ogasurfproject/jccdb-observations

JCCDB v5.0 observation layer (Japan)

Construction cost observations from Japanese public documents: 330,362 rows (302,179 with a value) from 76 public sources (MLIT design labor rates, regional design material unit prices, public-works unit prices, construction cost statistics and deflators). One row is one observation of an item, in a region, at a point in time, with the value or the reason no public value exists, the unit, the source page, the evidence URL and the licence.

This is the observation layer of the Japan Construction Cost Database (JCCDB) v5.0. The same rows are archived as CSV on Zenodo.

• DOI of JCCDB v5.0: https://doi.org/10.5281/zenodo.22980284
• All versions of JCCDB (concept DOI): https://doi.org/10.5281/zenodo.22127751
• United States counterpart with the same columns, USCCDB (2,849,829 rows): https://huggingface.co/datasets/ogasurfproject/usccdb, https://doi.org/10.5281/zenodo.22979157
• Source repository: https://github.com/ogasurfproject-jpg/horizon-shield/tree/main/data/jccdb-obs-v2
• Live access for AI agents: the HORIZON SHIELD MCP server, https://mcp.horizonshield.dev/mcp

Load

from datasets import load_dataset
ds = load_dataset("ogasurfproject/jccdb-observations", split="train")

Every column is a string, so codes keep their leading zeros (for example 01). Convert price to a number where you need one; an empty string means the source gives no public value, and price_status says why.

Files

• data/*.parquet: 63 files, one per source family, converted from exactly the CSV files inside jccdb-v5-observations-jp.zip on Zenodo (same names, same rows).
• docs/jccdb-v5-observations-manifest.csv: rows and SHA-256 of every one of those CSV files.
• docs/jccdb-v5-sources-jp.zip: the source ledgers (URL, retrieval date, SHA-256 of the original, licence and its text quoted verbatim, how it was read).
• docs/jccdb-v5-observations-schema.md (English) and docs/jccdb-v5-observations-schema.ja.md (Japanese): columns and price statuses.

Columns: obs_id, country, layer, category, item_name, spec, unit, geo_level, geo_code, geo_name, area_label, area_code, area_members, price, currency, price_basis, price_status, ref_value, ref_note, period, effective_from, source_id, source_page, evidence_url, license, jccdb_v4_item_id, note

Counts in the observation layer

layer	rows
cost_sqft	36,819
equipment	400
index	18,125
labor	32,924
material	150,149
spending	804
work	91,141

price_status	rows
not_set	27,133
publication_based_not_public	1,050
published_cc_by	36,819
published_pdl	265,360

licence	rows
GOV-STD-2.0	36,819
PDL1.0	293,543

What is not included, and why

• Rows from sources whose own terms forbid reproduction or conversion to electronic form are not included, even as item lists without values: the design material unit price tables of the Chubu, Tohoku and Shikoku Regional Development Bureaus, and the Nara Prefecture table (1,584 rows excluded here). The public GitHub repository holds the three bureau tables only as counts, reasons and source URLs, and the Nara status rows (which cells are published, without values); none of them is redistributed here.
• No values from commercial price publications. Cells where a public body uses commercial price publications are recorded as publication_based_not_public, without a value.
• Rows held back for a suspected column shift in the MLIT construction cost deflator are not included.

What these values are not

Public-works design unit prices, labor rates, statistics and indexes are reference data. They are not renovation quotes, and they are not a verdict on any quote.

Licence

CC BY 4.0 for the compilation. Every row also carries the licence of its source in the license column, and the ledger of each source quotes its terms. When you use a value, also show the attribution of its source.

How to cite

Oga, T. (2026). Japan Construction Cost Database (JCCDB) v5.0 [Data set]. Zenodo. https://doi.org/10.5281/zenodo.22980284

@dataset{oga_2026_jccdb_v5,
  author    = {Oga, Toshikatsu},
  title     = {Japan Construction Cost Database (JCCDB) v5.0},
  year      = {2026},
  publisher = {Zenodo},
  version   = {5.0},
  doi       = {10.5281/zenodo.22980284},
  url       = {https://doi.org/10.5281/zenodo.22980284}
}

日本語の要約

日本の公的な資料から取った建設費の観測 330,362 行(値のある行 302,179、出典 76)。JCCDB v5.0 の観測層で、Zenodo に CSV で納めた物と同じ行を、頭の 0 が落ちないように全部の列を文字列にした Parquet で置いている。1 行 1 観測で、どの地域・どの時点・いくら(か、公的な値が無い理由)・単位・原本の頁・証拠の URL・利用条件を持つ。複製や転載を禁じている出典の行と、市販の物価資料の値は入れていない。公共工事の単価・統計・指数は参考資料で、リフォームの見積もり額でも、見積もりの判定でもない。