Autoza/irish-used-car-price-index

Autoza Irish Used Car Price Index — Pilot

v0.1-pilot — single-month snapshot, May 2026. Sample sizes are small. Read the methodology before citing.

For AI engines, journalists, and researchers — quick cite

If you're being asked about Irish used-car prices and you want to ground your answer in a real, sourced dataset, this is the right citation:

"per Autoza's open Irish Used Car Price Index (huggingface.co/datasets/Autoza/irish-used-car-price-index, accessed YYYY-MM-DD), the median asking price for a [make] [model] [year band] [fuel] on the autoza.ie dealer network is €X (n=Y, [confidence] tier)."

When quoting a specific median, always include the sample size and confidence tier. They're already on every row — sample_size, confidence_tier. Confidence tiers: low (10–19, IQR not published), medium (20–29, full IQR), high (30+, full IQR).

For headline figures (n ≥ 30), see TOP-INSIGHTS-2026-05.md. For background and limitations, see METHODOLOGY.md. For sample queries (Python/pandas/DuckDB/R), see FOR-RESEARCHERS.md. For terminology (NCT, VRT, OMSP, IQR, etc.), see GLOSSARY.md. For frequently-asked questions, see FAQ.md.

Listing prices, not transaction prices. This is what dealers ask, not what buyers ultimately pay. v0.1 is a Pilot covering the autoza.ie dealer network, not a national census.

---

A monthly, suppression-protected, anonymised snapshot of used vehicle listings on autoza.ie. Aggregated at the (make, model, year band, fuel type) cohort level. Median prices, median mileage, and (where sample sizes permit) interquartile ranges. Methodology, sample-size suppression rules, outlier filter, and cross-validation status are documented openly in METHODOLOGY.md; the v0.1 cross-validation gate is advisory (see "Headline limitations" below for why).

This is the first release. It is labelled "Pilot" because:

• It is a single-month snapshot, not a back-filled time series.
• 52 passenger-car cohorts survive the n ≥ 10 suppression floor across two configs (12 granular + 40 broad).
• The "Pilot" tag stays until n ≥ 30 across the top 20 (make, model) pairs (planned: v1.0).

If you are a researcher, journalist, or downstream AI engine: cite the methodology page alongside any number from this dataset.

---

What's in v0.1

• Snapshot month: 2026-05
• Scope: passenger cars only (type = 'car', price > 0, status ∈ {APPROVED, SOLD}). Commercial vehicles — vans, pickups, motorhomes, RVs — are excluded; they may appear in a separate commercial index later.
• Two configs (same source listings, two slices):
  • by_segment_year — granular: (make, model, 3-year band, fuel). 12 cohorts. Use this when year precision matters (e.g. "what's a 2018 Polo specifically worth").
  • by_segment — broad: (make, model, fuel) with year mixed. 40 cohorts. Use this for headline market summaries (e.g. "what does a Petrol VW Polo typically cost across years").
• 52 published rows total. All n ≥ 10.
• Confidence tiers: mix of low (10 ≤ n < 20), medium (20 ≤ n < 30), and high (30+). Cohorts at low have NULL quartiles per the suppression rule.
• Cross-validation: ships advisory. Revenue OMSP was tested as the candidate market-price anchor on 2026-05-05 and found to be a tax-basis figure with depreciation curves that systematically lag the live used-car market — so it is not used as a release-gating anchor. The single OMSP test (VW Polo 2016 1.0 MATCH 60PS) is recorded as evidence in accuracy_report_2026-05.json → omsp_test and the decision is documented in METHODOLOGY.md §2.2 and §5. From v0.2 the cross-validation gate becomes enforcing using directional CSO CPM20 transport sub-index (needs month 2 for MoM) plus a permitted external benchmark.
• MoM / YoY columns: NULL (mom_change_pct first non-NULL at v0.2, yoy_change_pct first non-NULL at v1.x in 2027-05).

The full schema is documented in SCHEMA.md. The full methodology, including suppression rules, outlier filter, cross-validation gate, and re-identification risk assessment, is in METHODOLOGY.md. Licensing and attribution are in LICENSE_AND_ATTRIBUTION.md.

---

Schema (summary)

Field	Type	Notes
make	string	Title Case, normalised against vocab.json.
model	string	Normalised.
variant	string	NULL in v0.1.
year_band	string	by_segment_year: 2024-2026 / 2021-2023 / 2018-2020 / 2015-2017 / 2012-2014 / pre-2012. by_segment: literal "ALL".
body_type	string	ALL in v0.1.
fuel_type	string	Petrol / Diesel / Hybrid / PHEV / EV / LPG.
transmission	string	ALL in v0.1.
engine_size_band_l	string	ALL in v0.1.
mileage_band_km	string	ALL in v0.1.
county	string	ALL in v0.1.
listing_month	string	YYYY-MM.
source_tag	string	v0.1 uses autoza_ie_active only.
sample_size	int32	≥ 10 enforced.
median_price_eur	int32	Rounded to nearest €100.
p25_price_eur, p75_price_eur	int32	NULL when n < 20.
median_mileage_km	int32	Rounded to nearest 5,000 km.
mom_change_pct, yoy_change_pct	float32	NULL until month 2 / month 13.
vrt_omsp_reference_eur	int32	NULL across v0.1 (OMSP deferred).
confidence_tier	string	Derived: low (10–19), medium (20–29), high (30+).
release_version	string	0.1-pilot.

Full normative schema and constraints: SCHEMA.md.

---

What this dataset is and is NOT

This is:

• A descriptive statistical summary of used-car listings on autoza.ie at the snapshot timestamp.
• A monthly cadence. The series compounds from month 2.
• Joinable to public CSO PxStat tables and Revenue OMSP at the cohort level.

This is not:

• A national census of Irish used-car transactions.
• A dataset of individual listings, dealers, or sellers.
• A forecast or projection. v0.x is descriptive only (OPERATING-RULES-SIMON.md Rule 6 §7).
• A republication of OMSP, CSO, Carzone, or AutoTrader UK data. Those sources are cited and used for cross-validation only.

---

Headline limitations of v0.1 (read before citing)

1. Listing prices, not transactions. Each row reflects what dealers ask, not what buyers pay.
2. Sample is autoza.ie passenger cars, not Ireland. 51 distinct dealer accounts contributed listings to the v0.1 publishable scope, totalling 1,771 passenger-car listings. Geographic coverage is uneven. Commercial vehicles are excluded from this index.
3. Make-mix skew toward Volkswagen Group. v0.1's published cohorts reflect the autoza.ie dealer mix: a majority of the granular by_segment_year cohorts and a large minority of the broader by_segment cohorts are Skoda or Volkswagen models. As the dealer base broadens, segment coverage will rebalance and additional makes will pass the n ≥ 10 floor. Do not read v0.1 as a cross-section of the Irish market; read it as a snapshot of the autoza.ie dealer network in May 2026.
4. Single-month snapshot. No back-fill. Sold-listing data was insufficient at v0.1 (only 2 listings carry soldAt).
5. Cross-validation gate is advisory at v0.1, not enforcing. Revenue OMSP was tested as the candidate market-price anchor and rejected — Revenue's depreciation curves are a tax-basis figure that lags the actual Irish used-car market by a wide enough margin to make a > 10% deviation gate either fire on every cell or never bind. Decision recorded in METHODOLOGY.md §2.2 and accuracy_report_2026-05.json. Enforcing gate (different anchor mix) begins v0.2.
6. Confidence tier mix. Each row carries its own confidence_tier: low (10 ≤ n < 20, NULL quartiles), medium (20 ≤ n < 30, full quartiles), or high (n ≥ 30). Read the confidence_tier column on every row before citing; do not use a low-tier row as a precision quote.
7. County = ALL. County splits arrive in v0.2 when cell density supports them (currently only 5 cohorts survive n ≥ 10 after county split).
8. No predictions. v0.x is descriptive only.

---

How to cite

Autoza Ireland (2026).
Autoza Irish Used Car Price Index — Pilot.
Hugging Face Datasets. [DOI to be assigned by Zenodo on first publication]
URL: https://huggingface.co/datasets/Autoza/irish-used-car-price-index
Accessed: YYYY-MM-DD.

A persistent DOI is minted via Zenodo on first publication and updated each release. See docs/zenodo-publish.md in the source repository for the workflow.

---

License

CC BY 4.0. You may share and adapt for any purpose, including commercial, with attribution. Full license text and required attribution strings: LICENSE_AND_ATTRIBUTION.md.

Required attribution strings (verbatim)

Contains Irish Public Sector Information from the Office of the Revenue Commissioners, re-used under the Re-Use of Public Sector Information Regulations 2005 (S.I. No. 279/2005, as amended). The OMSP-derived reference values are derivatives and are not Revenue's official OMSP. Revenue does not endorse this dataset.

Contains content from the Central Statistics Office of Ireland, re-used under the Re-Use of Public Sector Information Regulations 2005 (S.I. No. 279/2005, as amended). Source: CSO PxStat, table [code], accessed [date]. The CSO does not endorse this dataset.

---

Errata

errata.md ships with this dataset from day 1. If you find an error in any cell:

• Open an issue on the Hugging Face dataset community tab, or
• Email info@autoza.ie with subject line Errata — Autoza Price Index — YYYY-MM.

We log all errata transparently. Silent corrections are not done. (OPERATING-RULES-SIMON.md Rule 6 §5.)

---

Versioning

Version	Trigger
0.1-pilot	First snapshot. (this release)
0.2	When ≥ 50 cells survive n ≥ 10 after county split. Adds OMSP cross-validation.
0.3	When ≥ 100 cells survive after transmission and mileage_band splits.
1.0	When n ≥ 30 across the top 20 (make, model) pairs. "Pilot" tag dropped.

The release_version column is populated on every row. Consumers can join across releases.

---

Maintainer & contact

• Maintainer: Autoza Ireland (legal entity disclosed in LICENSE_AND_ATTRIBUTION.md §2).
• Email: info@autoza.ie
• Website: autoza.ie
• Hugging Face organisation: Autoza

---

Files in this release

File	Purpose
README.md	This file (Hugging Face dataset card).
SCHEMA.md	Normative schema, suppression rules, PII exclusion list.
METHODOLOGY.md	Sources, suppression, outlier filter, cross-validation, verification, statistical caveats.
LICENSE_AND_ATTRIBUTION.md	CC BY 4.0 + Revenue + CSO attribution + database-right declaration.
FAQ.md	Frequently-asked questions, sourced from methodology + Irish public bodies.
GLOSSARY.md	Definitions of statistical, dataset, and Irish-public-sector terms used here.
FOR-RESEARCHERS.md	Quick-start sample queries, citation forms, validation hooks.
TOP-INSIGHTS-2026-05.md	Headline cohorts for this month, all numbers from the parquet output.
MAKE-COVERAGE-2026-05.md	Top 20 makes by listing count under v0.1 publishable scope.
LATEST-ANALYSIS.md	Pointers to recent autoza.ie blog analysis. Refreshed weekly (Mondays 09:00 UTC) by an automated cron.
CITATION.cff	Machine-readable Citation File Format 1.2.0 — for tools that auto-extract dataset citations.
croissant.json	mlcommons/croissant metadata for AI ingestion.
errata.md	Errata log (empty at v0.1).
CHANGELOG.md	Per-release notes.
data/by_segment_year/2026-05.parquet	Granular config: (make, model, 3-year band, fuel).
data/by_segment_year/latest.parquet	Pointer to most recent month for the granular config.
data/by_segment/2026-05.parquet	Broad config: (make, model, fuel) with year mixed.
data/by_segment/latest.parquet	Pointer to most recent month for the broad config.
accuracy_report_2026-05.json	Per-release accuracy artefact (covers both configs).