--- license: apache-2.0 language: - ar - bg - cs - de - en - es - fr - hi - hr - id - it - ja - ko - nl - pl - pt - ro - ru - sv - th - tr - vi - zh tags: - construction - cost-estimation - quantity-surveying - BIM - construction-rates - bill-of-quantities - tabular - parquet - multilingual size_categories: - 1M πŸ’‘ **Looking for a complete estimation platform?** CWICR powers the cost database inside [**OpenConstructionERP**](https://github.com/datadrivenconstruction/OpenConstructionERP) β€” an open-source modular ERP for construction (BOQ editor, BIM/CAD import, AI takeoff, GAEB XML, AGPL-3.0). The ERP consumes the parquets here and the Qdrant snapshots in the companion repo to provide localised cost estimation in 30 markets out of the box. ## How CWICR is used inside OpenConstructionERP ![BOQ Editor, Resource Catalog, Cost Database, Smart Item Search](./assets/feature-databases.jpg) The parquets in this repository are loaded into the ERP as four cooperating surfaces: - **BOQ Editor** β€” pick rates from the localised database and assemble Bills of Quantities in the project's currency. - **Resource Catalog** β€” drill into the per-rate decomposition (labour roles, machinery classes, material families) preserved verbatim in the parquet. - **Cost Database** β€” browse the full catalogue with national filters; the underlying schema matches the parquet 1-to-1. - **Smart Item Search** β€” backed by the companion Qdrant snapshots (BGE-M3 hybrid dense + sparse) for typo-tolerant multilingual lookup. ## Contents per country Each language directory (`AR/`, `BG/`, `CS/`, `DE/`, `EN/`, `ES/`, `FR/`, `HI/`, `HR/`, `ID/`, `IT/`, `JA/`, `KO/`, `MX/`, `NG/`, `NL/`, `NZ/`, `PL/`, `PT/`, `RO/`, `RU/`, `SV/`, `TH/`, `TR/`, `UK/`, `US/`, `VI/`, `ZA/`, `ZH/`) contains **5 files**, all uploaded for all 30 locales: | File | Description | |---|---| | `__workitems_costs_resources_DDC_CWICR.parquet` | Primary table β€” one row per (rate Γ— resource). ~900K rows per country, 84 columns. | | `DDC_CWICR___Catalog.csv` | Compact catalogue β€” one row per unique `rate_code` with title, unit, collection, hierarchy. ~55K rows per country. | | `DDC_CWICR___Catalog.xlsx` | Same catalogue, Excel-formatted (auto-widths, frozen headers, ~800 KB). | | `README.md` | Per-country quick reference in the source language (locale, collections, schema notes). | | `README_DDC_CWICR_TABULAR__.txt` | Plain-text full documentation in the source language β€” covers schema, classification hierarchy, units, currency, source publication and license notes for that country's rate book. | **File totals on this repository**: 30 parquets + 30 CSV catalogues + 30 XLSX catalogues + 30 per-country README.md + 30 localised TXT readmes = 150 country files (plus the top-level README and assets). Excel exports of the full table (`*_FORMATTED.xlsx`, `*_SIMPLE.xlsx`) and the localised PDF documentation are reproducible from the parquet + TXT readme and are kept in the source GitHub repository as Git LFS objects. ## Schema (parquet β€” 84 columns) The parquet preserves the full rate decomposition. A rate (e.g. *"Excavator 15 mΒ³, soil class 2"*) is unfolded into multiple rows, one per resource it consumes (the excavator itself, fuel, operator labour, transport, etc.). Each row carries both rate-level fields (replicated across the rate's rows) and resource-level fields (unique per row). Column groups: - **Identification**: `rate_code`, `rate_original_name`, `rate_final_name`, `rate_unit`, `is_abstract` (parent aggregator vs. specific variant) - **Hierarchy**: `category_type`, `department_code`, `department_name`, `section_name`, `subsection_name`, `collection_name` - **Physical params**: `mass_name`, `mass_value`, `mass_unit` (preserves original semantics: "масса 1 ΡˆΡ‚.", "Gewicht", etc.) - **Costs**: `rate_total_cost`, `rate_labour_cost`, `rate_materials_cost`, `rate_machinery_cost`, `rate_overhead`, `rate_profit`, `currency` - **Work composition**: `work_composition_text` β€” full natural-language description of the work performed - **Resources** (one row each): `resource_code`, `resource_name`, `resource_type`, `resource_quantity`, `resource_unit`, `resource_unit_cost`, `resource_total_cost` - **Machinery taxonomy**: `machine_class1_name` … `machine_class3_name` β€” classification of machinery resources - **Provenance**: `source_book`, `source_year`, `source_publisher`, `source_url` All text fields are in the **source language of the rate book** (Russian, German, Chinese, Hindi, etc.). Unit symbols and numeric values are kept verbatim from the original publication. ## Quick start (Python) ```python from datasets import load_dataset # Load all rates for one country (parquet β†’ in-memory Arrow table) ds = load_dataset("DataDrivenConstruction/cwicr-construction-rates", split="de") print(ds.column_names) print(ds[0]) # Or load lazily with polars (recommended for >100K rows) import polars as pl df = pl.scan_parquet( "hf://datasets/DataDrivenConstruction/cwicr-construction-rates/DE/DE_BERLIN_workitems_costs_resources_DDC_CWICR.parquet" ).filter(pl.col("collection_name").str.contains("Erdarbeiten")).collect() ``` ## Quick start (CLI) ```bash # Download one country huggingface-cli download DataDrivenConstruction/cwicr-construction-rates \ --include "DE/*" --repo-type dataset --local-dir ./cwicr_de # Convert to CSV for quick inspection python -c "import polars as pl; pl.read_parquet('cwicr_de/DE/DE_BERLIN_workitems_costs_resources_DDC_CWICR.parquet').write_csv('de.csv')" ``` ## Coverage **30 locale-specific rate books** spanning **23 languages**: | Region | Locales | |---|---| | Russian-speaking | RU (St. Petersburg / federal) | | English-speaking | EN (Toronto), UK (London / GBP), US (USA / USD), AU (Sydney), NZ (Auckland), NG (Lagos), ZA (Johannesburg) | | Spanish/Portuguese | ES (Barcelona), MX (Mexico City), PT (SΓ£o Paulo) | | Central/Eastern Europe | DE (Berlin), FR (Paris), IT (Rome), NL (Amsterdam), PL (Warsaw), CS (Prague), HR (Zagreb), BG (Sofia), RO (Bucharest), SV (Stockholm), TR (Istanbul) | | Middle East / South Asia | AR (Dubai), HI (Mumbai) | | East / Southeast Asia | ZH (Shanghai), JA (Tokyo), KO (Seoul), TH (Bangkok), VI (Hanoi), ID (Jakarta) | Each rate book uses its **native classification system** (Russian БНиП codes, German DIN structures, US MasterFormat-style hierarchies, Chinese GB classifiers, etc.). The schema is unified across countries but the *values* in `collection_name` / `department_code` / etc. are in the source language and follow each country's own taxonomy. ## Use cases - **Cost estimation**: BIM quantity takeoff β†’ match elements to rates β†’ compute project cost in local currency - **Cross-country price benchmarking**: compare the unit cost of equivalent works (e.g. concrete pouring) across markets - **Construction-specific LLM fine-tuning**: rich multilingual corpus of technical construction language with structured grounding - **Resource normalization research**: align machinery / material / labour ontologies across national catalogues - **Quantity surveying education**: realistic worked examples in 23 languages ## Related artifacts - **Vector DB snapshots** (Qdrant, hybrid BGE-M3 dense + sparse + resources, ready to restore): β†’ [`DataDrivenConstruction/cwicr-vector-db-bgem3-v3`](https://huggingface.co/datasets/DataDrivenConstruction/cwicr-vector-db-bgem3-v3) - **OpenConstructionERP** β€” production ERP that consumes this dataset (BOQ editor, BIM/CAD import, AI takeoff, GAEB): β†’ [`github.com/datadrivenconstruction/OpenConstructionERP`](https://github.com/datadrivenconstruction/OpenConstructionERP) - **Source code / pipelines** (encoders, classifiers, search demos, BIM matchers): β†’ [`github.com/datadrivenconstruction/OpenConstructionEstimate-DDC-CWICR`](https://github.com/datadrivenconstruction/OpenConstructionEstimate-DDC-CWICR) ## Exploring the data Below are a few summary visualisations produced directly from the parquet files (chart code lives in the [source repository](https://github.com/datadrivenconstruction/OpenConstructionEstimate-DDC-CWICR) under `0_Workflow and Pipelines CWICR/python/08-data-analysis/`). ### Cost distribution by department Boxplots of total cost per rate position, grouped by department. Heat-supply and rolling-equipment works dominate at the top end; electrical-installation rates are tightly clustered in the low-cost band. ![Cost distribution by department](./assets/cost_distribution_by_department.png) ### Material vs labour cost Scatter of (material cost, labour cost) per rate, coloured by department. Three regimes are visible: labour-heavy structural works (left vertical band), material-heavy mechanical installations (top-right cluster) and balanced general-purpose rates (diagonal). ![Material vs labour cost](./assets/labor_vs_material_ratio.png) ### Resource utilisation Distribution of how often each resource type (labour roles, machinery classes, material families) is consumed across all rates. ![Resource utilisation](./assets/resource_utilization.png) ### Beyond the database β€” AI takeoff in OpenConstructionERP ![AI takeoff from PDF drawings](./assets/feature-takeoff-ai.jpg) CWICR rates become live cost data once they are matched against quantities. In OpenConstructionERP that matching is automated: drawings (PDF / DWG / RVT / IFC) are converted to a canonical element list, CV/OCR detects symbols and dimensions, and the resulting quantities are paired with the closest CWICR rates via the companion Qdrant snapshots. ## Data sources CWICR aggregates publicly available national rate catalogues: - **RU**: Federal Unified Rates (Π€Π•Π  / Π’Π•Π ), Ministry of Construction RF, regional indices. - **EU**: National building cost compendia for each member state. - **US/CA**: RSMeans-style cost data, MasterFormat-aligned. - **CN/JP/KR/TH/VN/ID**: Government-published quota / norm catalogues. - **Other**: Country-specific official sources cited in each rate's `source_book` field. The CWICR pipeline normalises these into a single schema; the **rate codes, prices, and work compositions are preserved verbatim** from the originals. License applicability of underlying codes follows each catalogue's terms; this redistribution is for research and reference. Where redistribution is restricted, only derived metrics or anonymised samples are included. ## Reproducibility The CWICR pipeline (parsers, normalisers, classifiers) is open-source. Each parquet here is built by: 1. Source-specific parser β†’ standard `workitems_costs_resources` raw frame 2. `v3_classifier.enrich_v3` β†’ 29 indexed payload fields (collections, departments, physical params) 3. `ost_ifc_mapping.base_code` β†’ cross-language rate-code normalisation (lang/unit suffixes + inline tags) See the source repository for full pipeline scripts. ## Support the project ⭐ If this dataset is useful to you β€” or if you'd like to see more open construction databases like it β€” the most helpful thing you can do is **leave a star on the [OpenConstructionERP repository](https://github.com/datadrivenconstruction/OpenConstructionERP)**. Stars are how we gauge demand and prioritise the next country / language / source to publish. No account needed beyond GitHub, no signup, no money β€” just a one-click thank-you that directly steers what gets built next. ## License Released under **Apache-2.0** for the dataset structure, normalisation pipeline, and derived metadata. Underlying rate codes and prices are sourced from publicly available national catalogues β€” each catalogue's terms of use apply to commercial redistribution of raw numeric values. ## Citation ```bibtex @dataset{cwicr_rates_2026, author = {Boiko, Artem and DataDrivenConstruction}, title = {CWICR β€” Construction Works, Items, Costs \& Resources (30 Countries)}, year = {2026}, publisher = {Hugging Face}, url = {https://huggingface.co/datasets/DataDrivenConstruction/cwicr-construction-rates} } ``` ## Contact - Source code & issue tracker: - Author: Artem Boiko (DataDrivenConstruction) - Email: boikoartem@gmail.com