KDB dataset 2022-2026

Download

kdb-decisions-2022-2026.csv (3.8 MB, 2,511 rows, UTF-8, 41 columns). The data file reflects the state as of 6 November 2026. It contains every KDB case suitable for analysis in tabular form, with the meta fields (case number, dates), the initiator and complainant classification, the legal-basis citations, the fine fields, the project value and the sector tag. The download is public and may be used freely with attribution.

What the dataset contains

The dataset covers 2,511 KDB cases, for proceedings closed between 1 January 2022 and 31 August 2026. Each row is one decision. Field names are in English, the values in Hungarian. The main field groups:

• Identifiers and dates: case_id, case_number, decision_date, submission_date, days_to_decision, year, quarter, semester, semester_key.
• Initiator: initiator_type (on complaint, KH President for contract review, KH President on other grounds, other ex officio), initiator_body_name, is_ex_officio.
• Complainant: complainant_raw, complainant_type (private party, public institution, foreign, other), contracting_authority_raw.
• Outcome: outcome_bucket (well-founded, partially well-founded, unfounded, dismissed, terminated), outcome_bucket_reason_from_text, dismissal_reason.
• Legal basis: kbt_sections_cited, other_legal_refs.
• Fine: fine_huf_total, fine_bucket, has_fine, fine_huf_per_party, multiparty_fine_count, is_multiparty_fine, is_zero_fine_finding.
• Project value and sector: procurement_value_huf, procurement_value_currency, value_bucket, sector_hint, sector_canonical.
• Other: ekr_id, consolidated_into, was_challenged, extraction_status, extraction_confidence_self_reported, notes_flags, text_sidecar_filename, doc_type_self_reported, extraction_timestamp.

The full field descriptions and enum values are in the extraction schema document, which is available on request.

How the dataset was built

The dataset is built from the KDB decisions published on the Public Procurement Authority's portal, by automated case-level extraction followed by a four-layer ex post check: external comparison with the KH summary reports, independent re-extraction of 99 cases, a review of the extraction logic and a coverage check.

599 cases are excluded from the analysis: 590 consolidated child cases (parallel case numbers joined into a single decision), 5 source conflicts and 4 records without a source. That leaves 2,511 rows suitable for analysis out of 3,110 cases.

Known quality limits

The dataset is built on the text of the published decisions and orders. What is not in the text of a decision does not appear in the dataset either. The limits below apply to the dataset as a whole, so they hold for every analysis built on it.

About 12 per cent of the recorded Kbt. citations do not appear in the text of the decision

The automated extraction reads the legal-basis citations from the text of the decision. According to an independent re-extraction of 49 cases, about 88 per cent of the recorded citations can be found in the text of the decision, and the rest refer to sections of the Kbt. (the Public Procurement Act) that the decision does not cite. The citation list is also not always complete. This does not affect the aggregate fine data or the sector breakdown, but an analysis of a specific legal basis stays within the accuracy limit of the extraction.

6.6 per cent of cases have no classified main outcome

Some KDB decisions record a mixed outcome (partially well-founded on one point, unfounded on another). In 6.6 per cent of cases (165 cases) the automated classification could not determine the main outcome unambiguously, and the outcome_bucket field is empty for these. 98 of these rows also carry the compound_or_ambiguous_outcome tag in the notes_flags field. The independent re-extraction of 99 cases classified the main outcome identically in 77 cases.

74 cases from the first half of 2024 are missing in the KH President, other grounds category

For the first half of 2024, the KH summary report gives 102 cases brought by the KH President on other grounds, while the dataset recorded 28, a coverage gap of 74 cases. Most of the gap is probably explained by the 590 consolidated child cases, which are joined into a single decision but appear on the portal under separate case numbers. In the same category the KH reports 59 cases for the first half of 2025 against 29 in the dataset, and 20 against 21 for the first half of 2026, so the consolidation effect varies from year to year.

The figures published in the Public Procurement Authority's summary report differ somewhat from those in the dataset, because there every initiation submitted counts as a separate case, while here they are summed under the main case after joinder.

69 per cent of the fine fields match the independent re-extraction exactly

Of the fine data, 69 per cent match the independent re-extraction exactly. The large individual fines and the entity-level totals were checked by manual reconciliation, and the band and aggregate shares sit in a stable range over the large sample.

Fine amounts were extracted by text analysis of the operative part, and the entity-level totals are manually reconciled sums of the per-party entries.

Sector classification accuracy is moderate

The sector classification assigns the cases to 13 canonical categories by automated classification of the "subject of procurement" text field. A methodological audit of the classification (checked on 99 cases) found a full match with the classification of the independent re-extraction in 57 per cent of cases. Across the full dataset this gives a reliable aggregation (over a large sample the divergences balance out), but certainty at the level of individual cases is moderate. Sector shares can be read with confidence, the sector tag of a single case not necessarily.

182 cases, 7.2 per cent of the dataset, received no canonical sector classification. In these, the "subject of procurement" field of the decision was too general (for example "supply of goods") or spanned several sectors. These rows have the value unknown in the sector_canonical field.

Complainant and contracting authority names come from the decision text, unreconciled

The complainant_raw and contracting_authority_raw fields record the name given in the decision header, without linking it to an institutional identifier. In some of the cases with a combined "kérelmező/ajánlatkérő" (complainant/contracting authority) header field, the decision text named more than one complainant, but only the first was recorded.

Licence and source

The dataset is built on KDB decisions published on the Public Procurement Authority's portal, which are themselves public data. The compiled CSV can be downloaded freely and used for analysis and republication, with attribution: Sugallat.hu, 2026, KDB dataset 2022-2026. The original documents can be found on the Public Procurement Authority's portal by decision number.

Analyses built on the dataset

The articles in Sugallat's KDB series each unpack one aspect of the dataset:

• The full picture: a summary of 2,511 cases, from the initiators through sectors and fines to judicial review, the overview article of the series.
• The KH President's win rate: 98.46 per cent on the regulatory side.
• The contracting authority as complainant: well-founded complaints in 82 of 84 decisions on the merits.

Contact

For questions about using the dataset, an extended description of the schema or details of the known limits, the author can be reached through Sugallat.hu's contact page. The next release of the dataset is planned with targeted fixes for the known limits above.