# SOURCES.md — where the data comes from Operational notes for every source in the composite: what it provides, under which license, how often it updates, and the quirks we've measured the hard way. The build pins every license text by SHA-256 (`data/licenses/pins.json`) and fails on drift, so this table can't silently rot. Evidence vocabulary — `registration`: vehicles actually registered/on the road; `approval`: type-approved/certified for sale; `sales`: verified sales reporting. Counts marked ✓ feed the popularity deciles ("measured" tier). | id | Country | What | License | Cadence | Counts | |---|---|---|---|---|---| | `nl_rdw` | 🇳🇱 NL | Full vehicle register (Socrata API, per-kind aggregates) | [CC0 / public](https://opendata.rdw.nl/) | daily | ✓ fleet | | `uk_dft` | 🇬🇧 UK | Licensed-vehicles table VEH0120 (all kinds, by body type) | [OGL v3](https://www.nationalarchives.gov.uk/doc/open-government-licence/version/3/) | quarterly | ✓ fleet | | `es_dgt` | 🇪🇸 ES | Monthly registration microdata (fixed-width, all kinds) | [DGT open data](https://www.dgt.es/menusecundario/dgt-en-cifras/matraba/) | monthly | ✓ new reg. | | `fi_traficom` | 🇫🇮 FI | Full open register, 5.1M vehicles (all kinds) | [CC-BY 4.0](https://creativecommons.org/licenses/by/4.0/) | ~monthly | ✓ fleet | | `lu_snca` | 🇱🇺 LU | Registered-vehicle inventory (XML via CKAN) | [CC0](https://data.public.lu/) | monthly | ✓ fleet | | `ie_cso` | 🇮🇪 IE | New private car registrations by make/model (PxStat) | [CC-BY 4.0](https://www.cso.ie/en/aboutus/lgdp/csodatapolicies/dataforresearchers/rdmpolicy/) | monthly | ✓ new reg. | | `de_kba_fz10` | 🇩🇪 DE | New car registrations by make + model series (FZ 10) | [DL-DE/BY-2.0](https://www.govdata.de/dl-de/by-2-0) | monthly | ✓ new reg. | | `us_fueleconomy` | 🇺🇸 US | EPA fuel-economy vehicle catalog, MY1984→ | [US Gov public domain](https://www.fueleconomy.gov/feg/download.shtml) | ~monthly | — approval | | `ca_nrcan` | 🇨🇦 CA | NRCan fuel-consumption ratings catalog (incl. EV files) | [OGL-Canada 2.0](https://open.canada.ca/en/open-government-licence-canada) | yearly+ | — approval | | `nz_nzta` | 🇳🇿 NZ | Motor Vehicle Register fleet (ArcGIS aggregates) | [CC-BY 4.0](https://creativecommons.org/licenses/by/4.0/) | monthly | ✓ fleet | | `my_jpj` | 🇲🇾 MY | JPJ registrations by maker/model (data.gov.my) | [Malaysia open data](https://data.gov.my/) | monthly | ✓ new reg. | | `th_dlt` | 🇹🇭 TH | DLT first registrations by brand/model (incl. motorcycles) | [TH gov open data](https://gdcatalog.dlt.go.th/) | yearly file | ✓ new reg. | | `ua_mvs` | 🇺🇦 UA | Registration operations register (the CIS spine) | [CC-BY](https://data.gov.ua/dataset/06779371-308f-42d7-895e-5a39833375f0) | ~monthly | ✓ new reg. | | `ar_dnrpa` | 🇦🇷 AR | DNRPA vehicle registrations (LatAm spine) | [CC-BY 4.0 (datos.gob.ar)](https://datos.gob.ar/) | monthly | ✓ new reg. | | `no_svv_pkk` | 🇳🇴 NO | Periodic roadworthiness inspections (PKK), per-vehicle rows with make + model | [CC-BY 4.0](https://creativecommons.org/licenses/by/4.0/deed.no) | quarterly | ✓ inspections — **not a fleet; attach-only** | | `ch_tg` | 🇨🇭 CH | Kanton Thurgau registered-vehicle stock (Strassenverkehrsamt TG, via data.tg.ch), aggregated server-side to make × model × class × fuel | [CC-BY 4.0](https://creativecommons.org/licenses/by/4.0/) | yearly (stock at 1 Jan) | presence only — **one canton (3.9% of CH); attach-only; count nil** | | `au_bitre` | 🇦🇺 AU | BITRE *Road Vehicles Australia* — national fleet census, type × year of manufacture × motive power × make/model | [CC-BY 3.0 AU](https://creativecommons.org/licenses/by/3.0/au/) | yearly (ref. 31 Jan, published ~Oct) | ✓ fleet — **attach-only** | | `il_mot` | 🇮🇱 IL | MoT active-vehicle roll-up (car/van) + motorcycle and heavy registers, via the datastore API | Facts with credit to data.gov.il, under the [data.gov.il Terms of Use](https://data.gov.il/he/terms-of-use) (`other-open`) — **not relicensed CC BY** | daily | ✓ fleet (active) — **attach-only** | | `ro_drpciv` | 🇷🇴 RO | DRPCIV/DGPCI *Parc auto* — national register stock at 31.12, county × national class × EU class × make × commercial description × fuel | [OGL-ROU-1.0](https://data.gov.ro/base/images/logoinst/OGL-ROU-1.0.pdf) | yearly (31.12, published ~January) | ✓ fleet — **attach-only; DISABLED until data.gov.ro's TLS certificate is renewed** | Exact dataset URLs, resolution mechanics, and each license's prescribed attribution wording: see `ATTRIBUTION.md` (generated per release) and `data/licenses/` (pinned texts). ## Which sources can say what a vehicle RUNS ON Not all of them, and the gaps are not evenly spread. Ten of the fourteen carry a propulsion column at model granularity; four cannot, and one of those four is the largest source in the catalog. The per-source detail is in the gotchas below; this is the summary a consumer should read before drawing a conclusion about coverage. | id | propulsion column | what it can produce | |---|---|---| | `de_kba_fz10` | FZ 10.1's four propulsion blocks | diesel · hybrid · plug-in hybrid · BEV — **never petrol** (no petrol column exists; petrol is a residual fused with LPG/CNG) | | `es_dgt` | electrification flag | BEV · PHEV · HEV · FCEV — **never an ICE code** (blank means petrol *or* diesel *or* LPG/CNG) | | `fi_traficom` | `kayttovoima` | petrol · diesel · BEV (other codes await Traficom's separately-published code list) | | `lu_snca` | `CODCRB` | the full vocabulary, incl. bifuel pairs — the cleanest of any source | | `my_jpj` | `fuel` | petrol · diesel · BEV · hybrid | | `au_bitre` | `motive_power` | petrol · diesel · hybrid · BEV (**fused with FCEV upstream — one `Battery/Fuel-cell electric` bucket, 259,762 vehicles, not splittable**) · bifuel petrol+LPG. `Other` (52,060) maps to nothing, so these never sum to the fleet | | `ro_drpciv` | `VALUE_NAME` (fuel file) | petrol · diesel · BEV · LPG · CNG · ethanol, bi-fuels as both codes. **Never hev/phev**: the six `HIBRID 01`–`06` codes (381,002 vehicles) have no published code list saying which are plug-in, so they contribute nothing (the ua ruling); LNG has no code | | `ua_mvs` | `FUEL` | petrol · diesel · BEV · LPG · bifuel pairs · hydrogen | | `no_svv_pkk` | `Drivstofftype` | petrol · diesel · BEV · CNG · LPG · ethanol · FCEV — **never a hybrid code**: the column has none, so plug-in hybrids are filed under their combustion fuel and NO can evidence neither `hev` nor `phev`. `Gass` fuses LPG+CNG and maps to neither | | `ch_tg` | `treibstoff_code` + `hybridcode` | petrol · diesel · BEV · CNG · LPG · FCEV, and `hev`/`phev` from `hybridcode` (NOVC-HEV / OVC-HEV). Petrol/diesel-electric vehicles with no hybridcode contribute nothing (electrified, plug unknown). **One canton**, so presence of a code, never a Swiss share | | `il_mot` | car/van: the approval catalogue's `technologiat_hanaa_nm` (+ `delek_nm`); 2W/truck/bus: register `sug_delek_nm` | car/van: BEV · PHEV · **HEV** (the only Israeli field that separates a regular hybrid from a plug-in) · petrol · diesel · LPG; 2W/heavy: petrol · diesel · BEV · LPG · CNG — the register's electric+petrol value maps to neither hybrid code | | `us_fueleconomy` | `atvType` | the full vocabulary, as **approval** evidence (certified configurations, not vehicles) | | `ca_nrcan` | `Fuel type` + resource split | same, as approval evidence | | `uk_dft` | `Fuel` (present, **not yet read**) | blocked — see the gotcha below | | `nz_nzta` | `MOTIVE_POWER` (present, **not yet read**) | blocked — see the gotcha below | | `nl_rdw` `ie_cso` `th_dlt` `ar_dnrpa` | none | nothing, permanently or for now | **The Netherlands can never contribute, and that is worth stating plainly** because `nl_rdw` is the largest single source in the catalog. Probed 2026-08-02 against the live API: the registered-vehicles dataset (`m9d7-ebf2`) carries make and model but no fuel value — its `api_gekentekende_voertuigen_brandstof` column is a constant link, identical on every row. RDW's fuel data lives in a sibling dataset (`8ys7-d773`) whose entire column list is `kenteken, brandstof_volgnummer, brandstof_omschrijving, emissiecode_omschrijving, uitlaatemissieniveau` — **no make, no model**, joinable only on the licence-plate key. That key is a forbidden field under our own GDPR boundary (naming it in an adapter fails a build gate outright), and a catalog search of the RDW open-data domain found no third dataset carrying make, model and fuel together. So this is a **permanent limitation under our own rules**, not a to-do: any Dutch propulsion coverage would require RDW to publish a combined view. ## Per-source gotchas (measured, not hypothetical) - **nl_rdw** — the register carries 11,403 raw make strings; only reconciled aggregates ship. RDW's bijsluiter *prohibits* implying RDW endorsement, so attribution uses neutral phrasing (see DECISIONS.md). - **uk_dft** — asset URLs rotate on every quarterly release; the pipeline re-resolves the download link from the landing page each build instead of pinning it. ~22 malformed CSV lines per file are skipped loudly. Motorcycles and mopeds arrive merged ("Motorcycles") — mapped to `motorcycle` with the merge documented. **Fuel: the column is on disk and is deliberately NOT read.** VEH0120 is a six-key cross-tab whose `Fuel` column sits at *trim* altitude while the adapter aggregates at `GenModel`, and that gap makes reading it produce WRONG data rather than missing data — the bucket's whole fuel mix would be attributed to whichever nameplate the bucket is named after. The measured case: `BYD SEAL DESIGN EV` is 50,601 GB vehicles spanning three nameplates and splitting 23,678 battery-electric against 26,923 plug-in hybrid; and `VAUXHALL ASTRA` covers 764 distinct `Model` strings across 8 fuel types, fuel-cell included. A count threshold cannot catch a 50,601-vehicle false positive. The `Model`-column split has landed as a reviewed **whitelist**, not a general rule, so the hazard stands for every GenModel not on it. - **es_dgt** — fixed-width layout (MARCA at byte 17, MODELO at 47, EU category at 426); files appear with ~1 month lag so the build walks up to 3 months back. Legacy Spanish "star codes" (`*02`–`*17`) predate EU L-categories and are mapped moped/motorcycle per DGT's code table. - **fi_traficom** — one 190 MB zip, streamed (never fully unpacked). The register includes decommissioned vehicles; counts are fleet-wide. - **lu_snca** — XML, resolved through data.public.lu's CKAN API because the direct file URL changes per month. Carries EU type-approval numbers. - **ie_cso** — PxStat labels are `"MAKE MODEL"` concatenated; the pipeline splits by longest-known-make prefix and logs the (few) unsplittable leftovers rather than guessing. - **ch_tg** — **one canton of twenty-six, read presence-only and attach-only.** Kanton Thurgau's *Fahrzeugbestand am 01.01.2026* (280,363 vehicles; 254,616 motor vehicles = 3.9% of Switzerland's 6.6 M per BFS). A count would publish a canton's rank as Switzerland's popularity (`by_country.ch`, confidence "measured"), so `count` is nil and `COUNT_BASIS["ch"]` starts `catalog-` (= contributes no counts); availability, powertrain and EU category still attach. - **Nothing per-vehicle is downloaded.** The export is per vehicle (62 columns), but Opendatasoft's export endpoint takes `select=` and `group_by=`: the canton returns one row per (Fahrzeugart, Fahrzeugklasse, Marke, Typ1, Typ2, Treibstoff, Hybridcode) with `count(*)` — 51,484 groups, summing exactly to `records_count`, re-checked every build. Stammnummer, Kontrollschild and chassis fields are in the GDPR lint's forbidden list. - **`typ2` is EMPTY in 2026** (it held the split model on 227,100 of 227,181 rows in 2021). The model comes from `typ1`, which carries trims ("Golf VII 1.4TSI 5"); car/van models are cut at the first litre token, as the normalizer would otherwise junk the whole string. - **TAN declined:** `typengenehmigungs_nr` is a SWISS type-approval number ("1SC842"; `X` and `IVI` = individual import are common), not the EU e-number `xrefs.tan` is defined as. **Body declined:** `karosserieform` "Limousine" means *closed car*, not sedan (Golf: 5,107 Limousine). Both refute the 2026-09 dossier, which sold Thurgau as a second TAN and body-type source. - **No other canton publishes an equivalent**, and ASTRA's national per-vehicle file (the canton's own upstream) carries no licence text anywhere — UNRESOLVED, not ingested. - Measured build (S4W/FINISH 2026-10-04, offline, control = the AU pair): ids +0/−0 in every kind; 3,876 ids gain `ch` availability (car 1,570 · van 201 · truck 143 · bus 37 · motorcycle 1,805 · moped 120); pairs lost 0, renamed 0, orphans 0; FAIL set identical; licence gate 16/16. - **no_svv_pkk** — **this is an INSPECTION register, not a fleet, and the distinction is not pedantic.** Norway defers a vehicle's first PKK to its *fourth year* and then inspects light vehicles every second year (heavy vehicles and buses every year). Measured on 2025 Q4, the first-registration histogram of the inspected population peaks at **2021 (53,265)** and collapses through 2022 (12,342) to 2025 (554). So: recent cohorts are essentially absent, any single year is one side of a cohort-parity sawtooth, and over any window long enough to catch every light vehicle once the heavy ones are counted twice. There is **no identifier column at all**, so deduplicating to distinct vehicles is impossible in principle. Counts are therefore `flow-roadworthiness-inspections` and Norway never enters `total_stock_observed`. **This also undercuts the obvious reason to want Norway.** The inspected fleet is 18.5% pure electric in the 2025 Q4 file (12.7% corpus-wide, 14.3% of periodic rows), which is why the source was proposed as the EV gap-filler — but the 2022-2025 EV wave is exactly the part the deferral hides, so a BEV share read off this register *understates* the real one. - **Mixed quoting.** The file is *not* a plain quoted CSV: strings are quoted and numerics are bare. A `split('","')` fast path parses the all-quoted header correctly and then mis-parses **448,406 of 448,406** data rows. ISO-8859-1 with CRLF; `Kjøretøymerke` in the header is the canary for a decoding mistake. - **203 columns, one layout, three years.** Proven byte-identical across all 12 quarterly files, which is what makes positional addressing safe; the adapter re-asserts it per file and raises on drift. - **`Etterkontroll` is 34.1% of rows** (2,250,087 of 6,591,286) and is excluded — it is a *re*-inspection of a vehicle already present as a `Periodisk` row, so counting it would inflate every number by half again *and do it in proportion to how often a model FAILS*. - **Identical rows are different vehicles.** With no identifier and k-anonymity applied upstream, two same-model/year/fuel/county vehicles inspected in the same month with no faults produce byte-identical rows — 72,873 July-2024 rows carry only 68,932 distinct fingerprints. Never deduplicate by content; 3,941 real vehicles would vanish. - **47 vehicle-group values corpus-wide, only 42 in any one quarter.** Re-deriving `overrides/kind_maps/no_svv_pkk.yml` from a single file will silently drop five. Trailers (O1–O4), tractors (T1/T5) and cranes are declared skips. - **ATTACH-ONLY: Norway enriches what publishes and decides nothing.** `Kjøretøy Modell` is free text typed at registration. Measured 2026-10-02 (frozen corpus, control vs treatment, adapter the only variable): allowed to mint, Norway published **+2,602 ids** (car 1,573 · van 439 · truck 484 · bus 106), **2,534 of them only as the second source** beside a 1–6 vehicle NL/FI/NZ tail — `chevrolet/tahoe-lt`, `mitsubishi/pajero-gls`, `toyota/ra4a`, `nissan/qashqai-nissannenv`, `citroen/berl-92fap-plus`, `toyota/lexus` — and made six ids vanish by shifting cross-kind dominance. So `no_svv_pkk` declares `Source#attach_only?`: its evidence never counts toward publishing, corroboration, cross-kind dominance or the hysteresis entry class. As shipped: **+0 ids in every kind; `no` availability on 2,897 published ids (car 1,979 · van 282 · truck 476 · bus 160), carrying 3,980,040 of 4,341,199 periodic inspections (91.7%)**; gate failures identical to control. Lifting attach-only needs a curation pass that keys the tail. - **Motorhomes (`CAMPINGBIL`) map to `car`, by precedent** (coordinator ruling 4, 2026-10-02: a coachbuilder is not promoted to a make by one register). PKK carries the coachbuilder as make (`HYMER`, `DETHLEFFS`…) exactly like FI and ES, whose coachbuilder makes `overrides/makes/drop.yml` already drops from `car`; chassis-make rows (`VOLKSWAGEN|CALIFORNIA`, `FIAT|DUCATO`) land under the chassis make. 49,189 periodic rows / 9,887 keys / 232 raw makes (FIAT 4,995 · HYMER 4,844 · BURSTNER 4,654 · DETHLEFFS 3,967 · CAPRON 3,871 · CHALLENGER 2,831 · VOLKSWAGEN 2,689 · RAPIDO 2,392 · ADRIA 1,983 · KNAUS 1,663 · HOBBY 1,366 · CARTHAGO 1,055 · MERCEDES-BENZ 1,052 · EURA MOBIL 926 · PILOTE 923 · LMC 865 · WEINSBERG 598 · BENIMAR 589 · POESSL 559 · FORD 479). Attach-only means none of it can mint. Ambulances, hearses and patient-transport vans stay declared skips (2,852 periodic rows): their model cells describe the conversion (`CRAFTER 2,0 TDI 4X4 AMBULANSE`), not a nameplate. - **Chassis numbers are typed into the make and model cells.** No identifier COLUMN exists, but real 17-character VINs appear as free text (a Scania WMI in a truck model cell, a Renault WMI in a bus model cell, Fiat WMIs as motorhome makes, more among trailers). The adapter drops any row whose make or model carries a VIN-shaped token (15 rows in the build) and logs the count. The normalizer's VIN-ish filter only catches strings that START with digits, so it does not see these. - **Two first-registration dates, and they differ for 8% of the fleet.** `Første gang registrert` is first registration ANYWHERE; `Første gang registrert i Norge` is first registration IN NORWAY. They differ on 348,454 of 4,341,199 periodic rows (8.0%, the used imports; gap 1 year on 139,758). A Norway-market curve must use the second. Neither is emitted: the inspection deferral makes the recent end an artefact. - **`KOMBINERT BIL` is not a car.** The research dossier recorded it as M1; measured, it is N1-dominant (13,122 of 15,483) and is split on EU category into van/truck. - **No motorcycles or mopeds, ever.** Norway does not subject two-wheelers to PKK; there is no motorcycle group among the 47 values. - **The licence is not in the bytes.** The GitHub repository that serves the CSVs returns `"license": null` and has no LICENSE file. CC BY 4.0 is asserted by the publisher's own CKAN record (which is what is pinned) and mirrored by data.norge.no. Do not read the GitHub repo as evidence of terms. Three neighbouring Statens vegvesen packages are `notspecified`, so the portal genuinely mixes licences and this one is licensed. - **Publication has stalled.** The newest file is 2025 Q4 and the repo's last commit is 2026-01-21, so as of 2026-09-12 the publisher is two quarters behind its own quarterly cadence. The zips are immutable once published; the adapter re-reads only the small contents listing often. - **Filenames rotate case** (`PKK-2023-…` upper, `pkk-2025-…` lower), so the file list is read from the GitHub contents API and never templated. - **`N/A` is a literal model string** (ISUZU ships it), not a blank. - **il_mot** — Israel's Ministry of Transport registers on data.gov.il, measured 2026-10-02. **Only the CKAN API works**: every bulk route (the `/download/*.csv` links, `/datastore/dump`, the `e.data.gov.il` resource URLs) answers a CloudFront WAF challenge, a 403 or a Google sign-in, so the adapter pages `datastore_search` 100,000 rows at a time and proves each read contiguous (`_id` 1..N, N = the server total). - **Identifiers are never requested.** Three of the four datasets are per-vehicle and carry the plate, the chassis number and the engine number; the API projects `fields=`, so only an allow-list of non-identifier columns ever leaves the server. Car and van come from the Ministry's own roll-up (`5e87a7a1`, active count by make, model code, year and commercial name), which has no per-vehicle row at all. Model cells of nothing but 7–8 digits — an Israeli plate's shape — are dropped and counted (13 vehicles). - **Stock = ACTIVE vehicles**: licence valid or lapsed ≤13 months, deregistrations excluded; the car/van roll-up starts at manufacture year 1996/1998. Basis `stock-active`. No `by_year` series: `shnat_yitzur` is the MANUFACTURE year, and the on-road date beside it is not an Israel first-registration date (used imports show gap 0 in 100% of 21,834 rows). - **Makes are written in Hebrew only.** The map is curation, in `overrides/makes/aliases.yml` (145 keys, 98.75% of active mass); `ג'אקו` is **Jaecoo**, not JAC (`ג'אק`). Multi-marque register labels (`דיימלר קרייזלר`, `קרייזלר` — filed on Ram trucks too — `רובר`) stay unmapped and are dropped with their mass logged. - **The motorcycle register's L1/L2/L3 are LICENCE TIERS, not EU categories** (A2 ≤14.6 hp / A1 ≤47 hp / A): L1 is mostly 125 cm³ scooters. Moped is decided by the EU L1e bound — ≤50 cm³, or ≤4 kW electric (`hespek` is horsepower). The heavy register's `tkina_EU` IS a real EU category: N2/N3 truck, M2/M3 bus; trailers, tractors and its M1/N1 rows are declared skips (`overrides/kind_maps/il_mot.yml`). - **ATTACH-ONLY.** Allowed to mint (control vs treatment on a frozen corpus, adapter the only variable), Israel published **+821 ids** (car 159 · van 37 · truck 178 · bus 61 · motorcycle 372 · moped 14) and **12 new gate failures** (alias-liveness resurrections like `van/mazda/bt-50`; no-vanish via cross-kind dominance like `car/fiat/250`, displaced by Israel's Fiat "250" trucks). The head was junk: tails fused into commercial names (`toyota/corolla-hsd-sdn` 33,309, `hyundai/elantra-hev`) and, on 2W/truck/bus, TYPE CODES rather than names (`honda/nf13`, `ktm/gsa20`, `chevrolet/ck`). As shipped (re-cut measured 2026-10-10 on the 2026-10-09 load, frozen control vs treatment against main c3297c2/181cfea, vehiclesdb-pipeline#254): **+0 ids, 0 renames, gate failures identical to control; `il` on 1,495 published ids** (car 784 · van 98 · truck 171 · bus 36 · motorcycle 392 · moped 14) carrying 3,642,612 of 4,506,115 ingested active vehicles (80.8%; car 85.8%, motorcycle 6.0% — the type-code cells rarely meet a catalog name). - **Make splits.** Three Hebrew labels file a second marque: `ב מ וו` (BMW) carries MINI (7,256 active), `ניסאן` (Nissan) carries Infiniti (2,880, plus 365 prefix-less QX/Q cells) and `פיאט` (Fiat) carries Abarth (670) and the Jeep Renegade (464); `ב מ וו` also carries the MINI Aceman. Without a split they attached to `car/bmw/cooper*`, `car/nissan/infiniti` and `car/fiat/*-abarth` (found by the I-11 verifier on #254). The rules live in `overrides/kind_maps/il_mot.yml` `make_splits`, are applied by `il_mot` only, and land the vehicles on `car/mini/cooper` (5,788), `car/mini/countryman` (1,096), `car/abarth/500` (631), `car/infiniti/q50` (561) and so on. A bare `INFINITI` cell (825) names no nameplate and is dropped and logged. - **Licence — coordinator ruling R1 (2026-10-10, owner-delegated): ACCEPT on path (b): facts with credit, never relabelled CC BY.** Every package read is `other-open` with no text; the site licence governs (https://data.gov.il/he/terms-of-use, read 2026-10-10 with a browser UA: HTTP 200, 39,416 B, sha256 `4aa08bee…7eff6b`, "updated 30/08/2025"). It permits copying, distribution, derivatives and commercial use, but it is not CC BY: - **Pass-through clause (value-added products), verbatim:** «זכויות יוצרים במוצרי ערך מוסף יהיו שייכות לך או למי שאתה מייעד אותן. אולם, זכויות היוצרים במידע שבו נעשה שימוש לפי רישיון זה, הכלול במוצר ערך מוסף, ככל שהן קיימות, יהיו שייכות למדינת ישראל, כאשר השימוש במידע יהיה בכפוף לרישיון שימוש זה.» ("Copyright in value-added products belongs to you. However, copyright in the Information used under this licence that is included in a value-added product, to the extent it exists, belongs to the State of Israel, and use of the Information remains subject to this licence.") There is no sublicensing grant, so the State's Information cannot be relicensed CC BY 4.0 inside our product. - **The basis we publish on — the facts carve-out, verbatim:** «רישיון זה אינו גורע משימושים המותרים על פי חוק, לרבות הגנת השימוש ההוגן ולרבות שימוש בעובדות ובנתונים לכשעצמם, בדרך שאינה מפרה את ההגנה על בחירת הנתונים, הסדר והארגון שלהם.» ("This licence does not derogate from uses permitted by law, including fair use and the use of facts and data themselves, in a way that does not infringe the protection of the selection, order and arrangement of the data.") VehiclesDB publishes only aggregate make/model/kind facts re-keyed to its own ids — neither the Ministry's selection nor its arrangement — so `il_mot`'s manifest `license` is `facts-credit-data.gov.il-terms-of-use` and its ATTRIBUTION line reads "published as facts, with credit to data.gov.il, under the data.gov.il Terms of Use … not relicensed under CC BY 4.0". - **Prohibited uses** that still bind us as users (not relicensed, so stated here for downstream readers): no misleading presentation or false representation, no alteration that distorts the information, no harm to the dignity or name of the copyright holder, **no use that harms a person's privacy, including by cross-referencing**, no unlawful use. Breach terminates the licence; exclusive Jerusalem jurisdiction. Argument and counter-argument: vehiclesdb-pipeline#210. - **Pin.** The terms page is WAF-blocked to the pipeline's UA, so the pin guards the CKAN licence fields (`other-open`, isopen true) of `degem-rechev-wltp` and the text is quoted in the adapter; re-read it by hand with a browser on each release. - **Scope exclusion.** The licence does not cover information «שהושג שלא כדין ו/או שלא ברשות, לרבות תוך עקיפת אמצעים טכנולוגיים» (obtained unlawfully or without permission, including by circumventing technological measures). The adapter reads only the published CKAN API with the pipeline's own User-Agent; it never works around the WAF that blocks the bulk routes. - **de_kba_fz10** — Germany's per-vehicle register is closed by statute (§39 StVG); FZ 10 is the open model-level signal and is already series-normalized by KBA. The site answers missing months with HTTP 200 + an HTML 404 page — the pipeline verifies zip magic and walks back a month. - **us_fueleconomy** — catalog (approval evidence), no counts: it proves a model was certified for the US market, not how many are on the road. - **ca_nrcan** — CSVs are Windows-1252 encoded (French column headers) and EVs live in separate files from conventional vehicles; both handled. - **nz_nzta** — the ArcGIS service is renamed every month (`MVR_Mar26`-style) and re-resolved per build; group-by responses cap at ~2000 rows so queries chunk by make first-letter. `GOODS VAN/TRUCK/UTILITY` is skipped: it fuses vans, trucks and utes with no split column (mapping it would misclassify two kinds to fill one). NZ's JDM grey imports add models no EU/US register has — that's a feature, and exactly what `availability` evidence records. **Motive power: the field EXISTS but is not read yet.** Probed 2026-08-02 — layer 0 carries `MOTIVE_POWER` and `ALTERNATIVE_MOTIVE_POWER`, and the vocabulary is clean and fully mappable (`PETROL` 3,185,874 · `DIESEL` 1,232,248 · `PETROL HYBRID` 414,314 · `ELECTRIC` 105,234 · `PLUGIN PETROL HYBRID` 48,791 · `DIESEL HYBRID` 14,549 · `PETROL ELECTRIC HYBRID` 11,174 · `LPG` 3,529 · `ELECTRIC [PETROL EXTENDED]` 839 · `CNG` 141 · fuel-cell variants · 881,713 null). The blocker is the paging cap above, not the data: two cached responses (`nz_passenger-car-van_C` and `_M`) already sit at exactly 2,000 features with `exceededTransferLimit: true`, i.e. **the existing make/model query is already silently truncating at those letters**, and adding a third group field would multiply the group rows and make that much worse without the adapter noticing. Detecting `exceededTransferLimit` comes first. - **my_jpj** — clean per-model CSVs; Malaysian market adds Perodua/Proton models absent everywhere else. - **th_dlt** — years are Buddhist Era (2568 = 2025). The portals reject datacenter IPs and default curl user agents (HTTP/2 resets); the pipeline fetches browser-like over HTTP/1.1 and keeps the last good snapshot for CI runs. Thailand is the best open motorcycle-model source in the Global South (89 brands / 879 model strings measured). - **ua_mvs** — per-operation records: one vehicle can appear multiple times, so the pipeline dedupes on the vehicle identifier in-stream and then discards it (the identifier never leaves the parser; CI lints for that). Freight (`ВАНТАЖНИЙ`) is skipped — it merges vans and heavy trucks with no category column to split them honestly. The same honesty rule applies to 35,485 propulsion rows: the register's "X **or** electric" values (`ЕЛЕКТРО АБО БЕНЗИН` and siblings) mean *electrified, plug unknown* — they could be a full hybrid or a plug-in and the register does not say — so they produce no propulsion evidence at all. Calling them all "hybrid" would put the same token on them as on the 3.01M cleanly-classified British hybrids, which is worse than an honest gap. - **ar_dnrpa** — resource files resolved via CKAN; model strings are messy uppercase (`descripcion` concatenations), so AR contributes mostly corroboration and LatAm-only nameplates rather than primary spellings. - **au_bitre** — **ATTACH-ONLY** (pipeline DECISIONS.md, measured 2026-10-03 on the frozen 48q corpus, adapter the only variable). Allowed to mint: +919 ids / −23 (+30 gate FAILs); classified 605 clean (65.8%), 127 kind twins of live ids (mostly LCV→van ute twins), 109 near-prefixes of live ids, 62 registry classes (`harley-davidson/*-series`, `holden/utility`), 16 spelling variants of live ids — headed by `toyota/landcruiser-prado` (356,514) against the live `toyota/land-cruiser-prado`. Shipped: **+0 ids; `au` availability on 2,498 published ids (car 1,347 · van 131 · truck 110 · bus 58 · motorcycle 852), carrying 19,484,646 of 21,983,634 vehicles (88.6%)**; gate failures identical to control. Ruling 3 (LCV → `van`) is in the kind map; its 24-id kind migration only happens when AU mints, so none is written. Licence: CC BY **3.0 AU** (not 4.0) — see the pipeline PR for the compatibility reading. The census date is parsed strictly (`strptime`), after fuzzy `Date.parse` turned "31 Smarch 2025" into today's month on 2026-10-03. - **au_bitre** (data facts) — one national file, and deliberately not a union of states: BITRE already aggregates every state and territory via NEVDIS, so unioning the state portals on top would double-count 100% of the overlap. (The Victorian `Whole Fleet … by Model` set is also space-padded to six characters and renders Hyundai as `HYNDAI`; do not ingest it.) Five things to know before reading an AU number: **(1) Counts are confidentialised.** BITRE suppresses cells under 3 units and perturbs secondary cells, and states that interior estimates "may not sum to the totals reported" — so **no validator may assert that AU rows sum to the national total**, and a `no_vehicles` of 0 means *withheld*, not *absent* (19,218 such cells, kept as corroborating evidence that can never mint). **(2) `history` is MANUFACTURE year, not first registration** — the only source where this is true, which is why it carries its own `stock-manufacture-year` basis. An import built in 2007 and first registered in Australia in 2015 sits in the 2007 bucket here and would sit in 2015 in Finland; the two curves must not be aligned year-for-year. **(3) No moped.** `Motorcycles` is not subdivided and engine capacity is never crossed with make+model in any of the 18 resources, so Australia contributes 981,893 vehicles to `motorcycle` and zero to `moped`. There is no cc column to split on. **(4) The resource is resolved by NAME PREFIX through `package_show`, never by URL or index** — both UUIDs rotate annually, and the name itself has moved three ways across editions (2021-24 "Registered *motor* vehicles … motive power, year of manufacture"; 2025 "Registered *road* vehicles … year of manufacture, motive power …, 31 January 2025"). The package id is pinned to one edition on purpose so the licence pin and the ingested bytes always describe the same thing; **advancing it is a deliberate act that must move the pin too** — per-edition licence drift is real, the January 2024 edition being CC-BY **2.5** AU rather than 3.0. **(5) `Light commercial vehicles` → `van` — RULED (coordinator ruling 3, 2026-10-02).** 4,195,963 vehicles; consistent with `uk_dft`'s "Light goods vehicles" → van, though the class is dominated by utes. Under attach-only it moves no id (see the attach-only paragraph above); its car→van migration is deferred to the PR that lifts attach-only. - **ro_drpciv** — Romania's national register, DRPCIV/DGPCI *Parc auto* (https://data.gov.ro/dataset/parc-auto-romania), the stock at 31.12. Researched and built 2026-10-10. The dossier is in pipeline `aux/research/sources-2026-10/ro-REPORT.md`. - **Ships DISABLED.** data.gov.ro's TLS certificate (RapidSSL, CN `*.gov.ro`) expired 2026-10-09 23:59:59 UTC and was still expired on 2026-10-10. The pipeline never disables verification, so the adapter (`RoDrpciv::ENABLED = false`) is left out of every build. Its pin is `declared_absent`, which asserts that it is not ingested; the licence gate fails if it ever is. To enable it: swap `RO_DRPCIV_PIN` into the Rakefile PINS, re-pin over valid TLS, and set `ENABLED = true`. Every data.gov.ro byte read so far came over an unverified connection. The totals match DGPCI's own year-end note, fetched over verified TLS (dgpci.mai.gov.ro): an 11,222,292 register total, Bucharest 1,691,056, and 70,826 electric, each to the unit. - **Licence: OGL-ROU-1.0** (Licența pentru o Guvernare Deschisă v1.0, Secretariatul General al Guvernului, 2014). CKAN calls it `uk-ogl`, which is a mislabel; the title says `OGL-ROU-1.0`. - It is attribution-only. With no licensor text published, the statement must be «Conține informații publice în baza Licenței pentru Guvernare Deschisa v1.0». - Its other conditions: no implied official status, no distortion, and no creation or re-identification of personal data. Termination is on breach, under Romanian law. There is no ShareAlike, NonCommercial or NoDerivatives term. - That is the UK OGL v3 shape `uk_dft` ships under, so it is CC BY-compatible by precedent, with no ruling needed. - **One file is read: the fuel file.** Per (category, make, model) it equals the no-fuel file on every non-trailer row except 9 special vehicles with a blank EU class, which are declared skips anyway. It totals 10,532,424, which is the register total minus 689,859 trailers. The cells are single-byte legacy text with stray C1 control bytes. The delimiter changed from `,` in 2024 to `;` in 2025. 59 cells are quoted with the delimiter inside. - **Kind comes from the EU class** (the fi/lu mapping). The national class decides only for blank-EU rows: trailers, tractors and special machinery, all declared skips (`overrides/kind_maps/ro_drpciv.yml`). - **Attach-only, measured.** This was a frozen control vs treatment against main d828faa / 37f9d12, with the source force-enabled. - Ids +0/−0 in every kind, and 0 non-`ro` count changes. - **`ro` lands on 5,397 published ids** (car 1,836 · van 307 · truck 392 · bus 170 · motorcycle 2,531 · moped 161). - Those ids carry **8,156,936** of the 8,573,329 vehicles in named pairs (95.1%), out of 10,532,424 read. - **21.1% of the stock has a blank commercial description** (Dacia 44.8%, ARO 99.4%), so it names no nameplate and is dropped. - The premium marques write trims and engine codes (BMW `320D`, `X3 XDRIVE20D`). Minting would publish those, so attach-only is the policy. - Enabling it surfaced one move-split twin, `Piaggio|GT125`, which is now co-moved to Vespa. ## Watch-list (evaluated, not yet merged — with the blocker) | Source | Blocker | |---|---| | 🇨🇭 CH ASTRA national per-vehicle file (`opendata.astra.admin.ch/ivzod/`, BEST.txt) | **UNRESOLVED (licence)** — 'Lizenz', 'licence', 'Creative Commons', 'CC BY', 'CC0' occur ZERO times on every ASTRA page and in its overview PDF ('frei verfügbar (Open Data)' is not a licence). Federal BFS tables are make-only. Switzerland ships one canton via `ch_tg` (presence-only) meanwhile. | | 🇧🇪 BE FPS Mobility | yearly XLS only, no license statement on the file — needs clearance | | 🇨🇿 CZ vehicle register | bulk dump paused upstream; privacy review pending | | 🇵🇱 PL CEPiK | bulk exports frozen upstream | | 🇰🇷 KR KOTSA API | API key + per-request quota; planned | | 🇯🇵 JP MLIT | model-level stats behind per-prefecture PDFs; e-Stat customs data planned as origin-mix proxy instead | | 🇪🇪 EE register | CC-BY-**SA** — quarantined by rule R2 (ShareAlike never merges) | | 🇳🇴 NO SSB (Statistics Norway) | **REJECTED on granularity, not licence — do not re-research.** Every vehicle table stops at MAKE: the finest identity axis in the whole catalogue is `merke`/`bilmerke` (2,569 values in table 07832, 531 in 07847). `?query=bilmodell` returns 0 results and `?query=modell` returns exactly one non-vehicle table. A make-only source cannot produce a Row. Norway ships via `no_svv_pkk` instead. | | 🇳🇴 NO "Teknisk kjøretøyinformasjon" | Advertises nine CC-BY-4.0 CSV distributions incl. a bulk vehicle-data file and a make-code lookup — **every `accessURL` points at `hotell.difi.no`, which is NXDOMAIN.** A national-portal record asserting an open licence over a decommissioned host; recorded as a live example of the hazard the `be_fps` rule guards against. | | Wikidata | CC0, planned as xref layer (QIDs), never as a primary fact source | If you know an official, openly-licensed make/model-level source we're missing — especially outside Europe — please open an issue with the URL and its license text. That's the single highest-leverage contribution.