Skip to main content

Plan: Ingest kommune-level Elevundersøkelsen bullying ("mobbing") indicators

Ingests Udir's Elevundersøkelsen (pupil survey) bullying indicator and its three underlying questions, per kommune per grade (7th and 10th) per school year, as Atlas's second Udir source — a sharper, annual complement to the existing fhi-mobbing 3-year-rolling aggregate.

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

Status: Completed​

Goal: Add udir-elevundersokelsen-mobbing as a served Atlas source — an annual, per-grade bullying-prevalence signal at kommune resolution, sharper than fhi-mobbing's 3-year-rolling 7th+10th aggregate. Plugs Report #3 (Youth Outcomes) and strengthens Report #10 (School-Capacity Forecast)'s learning-environment axis.

Last Updated: 2026-10-03

Investigation: INVESTIGATE-new-norwegian-public-sources.md §4 ([Q10]-[Q13]), next in the Tier-1 sequence after PLAN-014-nav-helt-ledige.md

Prerequisites: None. udir is already a valid publishers.yaml provider (landed with PLAN-010-udir-gsi.md), topics.yaml already has the relevant categories validated, and this source reuses the exact same statistikkportalen.udir.no API client shape udir-gsi already proved out — no new infrastructure, but genuinely new response-shape handling (see Phase 1).

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

Checked live rather than trusting the investigation's description (last touched 2026-05, well before udir-gsi found the real API). This source's real shape is structurally different from every prior source this session — not a NAV-Excel pivot, not an SSB PxWebAPI call, and not even the same response shape as its own sibling udir-gsi despite sharing one client.

Tasks​

  • 1.1 Confirm the real API and find the live report. Confirmed — same system udir-gsi already found, statistikkportalen.udir.no/api/rapportering. Listed every Rapportside (GET /rest/v1/Rapportside, 136 report pages total) and found the live bullying report has migrated schemas since the investigation's own framing: the classic ElevundersoekelsenG table's own Rapportside (GSK_EU_mobbing) is stale — its filterDefaultVerdier.TidID is frozen at [201901, 202001, 202101]. The live, current table is EUG (GSK_EUG_mobbing, endpoint rest/v1/Statistikk/GSK/EUG/5/5), whose own default year is 202512 ("2025-26") — confirmed by fetching both Rapportside definitions directly and comparing, not assumed from naming similarity. Don't reuse the ElevundersoekelsenG table name anywhere in this source — it is the retired predecessor.
  • 1.2 Confirm available years. Confirmed — filterVerdier on TidID returns five school years: 2025-26, 2024-25, 2023-24, 2022-23, 2021-22 (ids 202512, 202412, 202312, 202212, 202112). All five are live-queryable today, not an archive-vs-live split like NAV's tables.
  • 1.3 A fundamentally different response shape than udir-gsi's — read this before writing a parser. udir-gsi's EnhetID (geography) is the row hierarchy, filtered to kommune depth by radSti. For this report, EnhetID is a column dimension, not a row. The row hierarchy is Indikator/Spørsmål (max 2 levels — confirmed by a 400 error when a 3-level radSti was tried: "For dyp radsti: for denne rapporten støttes maksimalt 2 nivåer"). Geography has no wildcard-and-filter-by-depth mechanism the way udir-gsi's did:
    • A wildcarded EnhetID (e.g. -12.*, "all of Norway, one level down") returns the entire national column tree in one response — nested columns[] arrays down through landet → fylke → kommune → individual school, each carrying a columnCount (how many leaf columns live under it). Confirmed live: a real response for 2025-26 nests all 17 fylker, then every kommune under each, then every individual school under each kommune.
    • A specific, non-wildcarded EnhetID (e.g. -720, Arendal's internal id) collapses the response to exactly one column, labelled "Alle skoler" — confirmed this is the kommune's own aggregate value, not a drill into its schools, by comparing against the all-of-Norway response's column structure for the same kommune.
    • [Q1] resolved 2026-10-03, at the start of Phase 2 — per-kommune calls, deliberately, not the bulk decode. The wildcard tree's leaf columns were never confirmed to contain a distinct per-kommune aggregate separate from its individual schools — decoding it without that guarantee risks silently extracting a specific school's number as if it were the kommune's, exactly the defect class this project's "verified correct, not merely clever" discipline (see udir-gsi's own README) exists to avoid. The per-entity-collapse technique is unambiguous ("Alle skoler" label, cross-checked against the national tree for the same kommune) and costs one HTTP call per region node per grade — filterVerdier's own EnhetID list has 351 nivaa-3 nodes (349 real kommuner + Svalbard's 2100 + Utlandet, uspesifisert's 2599 — fewer than the ~357-359 seen in every prior kommune-grain source, because a kommune with zero schools reporting into Elevundersøkelsen that year is simply absent from the tree, not present-and-suppressed; confirm this distinction with a real row-count check in Phase 2) × 2 grades = ~702 calls for one year, plus 2 setup calls. [Q2]'s scope follows from this: latest year only for v1 (all 5 years would be ~3,510 calls — defer backfill, same as every annual Udir table's own convention). A deliberate pacing delay between calls is warranted given the API's own "not intended for external use" disclaimer — courtesy, not a documented rate limit.
  • 1.4 Confirm the grade ("Trinn") filter semantics. Confirmed, and this corrects an assumption carried from udir-gsi. udir-gsi's TrinnID(-10)/KommunalitetID(-10) sentinel meant "alle" (all grades/ownership types summed). For this report, -10 is NOT a valid "alle" sentinel at all — TrinnID's real filterVerdier only lists ids 4–9 (grades 5–10), no "alle" option. Passing TrinnID(-10) did not error; it silently returned the last value in the report's own default list ([6, 9] → grade 10) without flagging anything wrong — caught only by cross-checking the result against an explicit TrinnID(9) call and finding identical values. This is exactly the kind of silent-wrong-default trap this project's "verify, don't assume" discipline exists for. TrinnID(6,9) (comma-joined, guessing at multi-value syntax) returns a genuine 400 ("Ugyldige tegn i filter") — there is no single call for "both grades at once"; v1 must issue one call per grade. Matches fhi-mobbing's own established 7th+10th-grade axis, so this is also the natural v1 scope, not an arbitrary restriction.
  • 1.5 Confirm the suppression marker. Confirmed — literal "*", same convention as every other Atlas source this session. Verified on Bykle (Agder, Norway's smallest kommune by population): its 10th-grade 2025-26 "Mobbing på skolen" indicator and one of its three questions are suppressed ("*"), while the other two questions report real "0,0" values in the same response — suppression is per-row, not per-kommune, matching udir-gsi's and every NAV source's own per-cell suppression.
  • 1.6 A genuinely new sentinel, not seen in udir-gsi: Norwegian schools abroad. The national column tree's top level has two siblings, "Hele landet" (id -12) and "Utlandet" (id -13, kode "U") — confirmed live, Elevundersøkelsen is answered by Norwegian schools operating abroad (e.g. "Den norske skole - Costa Blanca"). Under Utlandet's own fylke-equivalent node (kode "25" — ⚠️ the retired pre-2024 Finnmark fylke code, a real collision worth flagging even though fylke-level codes are out of scope for this kommune-grain source) sits exactly one kommune-equivalent node: "Utlandet, uspesifisert", kode 2599 — which matches classify_region_code's existing ^\d{2}99$ unspecified_within_fylke branch exactly, confirmed by inspection, no new macro branch needed. udir-gsi's own README and parse.ts never mention this sentinel; not assumed to carry over, checked fresh for this source.
  • 1.7 Confirm licence. NLOD, confirmed directly on Udir's own data portal (udir.no/om-udir/data/, which this source shares with udir-gsi — same portal, same organisation, re-fetched and read for this source rather than assumed to transfer): "lisens for offentlig data (NLOD)."
  • 1.8 Check data.norge.no for a cleaner distribution. Not separately re-checked this round — udir-gsi's own Phase 1 already established statistikkportalen.udir.no as the real, current, sole surface for every Udir statistics table including Elevundersøkelsen (confirmed by listing every table under schema GSK, which includes EUG/ElevundersoekelsenG alongside GSI). No reason to expect a different conclusion for this specific table; flagged as inherited, not independently re-verified, per this session's own discipline about not assuming licence/surface findings transfer silently between sibling sources — if this matters, re-check before shipping.

Validation​

✅ Confirmed 2026-10-03. Real API calls made directly against statistikkportalen.udir.no for the Rapportside definition, filter values, and real data (Arendal and Bykle, both grades, latest year) — not inferred from udir-gsi's shape or the investigation's five-month-old description.

Open Questions​

  • [Q1] Bulk-decode vs per-kommune calls — RESOLVED 2026-10-03, per-kommune calls. The wildcard-and-decode route was never confirmed to carry a distinct per-kommune aggregate leaf separate from its individual schools — decoding it on that unverified assumption risks silently extracting one school's figure as if it were the kommune's. The per-entity-collapse technique (EnhetID pinned to one specific id → response collapses to one column labelled "Alle skoler") is unambiguous, already cross-checked against the national tree for the same kommune, and costs one call per region node per grade: filterVerdier's own EnhetID list has 351 nivaa-3 nodes (349 kommuner + Svalbard's 2100 + Utlandet, uspesifisert's 2599) × 2 grades = ~702 calls for one year — far more than any prior source. At Phase 1's exploratory pace this looked cheap (tens of ms per call); a real full run (Phase 2) found that assumption wrong — only the first ~22 calls stay fast, every call after settles into a steady ~3.5-6s each, for ~60 minutes total wall time. Correctness over call count was still the right call — the alternative (bulk-decode) risked a silent misparse, not just a slower run — but the real cost of this decision turned out to be wall-clock minutes, not request count.
  • [Q2] Historical backfill — RESOLVED 2026-10-03, latest year only for v1. Follows directly from [Q1]: per-kommune calls make every extra year ~702 more requests (all 5 years would be ~3,510). Defer backfill, same convention as every other annual Udir table — revisit if a real consumer need for historical Elevundersøkelsen trend data emerges.
  • [Q3] Scope — mobbing only, or fold in trivsel/other Elevundersøkelsen indicator families too? Elevundersøkelsen covers many indicator families (GSK_EUG_mobbing, GSK_EUG_indikator / learning-environment, GSK_EUG_tema, and more) under one survey, structurally closer to Bufdir's multi-indicator monitors than to NAV's one-indicator-per-source tables. Recommendation: udir-elevundersokelsen-mobbing v1 ingests only the bullying indicator family (direct complement to fhi-mobbing, the gap this plan is named for) — matching Bufdir's own precedent of one source folder per indicator family (bufdir-barnefattigdom, bufdir-barnevern), not one mega-source. A future udir-elevundersokelsen-trivsel or similar would reuse this exact client.
  • [Q4] GSK (grunnskole) only, or also VGO (videregående)? ElevundersoekelsenV/EUG-VGO exists under schema VGO with the identical mechanism. Recommendation: GSK only for v1, matching udir-gsi's own grunnskole-only scope — VGO as a natural, cheap follow-up reusing this client, not folded in here.
  • [Q5] region_code handling for Utlandet/2599. See Phase 1.6 — this falls through classify_region_code's existing unspecified_within_fylke branch cleanly by inspection; verify with a real dbt test during Phase 3 rather than assuming the inspection is sufficient, same discipline as every prior sentinel this session.
  • [Q6] Presentation sensitivity. Per-kommune, per-school-year bullying share for minors is sensitive in the same way fhi-mobbing already is — same presentation_policy: 'sensitive' consideration as every prior NAV-adjacent source. Not resolved here; a Phase 2/3 decision.

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

Tasks​

  • 2.1 [Q1] resolved — per-kommune calls, not bulk-decode. See Phase 1.3/[Q1]. ~702 calls for the latest year (351 region nodes × 2 grades), with a deliberate pacing delay between calls.
  • 2.2 Created atlas-data/ingest/src/sources/udir-elevundersokelsen-mobbing/: manifest.yml, index.ts, parse.ts, fetch_retry.ts, README.md, __tests__/ with real captured fixtures covering: an ordinary kommune (Arendal, both grades), a suppressed row (Bykle), the Utlandet, uspesifisert/2599 sentinel, Svalbard/2100, and — found during the first real run, not anticipated in Phase 1 — a region/grade pair with zero rows at all (Hægebostad/4226, 10th grade: no 10th-grade cohort reports into this table). 24 tests, all passing.
  • 2.3 Migration 064_raw_udir_elevundersokelsen_mobbing.sql — raw.udir_elevundersokelsen_mobbing( region_code, grade, year, measure, measure_label, value, loaded_at), PK (region_code, grade, year, measure).
  • 2.4 Dagster registration — added to _ANNUAL_SOURCE_IDS/OTHER_SOURCES's weekly-polled group alongside udir-gsi, no new job. dagster definitions validate passes. Flagged explicitly in raw_other.py's own docstring and schedules.py's job description as a real outlier in request count and wall time (see Validation below) — not a hang if a weekly poll looks "stuck" on this asset for 45-60 minutes.
  • 2.5 ingest:udir-elevundersokelsen-mobbing npm script added first, verified via check-every-source-has-an-ingest-script.sh and the real npm run invocation.

Validation​

Real run against the live API, completed 2026-10-03: 702 calls, 2,804 rows written, 60.3 minutes wall time (duration_ms: 3615723). Zero rows silently dropped: 351 regions × 2 grades × 4 measures = 2,808 expected if every call returned data; the 4-row deficit is exactly the one confirmed-empty call (Hægebostad/10th grade), verified directly against Postgres — every other region/grade pair has exactly 4 rows, no partial writes.

A genuinely new finding, not predicted in Phase 1: per-call latency was NOT small and fast as assumed from the Phase 1 exploratory calls — the first ~22 calls returned in 30-45ms, then every call after that settled into a steady ~3.5-6 seconds each, for the rest of the run. No pattern found explaining the transition (not fylke-aligned, not a clean request-count threshold). This changed the real cost of this source from "~702 small/fast calls" (Phase 1's assumption) to "~702 calls, ~60 minutes wall time" — documented prominently in the source's own README and the Dagster asset docstring so this isn't mistaken for a hang later.

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

Tasks​

  • 3.1 Added raw.udir_elevundersokelsen_mobbing to models/indicators/sources.yml.
  • 3.2 indicators__udir_elevundersokelsen_mobbing.sql — kommune_nr/region_kind via region_code_to_kommune_nr/classify_region_code. Passed clean on the first dbt build, as expected — both sentinels resolved exactly as predicted (see Validation below).
  • 3.3 Documented columns in schema.yml; mart_indicators__udir_elevundersokelsen_mobbing.sql api passthrough + marts/api/schema.yml entry.
  • 3.4 dbt build --select indicators__udir_elevundersokelsen_mobbing mart_indicators__udir_elevundersokelsen_mobbing against real loaded data — PASS=16 WARN=0 ERROR=0 SKIP=0 TOTAL=16.

Validation​

Real local Postgres, not an empty schema. Explicitly confirmed by direct query:

 region_code | kommune_nr |       region_kind        | count
-------------+------------+---------------------------+-------
 2100        |            | svalbard                  |     8
 2599        |            | unspecified_within_fylke  |     8

2100 resolves to region_kind='svalbard' and 2599 resolves to region_kind='unspecified_within_fylke', both kommune_nr=NULL, exactly as predicted — no macro change needed.

Full dbt check-suite (all 17 scripts), a full dbt build, the full ingest test suite (351 tests), and the website build all pass clean. The only non-pass results in the full build (extracted_columns_are_still_populated, raw_sources_were_refreshed_recently, the mart_atlas_inventory SKIP cascade) are the same pre-existing environmental staleness confirmed unrelated to every prior source this session — udir_elevundersokelsen_mobbing does not appear in either failing test's result set.

Phase 4: Deploy and verify arrival — DONE (verified 2026-10-03)​

PR #528 merged to main at c4a450c. Image build (run 37116024054) succeeded; both digests read from the release's own uis-artifact.json, not reconstructed, and cross-checked against the full build log before sending:

artifact_digest  sha256:a31efe43001d7bd5d0ee8fb72ffb012b8d9e5e754f0282eb2424c466f0a3de53  ghcr.io/terchris/atlas-data/uis:v20261003-c4a450c
image_digest     sha256:d0cb88bf7473dbcbd585d186b034196ee5171890eac69f5f4b103c020269b573  ghcr.io/terchris/atlas-data:v20261003-c4a450c

lands-with.sh 7de652a..c4a450c output was clean: annual_sources_refresh (re-fetch) then transform_and_publish, plus a template-info pin.

Two tasks filed, per the pin/run split ([[a-pin-and-a-run-are-two-actions]]):

  • Deploy request to imac: urb-agents#1822 — titled explicitly with a "SLOW RUN, NOT STUCK (~60min)" warning, and led with that same warning in the body, before the digests — this source makes ~702 HTTP calls and the real run took 60.3 minutes in dev (see Phase 2's Validation). Names raw.udir_elevundersokelsen_mobbing (2,804 rows as of today's validation run), and — learning from [[mart-prefix-is-never-a-served-endpoint]] (PLAN-014's own closing finding) — predicts only indicators__udir_elevundersokelsen_mobbing as a served relation, never a second mart_indicators__... entry.
  • Pin nomination to ops-dev: urb-agents#1823 — both digests copied verbatim from uis-artifact.json, not reconstructed.

imac's deploy report (urb-agents#1822): annual_sources_refresh SUCCESS in 1410.8s (~23.5 minutes) — well inside the ~60-minute envelope flagged in the request, confirming the warning did its job rather than triggering an unnecessary escalation; the driver pinged twice mid-run and imac confirmed real progress each time instead of guessing. raw.udir_elevundersokelsen_mobbing row count 2,804, exact match. transform_and_publish SUCCESS in 516.9s — watched PostgreSQL's log live throughout both runs for #1810-style corruption; zero events. Regression check against 6 prior sources — all unchanged. No discrepancy this time (unlike PLAN-014's own Phase 4): the corrected row-count prediction (only indicators__udir_elevundersokelsen_mobbing, never a second mart_indicators__... entry) matched exactly.

ops-dev's pin report (urb-agents#1823): tag v20261003-c4a450c pushed to dev-templates main cleanly on the first pass — a routine single-release bump, no stale markers left over this time. The operational.automation discrepancy from PLAN-014's own Phase 4 (urb-agents#1814) is unchanged and correctly not re-litigated — it is unrelated to this source and already tracked in 1PRIORITY.md.

Both independently re-verified against the live public API before closing either task — neither report was taken on its own word:

  • GET /indicators__udir_elevundersokelsen_mobbing?limit=1 → real row (Arendal, 7th grade, value 15.8 — matches the captured fixture exactly); Content-Range confirms 0-2803/2804.
  • GET /meta_sources?source_id=eq.udir-elevundersokelsen-mobbing → served_as: ["indicators__udir_elevundersokelsen_mobbing"], exactly one endpoint.
  • GET /mart_indicators__udir_elevundersokelsen_mobbing → 404, confirming the corrected prediction (no second served endpoint under the mart_ name).
  • Spot-checked both sentinels and the empty-grade finding directly: Svalbard (2100 → region_kind=svalbard, kommune_nr=NULL, value=NULL, suppressed) and Utlandet, uspesifisert (2599 → region_kind=unspecified_within_fylke, kommune_nr=NULL, value=12.8) both match the captured fixtures exactly; Hægebostad (4226) has exactly 4 rows live, all grade=7 — confirming the genuinely-empty-10th-grade finding from dev holds in production too, not a local-environment artifact.
  • Independently fetched dev-templates' own committed template-info.yaml via gh api rather than trusting ops-dev's report — tag, digest, and all headline counts (52 sources, 89 relations, 61 raw tables) confirmed landed correctly.

Both bus tasks closed with the verification evidence attached as comments.

Acceptance Criteria​

  • The mechanism is verified live — Udir's EUG table via statistikkportalen.udir.no, confirmed by direct API calls, not assumed from udir-gsi's shape or the investigation's five-month-old description.
  • Licence independently confirmed — NLOD, re-fetched from Udir's own data portal for this source specifically.
  • The bulk-decode-vs-per-kommune-calls question ([Q1]) is resolved with evidence, not assumed — per-kommune calls, correctness over call count, real cost measured at ~60 minutes wall time.
  • udir-elevundersokelsen-mobbing ingests cleanly with zero rows silently dropped, including the 2100/2599 sentinels and at least one suppressed row both represented — 2,804 rows written, exactly matching 351 regions × 2 grades × 4 measures minus the one confirmed-empty region/grade pair (Hægebostad's 10th grade).
  • indicators__udir_elevundersokelsen_mobbing and its mart build and test clean against real loaded data, with 2100/2599 resolving through classify_region_code exactly as predicted (PASS=16 WARN=0 ERROR=0).
  • udir-elevundersokelsen-mobbing appears in meta_sources.served_as after a real deploy, independently verified via live curl — served_as: ["indicators__udir_elevundersokelsen_mobbing"].
  • Golden-file tests cover: an ordinary kommune, a suppressed row, the Utlandet sentinel, both grades, and a genuinely-empty region/grade pair (found on the real run, not predicted) — 24 tests, all passing.
  • The investigation and 1PRIORITY.md are updated to mark this candidate shipped.

Implementation Notes​

  • This source's response shape is NOT udir-gsi's shape, despite sharing one client. Geography is a column dimension here, not a row dimension — do not port udir-gsi's depth-filtered-radSti approach; it does not apply (confirmed live, a 3-level radSti errors).
  • TrinnID(-10) is NOT "alle" for this report — it silently resolves to the wrong single grade. Always pass an explicit grade id (6 = 7th, 9 = 10th); never carry the -10-means-"alle" assumption from udir-gsi/nav-aap into a new Udir report without checking that report's own filterVerdier first.
  • [Q1] resolved — per-kommune calls (~702 for the latest year), not bulk-decode. The cheaper single-call-per-grade route was rejected because it was never confirmed to carry a distinct per-kommune aggregate leaf, only individual schools — correctness over call count. The real cost of that call count turned out to be ~60 minutes wall time, not request volume — only the first ~22 calls are fast; every call after settles into a steady ~3.5-6s each, confirmed on the real Phase 2 run. Budget for this; it is not a hang.
  • A region/grade pair can be genuinely absent, not suppressed. Confirmed live on Hægebostad's 10th grade: zero rows, no column at all for that grade — not predicted in Phase 1, found on the first real run. parse.ts returns zero rows for this case rather than throwing.
  • One new sentinel confirmed, already has precedent: Utlandet/2599 → classify_region_code's existing unspecified_within_fylke branch, no macro change needed — confirmed with a real test in Phase 3.

Outcome​

Shipped end to end, 2026-10-03: ingest (2,804 rows, zero dropped) → dbt staging and api_v1 publication → live cluster deploy → independently re-verified arrival. Atlas's second Udir source, reusing the statistikkportalen.udir.no client udir-gsi already proved out — but a genuinely different response shape: geography here is a column/filter dimension, not udir-gsi's row hierarchy, resolved in [Q1] as one HTTP call per kommune-equivalent region node per grade (~702 calls) rather than decoding an unverified bulk response. Correctness over call count was the right call, but the real cost of that decision turned out to be wall-clock minutes, not request volume — a genuinely new finding for this session, not anticipated in Phase 1's exploratory calls: per-call latency settles at ~3.5-6s each after an initial fast burst, for ~60 minutes total in dev and ~23.5 minutes in production.

Two genuinely new findings, both caught against real data rather than assumed from Phase 1: a region/grade pair can be entirely absent, not suppressed (Hægebostad has no 10th-grade cohort reporting into this table at all — confirmed in both dev and production); and "Utlandet, uspesifisert" (2599, Norwegian schools abroad) is a sentinel udir-gsi never encountered, falling through classify_region_code's existing unspecified_within_fylke branch cleanly, no macro change needed — alongside the already-known Svalbard (2100) sentinel, which did carry over from udir-gsi.

One deploy-Phase finding applied, not just inherited: PLAN-014's own closing lesson ([[mart-prefix-is-never-a-served-endpoint]]) was used proactively here — the deploy request predicted only indicators__udir_elevundersokelsen_mobbing as a served relation from the start, and that prediction matched exactly, with no discrepancy to flag this time. The deploy request's explicit "SLOW RUN, NOT STUCK (~60min)" warning also did its job — imac confirmed progress on two mid-run driver pings rather than escalating, and the real production run finished in 23.5 minutes, well inside the flagged envelope.

Plugs a sharper, annual, per-grade bullying-prevalence signal at kommune resolution into Report #3 (Youth Outcomes), complementing the existing fhi-mobbing 3-year-rolling aggregate.

Files to Modify​

  • atlas-data/ingest/src/sources/udir-elevundersokelsen-mobbing/manifest.yml (new)
  • atlas-data/ingest/src/sources/udir-elevundersokelsen-mobbing/index.ts (new)
  • atlas-data/ingest/src/sources/udir-elevundersokelsen-mobbing/parse.ts (new)
  • atlas-data/ingest/src/sources/udir-elevundersokelsen-mobbing/fetch_retry.ts (new, copied)
  • atlas-data/ingest/src/sources/udir-elevundersokelsen-mobbing/README.md (new)
  • atlas-data/ingest/src/sources/udir-elevundersokelsen-mobbing/__tests__/ (new)
  • atlas-data/ingest/package.json (ingest:udir-elevundersokelsen-mobbing script — add this BEFORE shipping, verify with the real npm run invocation)
  • atlas-data/migrations/<next>_raw_udir_elevundersokelsen_mobbing.sql (new)
  • atlas-data/dagster/atlas_data/assets/raw_other.py or a Udir-specific asset module (registration — annual cadence, existing weekly-polled job pattern)
  • atlas-data/dbt/models/indicators/sources.yml, indicators__udir_elevundersokelsen_mobbing.sql (new), schema.yml
  • atlas-data/dbt/models/marts/api/mart_indicators__udir_elevundersokelsen_mobbing.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 shipped)
  • website/docs/ai-developer/plans/backlog/1PRIORITY.md (mark shipped)