README.md

metadata

license: other
language:
  - es
pretty_name: OpenPLACSP Historical Spanish Public Procurement
tags:
  - public-procurement
  - government
  - spain
  - time-series
  - public-data
configs:
  - config_name: awards
    data_files:
      - split: train
        path: data/awards/*.parquet
  - config_name: cpv_codes
    data_files:
      - split: train
        path: data/cpv_codes/*.parquet
  - config_name: lots
    data_files:
      - split: train
        path: data/lots/*.parquet
  - config_name: versions
    data_files:
      - split: train
        path: data/versions/*.parquet

OpenPLACSP Historical Procurement

Dataset summary

OpenPLACSP Historical Procurement contains public procurement notices published by contracting authorities hosted on the Spanish Public Sector Procurement Platform. Every published version of a procurement procedure is retained, including deletion markers, rather than reducing each procedure to its latest state.

The snapshot covers updates from 2 January 2012 through 30 December 2024. It excludes minor contracts, linked documents, and notices published only through external aggregation platforms.

Property	Value
Publisher	Directorate-General for State Assets, Spanish Ministry of Finance
Geographic scope	Spain
Source format	ATOM feeds containing CODICE XML
Published format	Zstandard-compressed Parquet
Configurations	versions, lots, cpv_codes, awards
Split	One train split per configuration
Total rows	17,104,600
Compressed size	951.20 MB

Dataset structure

The four configurations form a relational historical dataset. version_id is the primary key of versions and the foreign key used by the other configurations.

Configuration	Rows	Description
versions	2,993,582	One row per published state or deletion marker
lots	2,080,084	Lots attached to each published version
cpv_codes	8,944,813	CPV classifications at procedure or lot level
awards	3,086,121	Procedure results, winning parties, and awarded amounts

Versions

Fields	Description
version_id	Deterministic identifier for the published version
atom_id	Identifier of the source ATOM entry
updated_at, published_at	Source update and publication timestamps in UTC
title, summary, entry_url	Public description and URL of the source entry
is_deleted	Whether the row is an ATOM deletion marker
folder_id, status_code	Procurement file number and source status
project_name	Object or title of the procurement project
contract_type_code, contract_subtype_code	Source contract classification codes
estimated_value, budget_tax_exclusive, budget_total, currency	Published values, budgets, and currency
procedure_code	Source procurement procedure code
contracting_party_name, contracting_party_nif	Published authority name and Spanish tax identifier
contracting_party_platform_id, contracting_party_type_code	PLACSP authority identifier and source type
buyer_profile_url	Public buyer-profile URL
source_archive, source_member	Source ZIP archive and ATOM member

Lots

Fields	Description
version_id	Parent version identifier
lot_id	Lot identifier within the procedure
name	Lot name or object
budget_tax_exclusive, budget_total, currency	Published lot amounts and currency

CPV codes

Fields	Description
version_id	Parent version identifier
lot_id	Related lot, or null for a procedure-level classification
cpv_code, cpv_name	Common Procurement Vocabulary code and source label

Awards

Fields	Description
version_id	Parent version identifier
result_position, winner_position	Stable positions within the source version
result_code, description	Published result code and description
award_date, received_tenders	Award decision date and number of tenders received
lot_id	Related lot when the result is lot-specific
winner_name, winner_nif	Published winner name and Spanish tax identifier
award_tax_exclusive, currency	Published award amount excluding taxes and its currency

Version semantics

Repeated atom_id and folder_id values are expected. Each row in versions represents a state published at a particular time. A current-state view can be constructed by selecting the greatest non-null updated_at for each atom_id and retaining deletion markers when the latest row has is_deleted = true.

All configurations use a single train split. Random row-level splits would place versions of the same procedure in different partitions and create temporal leakage. Downstream prediction tasks should use time-aware splits and keep all versions of a procedure together.

Processing

ATOM entries and CODICE elements are normalized into four relational tables. Source values are preserved without currency conversion, inflation adjustment, imputation, or inferred classifications. Linked procurement documents are not included.

Stable version identifiers are computed from the source ATOM identifier, update timestamp, and deletion state. Parquet files use explicit schemas, Zstandard compression, and shards of at most 250,000 rows.

Quality

All 17,104,600 rows passed the required-field checks. The snapshot contains 79,140 deletion markers, no null values in required identifiers or relationship keys, and 70 Parquet shards with verified SHA-256 checksums.

Machine-readable profiles, schemas, quality results, provenance, and checksums are included under artifacts/.

Known limitations

• The snapshot ends on 30 December 2024 and contains no later notices.
• Coverage is not uniform because contracting authorities joined the platform at different times.
• Fields introduced by later CODICE revisions are absent from older records.
• A procedure may appear many times because every published update is retained.
• Monetary values are not converted or adjusted and should not be aggregated across currencies without conversion.
• Published organization names and tax identifiers are source metadata, not independently verified identity data.
• Linked notices and documents may have changed or become unavailable after the snapshot date.

Source and reuse

The source data was published by the Directorate-General for State Assets through the Spanish Public Sector Procurement Platform and the Spanish Ministry of Finance open-data catalogue. The source describes the records as public information available for reuse. Attribution to the original publisher and the supplied snapshot provenance must be retained in derived releases.

The card uses license: other because the source-specific reuse terms have not been mapped to a standard Hugging Face license identifier. The original publisher remains responsible for the source records; this dataset provides a normalized historical representation.