bettergovph/philgeps-v2

PhilGEPS procurement notices and awards, 2000–2025 (v2)

This dataset holds every row of the 70 notice and award exports published by the Philippine Government Electronic Procurement System (PhilGEPS), covering 2000-11--2000-12 to 2025-07--2025-09. That is 18,515,260 rows in one schema, one per line item of a bid notice, with its award where one was published. Every row says which export and which row it came from.

Independent and unofficial. This is csiiiv's normalization of the published PhilGEPS exports, published by BetterGov.ph. It is not an official record.

• Rows are kept as published. Republished quarters overlap, duplicates are kept on purpose, and nothing is corrected.
• The exports name no currency for their amounts. PHP is assumed but not confirmed.
• A notice, an award, an amount or a flag does not by itself establish irregularity or wrongdoing by any agency, supplier or official.

Which file to use

File	Config	Contents
philgeps-2000-2025.parquet	default	Start here. The normalized layer: 58 columns, every value a string exactly as published, with the source of each row
philgeps-2000-2025-typed.parquet	typed	The same rows and order, typed for analysis: dates are timestamps and amounts are DECIMAL(38,4). Adds three name-comparison keys and per-row cleaning flags (64 columns)
sources.parquet	sources	The 70 exports: file name, period, format, size, SHA-256, layout, row count, and Drive link
source-catalog.json	—	csiiiv's full source catalog, including each export's column headers
manifest.json	—	Every file, with its bytes, SHA-256, row count, and the SHA-256 of csiiiv's original Parquet files

The default layer keeps blanks and literals such as NULL and N/A exactly as published. The typed layer turns them into real nulls, so count values there.

Every row carries _source_file_id, _source_sheet and _source_row_number. Join _source_file_id to sources.source_file_id to find the export and the row a value came from.

Left out: names of private individuals

The exports, and csiiiv's files, include two columns that name private individuals. We leave both out:

• created_by is the government staff member who posted the notice: 8,392,977 rows, 62,829 distinct names.
• awardee_contact_person is the supplier's contact person: 2,716,421 rows, 89,025 distinct names.

Both appear only in the 43-column exports. Every other column, every row and the row order are unchanged. The release check compares an order-sensitive fingerprint of every kept value against csiiiv's files.

Not here either: the raw exports, and the OCDS-shaped tables csiiiv built from the typed layer.

Before you add anything up

• One row is a notice line item, not an award.
  • 18,515,260 rows carry 10,861,545 distinct bid references.
  • 6,970,129 rows carry an award reference, for 4,903,849 distinct awards.
  • An award with several line items repeats its contract_amount on every one of them. Of the 456,917 awards spread over more than one row, 456,655 repeat the same amount. Summing contract_amount across rows overcounts, so count each award once.
• Some quarters were exported more than once. About 2.5 million rows repeat another row's bid reference, line item, award reference, awardee, amount and item name. They are kept, because choosing which copy to keep is a judgment the data can't make for you.
• The layouts differ, so coverage differs:
  • Procuring entity location columns exist only in the 46-column CSV exports.
  • client_agency exists only in the 43-column exports.
  • A column a layout doesn't have is null for its rows. The column table below shows how many rows have a value.
• Dates:
  • Timestamps carry no time zone. Asia/Manila is likely, but the exports don't say.
  • Bid notices run from 2000-12-01 to 2025-09-30.
  • Award dates run from 1920-01-08 to 2034-10-04, so some are clearly wrong as published. The typed layer flags them; it doesn't fix them.
• item_budget uses -1 for "not provided". The typed layer turns it into null and flags the row.
• Names are kept as printed. The *_key columns group names only by case and spacing. No agency or supplier has been merged with another, and no official identifiers are assigned.
• 74,541 rows name a joint venture that isn't split into its partners.
• 10 cells contain encoding damage (U+FFFD, for example LOS BAÑOS). They are kept as is and listed in _replacement_character_columns.

Export layouts

_source_schema_id	Exports	Rows
modern_43	40	8,544,543
location_46	18	8,050,724
legacy_40_uom	11	1,595,484
legacy_40_unit_of_measurement	1	324,509

The 2000–2010 workbooks have their header on row 4. Four 2021–2022 CSVs have no header row and are read by verified column position.

Cleaning flags (typed layer)

The flags are findings for review, not corrections. 806,356 rows have at least one.

Flag in _cleaning_issues	Rows
item_budget_not_provided_sentinel	589,612
award_before_bid_published	179,717
contract_end_before_notice_to_proceed	30,968
contract_amount_exceeds_approved_budget	9,561
item_budget_exceeds_approved_budget	5,200
invalid_contract_end_date	11
replacement_character	8
missing_bid_reference	1

Columns

The "PhilGEPS" notes list the export headers each column was read from.

Column	In default	Type in typed	Rows with a value (typed)	Notes
_source_file_id	✓	string	18,515,260	The export the row came from; joins sources.source_file_id (SHA-256 of the file).
_source_sheet	✓	string	10,464,536	Worksheet in that export (null for CSV).
_source_row_number	✓	int64	18,515,260	Row in that export, as a spreadsheet shows it.
_source_schema_id	✓	string	18,515,260	Which of the four export layouts the row was read with.
procuring_entity_name	✓	string	18,515,259	PhilGEPS: “Organization Name”, “Procuring Entity”, “Procuring Entity (PE)”
bid_reference_no	✓	string	18,515,259	PhilGEPS: “Bid Reference No.”, “Reference ID”
solicitation_no	✓	string	10,450,720	PhilGEPS: “Solicitation No.”
notice_title	✓	string	18,513,764	PhilGEPS: “Notice Title”
bid_published_date	✓	timestamp[us]	18,352,964	PhilGEPS: “Publish Date”, “Published Date”
classification	✓	string	18,515,259	PhilGEPS: “Classification”
notice_type	✓	string	10,464,535	PhilGEPS: “Notice Type”
business_category	✓	string	18,515,259	PhilGEPS: “Business Category”
funding_source	✓	string	18,515,186	PhilGEPS: “Funding Source”
funding_instrument	✓	string	18,515,130	PhilGEPS: “Funding Instrument”
procurement_mode	✓	string	18,515,259	PhilGEPS: “Procurement Mode”
trade_agreement	✓	string	18,515,259	PhilGEPS: “Trade Agreement”
approved_budget	✓	decimal128(38, 4)	18,515,259	PhilGEPS: “Approved Budget of the Contract”
area_of_delivery	✓	string	17,138,754	PhilGEPS: “Area of Delivery”
contract_duration	✓	int64	18,170,001	PhilGEPS: “Contract Duration”
calendar_type	✓	string	17,889,397	PhilGEPS: “Calendar Type”
line_item_no	✓	int64	11,818,682	PhilGEPS: “Line Item No”
item_name	✓	string	11,818,143	PhilGEPS: “Item Name”
item_description	✓	string	11,810,237	PhilGEPS: “Item Desc”, “Item Description”
quantity	✓	decimal128(38, 6)	11,818,682	PhilGEPS: “Quantity”
unit_of_measurement	✓	string	11,818,624	PhilGEPS: “UOM”, “Unit of Measurement”
item_budget	✓	decimal128(38, 4)	11,229,070	PhilGEPS: “Item Budget”
prebid_date	✓	timestamp[us]	2,581,455	PhilGEPS: “PreBid Date”
closing_date	✓	timestamp[us]	18,174,604	PhilGEPS: “Closing Date”
bid_notice_status	✓	string	18,515,259	PhilGEPS: “Bid Notice Status”, “Notice Status”
award_reference_no	✓	string	6,970,129	PhilGEPS: “Award No.”, “Award Reference No.”
award_title	✓	string	6,970,100	PhilGEPS: “Award Title”
award_type	✓	string	3,061,264	PhilGEPS: “Award Type”
unspsc_code	✓	string	6,301,316	PhilGEPS: “UNSPSC Code”
unspsc_description	✓	string	6,301,316	PhilGEPS: “UNSPSC Description”
awardee_organization_name	✓	string	6,919,984	PhilGEPS: “Awardee Corporate Title”, “Awardee Organization Name”
contract_amount	✓	decimal128(38, 4)	6,962,376	PhilGEPS: “Contract Amount”
contract_no	✓	string	2,061,617	PhilGEPS: “Contract No”
award_published_date	✓	timestamp[us]	6,970,125	PhilGEPS: “Publish Date(Award)”, “Published Date(Award)”
award_date	✓	timestamp[us]	6,970,070	PhilGEPS: “Award Date”
notice_to_proceed_date	✓	timestamp[us]	5,043,024	PhilGEPS: “Notice to Proceed Date”
contract_effectivity_date	✓	timestamp[us]	5,043,024	PhilGEPS: “Contract Efectivity Date”, “Contract Effectivity Date”
contract_end_date	✓	timestamp[us]	5,068,751	PhilGEPS: “Contract End Date”
reason_for_award	✓	string	3,060,634	PhilGEPS: “Reason for Award”
award_notice_status	✓	string	6,970,128	PhilGEPS: “Award Notice Status”, “Award Status”
client_agency	✓	string	282,983	PhilGEPS: “Client Agency”
procuring_entity_region	✓	string	8,050,724	PhilGEPS: “Region”
procuring_entity_province	✓	string	8,050,724	PhilGEPS: “Province”
procuring_entity_city_municipality	✓	string	8,050,724	PhilGEPS: “City/Municipality”
government_branch	✓	string	8,050,724	PhilGEPS: “Government Branch”
procuring_entity_organization_type	✓	string	8,050,724	PhilGEPS: “PE Organization Type”
procuring_entity_organization_type_grouped	✓	string	8,050,724	PhilGEPS: “PE Organization Type (Grouped)”
awardee_country	✓	string	3,902,161	PhilGEPS: “Country of Awardee”
awardee_region	✓	string	3,896,195	PhilGEPS: “Region of Awardee”
awardee_province	✓	string	3,897,502	PhilGEPS: “Province of Awardee”
awardee_city_municipality	✓	string	3,899,781	PhilGEPS: “City/Municipality of Awardee”
awardee_size	✓	string	3,716,039	PhilGEPS: “Awardee Size”
awardee_joint_venture	✓	string	74,541	PhilGEPS: “Awardee Joint Venture”
_extra_values	✓	list<element: string>	0	Unexpected trailing values on the row (none in this release).
procuring_entity_name_key	—	string	18,515,259	Upper-case comparison key for grouping names; entities are not merged.
client_agency_key	—	string	282,983	Upper-case comparison key for client_agency.
awardee_organization_name_key	—	string	6,919,984	Upper-case comparison key for awardee_organization_name.
_replacement_character_columns	—	list<element: string>	18,515,260	Columns on the row that contain U+FFFD (encoding damage, kept as is).
_cleaning_issues	—	list<element: string>	18,515,260	Findings for review on this row, not corrections: see the flags table.
_cleaning_issue_count	—	int32	18,515,260	Number of findings on the row.

Use it

from datasets import load_dataset

philgeps = load_dataset("bettergovph/philgeps-v2", split="train", streaming=True)

-- DuckDB: awards by year, counting each award once
WITH awards AS (
  SELECT DISTINCT award_reference_no, year(award_date) AS year, contract_amount
  FROM 'hf://datasets/bettergovph/philgeps-v2/philgeps-2000-2025-typed.parquet'
  WHERE award_reference_no IS NOT NULL
)
SELECT year, count(*) AS awards, sum(contract_amount) AS amount
FROM awards GROUP BY year ORDER BY year;

Sources and method

• Source: the PhilGEPS exports, quarterly and annual XLSX and CSV files, collected in the PHILGEPS-V2 Drive folder.
• Pipeline: csiiiv/philgeps_data_v2, at the commit in manifest.json.
  • Stage 0 downloads the exports and registers each one by its SHA-256.
  • Stage 1 maps every layout onto one schema by column name. Nothing is dropped, deduplicated or coerced, so the result is the normalized layer.
  • The typed layer then applies config/cleaning_rules.json: null literals, date and decimal typing, Unicode NFC and trimming, and the comparison keys.
• This release:
  • leaves out the two name columns;
  • renames the files;
  • rewrites them with Zstandard compression in 100,000-row groups;
  • checks the row count, column types and an order-sensitive fingerprint of every kept value against csiiiv's files.

How this relates to bettergovph/philgeps-data

bettergovph/philgeps-data is the earlier release: 5,481,161 awards in 12 columns. This one has the full line-item exports, with every published column except the two above, and the source of every row.

License

CC0 1.0. PhilGEPS notices are works of the Government of the Philippines, and no copyright subsists in them (Republic Act 8293, section 176). Credit is welcome: "PhilGEPS exports normalized by csiiiv (philgeps_data_v2), published by BetterGov.ph".

Report a problem

Open a discussion on this dataset's Community tab. Include the _source_file_id and _source_row_number.

Citation

@misc{bettergov_philgeps_v2,
  title        = {PhilGEPS procurement notices and awards, 2000--2025 (v2)},
  author       = {{BetterGov.ph} and csiiiv},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/datasets/bettergovph/philgeps-v2}}
}