as this session's own [[local-postgres-for-real-dbt-evidence]] note.

Phase 4: Deploy and verify arrival — IN PROGRESS (submitted 2026-10-03)​

PR #536 (Phases 2-3) merged to main as 0ffeb79. Image build (run 37155371998) completed and published:

artifact_digest  sha256:fe8ab71e9dcf91b4b5dc6554a3f79bf198ed494818bab2c9cd78716082866530  ghcr.io/terchris/atlas-data/uis:v20261003-0ffeb79
image_digest     sha256:bd146257a33a215514e1f8bf99282ff328efb28f2c99973a2de7a4cacba06253  ghcr.io/terchris/atlas-data:v20261003-0ffeb79

Both copied verbatim from the release's own uis-artifact.json, not reconstructed. render-template-info.sh v20261003-0ffeb79 confirmed clean (63 raw = migrations, 93 marts stated once, 54 automated sources covered).

atlas-data/uis/lands-with.sh 59e8bc3..0ffeb79 confirmed: annual_sources_refresh then transform_and_publish, plus a separate template-info pin (pin and a run are two actions, see [[a-pin-and-a-run-are-two-actions]]).

Sent:

  • urb-agents#1827 → imac: Dagster deploy request, row-count prediction 21,290 for raw.udir_fravar/indicators__udir_fravar. This source's call volume (22 calls total across 11 backfilled years, see 1.9) is cheap, unlike udir-elevundersokelsen-mobbing's — measured live in dev at 6.1s wall time, not assumed from a Phase 1 sample (per [[exploratory-calls-dont-reveal-sustained-api-latency]]). No "slow run" warning needed.
  • urb-agents#1828 → ops-dev: template-info pin nomination for the same tag.

Only predict indicators__udir_fravar as a served relation — never a second mart_indicators__... entry, see [[mart-prefix-is-never-a-served-endpoint]].

✅ DONE (verified 2026-10-03). imac's report (urb-agents#1827): annual_sources_refresh hit the slow end of this API's latency range (~4-6s/call, confirming [[exploratory-calls-dont-reveal-sustained-api-latency]]'s own run-to-run-variance finding again) — ~86 minutes total, correctly distinguished from a hang by polling Dagster's own API directly rather than trusting the --wait client's 3600s timeout. raw.udir_fravar: 21,290 rows, exact match to prediction. transform_and_publish succeeded; zero PostgreSQL corruption events across the full window; a 6-source regression check confirmed unchanged.

Independently re-verified against the live public API, not taken on imac's report alone: GET /indicators__udir_fravar?limit=1 with count=exact → content-range: 0-0/21290, exact match; real row returned with region_kind=kommune/kommune_nr correctly resolved. meta_sources?source_id=eq.udir-fravar → served_as: [indicators__udir_fravar] only, no second mart_indicators__... entry.

ops-dev's pin report (urb-agents#1828): cross-checked every reported number against the actual committed template-info.yaml diff (sources: 53→54, publishers: 9 unchanged, public_relations: 90→91, 62/92/78→63/93/79 raw/marts/views, NLOD for 50 of 54) — all exact, no discrepancy.

Both urb-agents#1827 and #1828 closed with the verification evidence attached as comments.

Acceptance Criteria​

  • The mechanism is verified live — Udir's FravaerG table via statistikkportalen.udir.no, confirmed by direct API calls, not assumed from any sibling Udir source's shape.
  • The videregående half's genuinely different geography is documented, not silently dropped — VGO_fravaer has no kommune-level resolution in this API at all; deferred as [Q4], not shipped as a degraded or invented kommune mapping.
  • Licence independently confirmed for this surface — NLOD, same portal already confirmed explicitly earlier this session, inherited consistent with established discipline.
  • udir-fravar ingests cleanly with zero rows silently dropped, including the 2100 and 2599 sentinels (both carrying real data) and at least one suppressed row all represented.
  • indicators__udir_fravar and its mart build and test clean against real loaded data, with 2100/2599 resolving through classify_region_code exactly as predicted.
  • udir-fravar appears in meta_sources.served_as after a real deploy, independently verified via live curl.
  • Golden-file tests cover: an ordinary kommune, a suppressed row, Svalbard, and the Utlandet anchor's real data.
  • The investigation and 1PRIORITY.md are updated to mark this candidate shipped, with [Q4] (the fylke/school-grain VGO_fravaer decision) carried forward explicitly as still open.

Implementation Notes​

  • One source id, one report, one geography. GSK_fravaer only — VGO_fravaer is a different report with a genuinely different, non-kommune geography; do not fold it into this source later without re-reading [Q1]/[Q4] first.
  • This report's EnhetID is a row hierarchy, like udir-gsi/udir-nasjonale-prover — confirmed fresh via rowHierarchy in the response metadata, not inherited by assumption.
  • The Utlandet/-13 anchor carries real, non-suppressed data every year — the opposite of this plan's own original Phase 1 finding, which was a buggy manual test (an explicit, conflicting EnhetID(-12) filter alongside the -13.*.* radSti anchor). Corrected at the start of Phase 2; see 1.8.
  • No new Dagster job. Same weekly-polled group as the other three Udir sources.
  • This source is structurally 10th-grade-only — not a filtered slice Atlas chose, Udir's own FravaerG table has no other grade to select. Do not add a grade column that would imply otherwise.
  • Backfills all 11 available years, not latest-only — a correction made at the start of Phase 2 (see 1.9): udir-gsi's real behavior (checked in its code) backfills fully when the call cost is cheap, and this report's cost matches. Real call volume is cheap (~22 calls total) — verify this holds in Phase 2, don't assume it from Phase 1's sample, per [[exploratory-calls-dont-reveal-sustained-api-latency]].

Outcome​

Shipped end to end, 2026-10-03/04. Atlas's fourth Udir source: kommune-level, 10th-grade documented absence statistics (GSK_fravaer), same EnhetID-row-hierarchy shape as udir-gsi. Ingested 21,290 rows (region × year × measure, 11 backfilled school years), published to api_v1, deployed to the live cluster by imac (urb-agents#1827), independently re-verified against the public API — GET /indicators__udir_fravar?limit=1 with count=exact confirms content-range: 0-0/21290.

Two real corrections, both caught during Phase 2, not Phase 1:

  1. Backfill convention was wrong. Phase 1 recommended "latest year only," matching what it believed was every annual Udir source's convention. Checking udir-gsi's own code (not its README's prose) found it backfills fully when the call cost is cheap — this report's cost matches (22 calls total), so v1 ships all 11 available years.
  2. The Utlandet/2599 anchor's "empty response" finding was a Phase 1 testing bug, not a real finding. The manual curl test that produced an apparent zero-row result had an explicit, conflicting EnhetID(-12) filter alongside the -13.*.* radSti anchor — an invalid combination. The real ingest module (which never sets EnhetID in the filter string) confirmed Utlandet carries real, non-suppressed data every single year, the same shape udir-nasjonale-prover already found for its own Utlandet anchor.

VGO_fravaer (videregående) was deliberately not shipped — its EnhetID hierarchy has no kommune level at all (fylke/school-grain only, confirmed live), a structural fact about how Norway organises videregående skole, named as [Q4] and carried forward to the investigation doc as a cross-cutting decision rather than resolved unilaterally.

Deploy took ~86 minutes for annual_sources_refresh — this API's per-call latency hit the slow end of its own range again (confirming [[exploratory-calls-dont-reveal-sustained-api-latency]]'s run-to-run-variance finding), correctly distinguished from a hang by imac polling Dagster's API directly rather than trusting a client timeout. Zero data-quality issues; a 6-source regression check confirmed nothing else moved.

Files to Modify​

  • atlas-data/ingest/src/sources/udir-fravar/manifest.yml (new)
  • atlas-data/ingest/src/sources/udir-fravar/index.ts (new)
  • atlas-data/ingest/src/sources/udir-fravar/parse.ts (new)
  • atlas-data/ingest/src/sources/udir-fravar/fetch_retry.ts (new, copied)
  • atlas-data/ingest/src/sources/udir-fravar/README.md (new)
  • atlas-data/ingest/src/sources/udir-fravar/__tests__/ (new)
  • atlas-data/ingest/package.json (ingest:udir-fravar script — add this BEFORE shipping, verify with the real npm run invocation)
  • atlas-data/migrations/066_raw_udir_fravar.sql (new)
  • atlas-data/dagster/atlas_data/assets/raw_other.py, schedules.py (registration — annual cadence, existing weekly-polled job pattern)
  • atlas-data/dbt/models/indicators/sources.yml, indicators__udir_fravar.sql (new), schema.yml
  • atlas-data/dbt/models/marts/api/mart_indicators__udir_fravar.sql (new), schema.yml
  • atlas-data/dbt/models/marts/api/mart_atlas_inventory.sql (depends_on + counted list)
  • atlas-data/dbt/scripts/generate_api_v1.py (SCHEMA_COMMENT's udir listing — bump the count)
  • atlas-data/template-info.yaml, website/docs/developers/index.md (regenerated counts)
  • website/docs/ai-developer/plans/backlog/INVESTIGATE-new-norwegian-public-sources.md (mark GSK_fravaer shipped, carry [Q4]/VGO_fravaer forward as still open)
  • website/docs/ai-developer/plans/backlog/1PRIORITY.md (mark shipped)