SECOND MEASUREMENT SM-019

A federal grant sub-award must be reported by the end of the month after it is made. GAO found 41% of non-pandemic sub-awards late through 2023; on every sub-award at or above the reporting threshold made since October 2017 it is 56.6%, and state agencies carry three-quarters of the late reports

Markovian ProtocolMeasured 2026-10-12Status: published and corrected 2026-10-13; questions sent 2026-10-13DOI: 10.5281/zenodo.23330729

When a federal grant is passed down to a sub-recipient, a school district, a county, a university lab, the prime recipient must report the sub-award to the government by the end of the month after it was made; the rule is an award term in federal grants, 2 CFR 170, Appendix A, and covers every sub-award of $30,000 or more, $25,000 before 12 November 2020. The reports are public on USAspending.gov with two dates on each: the day the sub-award was made and the month the report was published. Until March 2025 recipients filed through FSRS, the Federal Funding Accountability and Transparency Act Subaward Reporting System; since then the same reports go through SAM.gov, the government’s award-registration site. GAO read the file in 2023 and found 2.4 million of nearly 6 million records late. We recomputed it on the file as it stands on 12 October 2026, every grant sub-award made since October 2017 at or above the threshold, 2,125,343 of them once duplicate reports are removed and the 936,249 below the threshold, which carry no deadline, are set aside: 1,202,449 were reported after the deadline (56.6%), 929,851 three or more months after. The late ones sum to $809.5 billion, after setting aside 6,061 whose amount is larger than its own prime award, a category GAO itself flagged as likely reporting error (the raw sum, those included, is $939.4 billion). Every fiscal year here reads lower than it will finish, because a sub-award not yet reported is not in the file; FY2024, the most mature year before reporting moved into SAM.gov, stands at 57.9% today, and stood at 52.6% at the age FY2025 is now. Prime recipients registered as state governments carry 924,450 of the 1,202,449 late reports (76.9%), most of them agencies passing federal education and nutrition money to districts and counties; EDUCATION, CALIFORNIA DEPARTMENT OF, the name as filed, reported 74,154 of its 99,692 sub-awards late. GAO’s file, read on 1 August 2023, could not contain any report filed after that day: of the 1,396,774 sub-awards in our set made before August 2023, 257,875 were reported in August 2023 or later, so GAO’s count was a floor; on the reports GAO could see, our test gives 52.0% late.

Corrected 13 October 2026: the first version scored every sub-award in the file, including 936,249 whose amount is below the reporting threshold and so carry no deadline, and said such sub-awards were not in the file. With them set aside: 1,202,449 of 2,125,343 late (56.6%), was 1,730,713 of 3,061,592 (56.5%); state governments 76.9% of late reports, was 81.4%; Pennsylvania’s education department 32,203 of 43,529 late, was 160,772 of 255,665. The late sum went up, to $809.5 billion from $806.9 billion: the late sub-awards set aside total minus $2.6 billion, because 77,652 of the 936,249 carry zero or negative amounts. The first version stays on Zenodo, 10.5281/zenodo.23328860, and in the log at leaf 9981.

Disclosure: Markovian Protocol holds no financial position in any organisation named here, was paid by no one for this work, and showed it to no one before publication. How we work.

What the rule says, what we found

Fiscal year by fiscal year

Federal fiscal years, October to September; FY2026 runs to 30 September 2026 and reads low because sub-awards not yet reported are absent; the 445 sub-awards made 1 to 12 October 2026 are in the rows file and in no fiscal-year row. FY2026’s twelve-month column is blank by construction. Every year is a floor, not only FY2026: at the age FY2025 is now, FY2019 read 41.3% and has since risen to 49.0%; FY2021 read 54.2% and rose to 67.0%; the full year-by-year comparison is censoring_at_13_months in the summary json. “Late, total amount” excludes sub-awards whose amount is larger than their own prime award (see Details); FY2022’s raw sum before that exclusion is $193.3 billion, $69.1 billion of it one category GAO flagged. Every scored sub-award, one row per prime award, sub-award number and action date with the earliest report, the sub-recipient’s name and identifier and the prime’s UEI (Unique Entity Identifier, the number SAM.gov assigns a recipient when it registers): subaward_lag_rows.csv.gz.

By awarding agency

Every agency with 1,000 or more sub-awards since October 2017, most late reports first. The agency is the one that made the prime award; the report is owed by the prime recipient.

By prime recipient

The report is owed by whoever received the federal grant and passed part of it on. For the big formula programs, Title I school funding, special education, school nutrition, child care, that is a state agency passing money to every district or county in the state, hundreds or thousands of sub-awards a year from one office. Recipients are grouped by their UEI, which all 903 with 200 or more sub-awards carry; the name shown is the one they used most, as filed.

The twenty-five recipients with the most late reports; 24 of the 25 are self-tagged “State Government” in the file, and the 25th, a state education department, is not tagged but is one by name. Together they carry 493,104 of the 1,202,449 late reports. Every recipient with 200 or more sub-awards: subaward_lag_summary.json.

The same job done on time, recipients with 500 or more sub-awards and the lowest late share:

How this ends

One act is already available without anyone’s help: a prime recipient files the report in the month after the sub-award, and the next monthly read would see its late share for new sub-awards fall toward zero. The other acts are GAO’s own four recommendations, none confirmed done as of this read. GSA was told to add automated validations for known data-quality limits such as wrong amounts (Recommendation 1) and to give recipients guidance against duplicative reporting (Recommendation 2); it now runs the reporting system inside SAM.gov, and our question to GSA above asks whether either has been built in. Treasury was told to disclose subaward data-quality limitations on the USAspending.gov pages where users see the data (Recommendation 3); this page is one more such disclosure. And OMB was told to clarify what agencies must do about a recipient that reports late (Recommendation 4) — GAO found, and the agency officials it interviewed confirmed, that current OMB guidance does not hold agencies responsible for subaward data quality and is unclear on what, if anything, they owe a non-compliant recipient. We re-pull the newest two fiscal years on the first of each month, re-score every sub-award in them, and print the result on the rechecks page by fiscal year and for the recipients named here.

The evidence

Every quote below was checked against a saved copy. Copies and SHA-256 hashes: SHA256SUMS.

Our questions

For the three prime recipients with the most late reports, for the Department of Education, the awarding agency behind all three, for GSA, which runs the reporting system, and for the GAO team behind the 2023 report. Answers will be printed here as written.

1. To the recipients: the file shows your office reporting most of its sub-awards after the end of the month following the sub-award. Is the sub-award action date in the file the date the sub-award was made, and what sets the interval before the report?
2. To the Department of Education: GAO found OMB’s guidance unclear on what, if anything, an agency must do when a recipient reports late, and recommended OMB clarify it. Has the department raised late sub-award reporting with a recipient since 2023 anyway, and does it track the share?
3. To GSA: GAO recommended automated validations in 2023. Since reporting moved into SAM.gov, does the system tell a filer at submission that a sub-award is past its reporting month, and does GSA count how many are?
4. To GAO: figure 5 compares the sub-award action date with “the subaward report date (the date the award was reported in FSRS)”. Was that the report’s publication month in the public file, or a submission timestamp from the system, were duplicate reports of one sub-award counted once or each, and what pandemic DEFC codes did figure 5 use?

Their reply, scored

What we told whom, and when.

What we can’t be sure of

• The file is the recipients’ own reports. A sub-award never reported is not in it, so every count here is a floor on what is owed and late; the newest months read best for that reason.
• Every fiscal year here can still rise as late reports arrive, as the fiscal-year table above shows; at the age FY2025 is now, FY2019 read 41.3% and has since risen to 49.0%, FY2021 read 54.2% and rose to 67.0%. The full year-by-year comparison is censoring_at_13_months in the summary json.
• The report month is the file’s publication month and the last-modified date, mapped by the dictionary to the system’s date-submitted field, never falls outside it on any of the 2,125,343 records (clock_check in the summary). GAO used a submission date from inside the system; if that date can differ from the published month for reports created in one month and published in the next, GAO’s clock and ours differ by up to a month on those.
• One sub-award is one prime award, sub-award number and action date. A recipient that reports the same sub-award under two numbers is counted twice; one that corrects a date creates what we read as a new sub-award. GAO’s duplicate estimate was a quarter of records; ours removed 1,117,877 later reports of the same sub-award. Scored GAO’s way instead, every report counted, the late share only climbs, so this is not where our figure diverges from GAO’s. 61,792 late sub-awards share a recipient, sub-award number and action month with a report filed in an earlier month; 59,013 of them sit under a different prime award, which owes its own report. 5,360 match an earlier report under another prime award on sub-recipient, amount and day, the likely re-reports; counting them as duplicates would move the headline by 0.1 points.
• Dollar totals exclude 6,061 late sub-awards whose amount is larger than their own prime award, $129.8 billion set aside from the $939.4 billion raw sum; GAO found 25,260 such records across the whole file and quoted OMB calling the pattern likely reporting error, but GSA told GAO a few are legitimate, so this is a conservative filter, not a certainty every excluded record is wrong.
• Recipients also report sub-awards below the $30,000 threshold ($25,000 before 12 November 2020); the 936,249 whose earliest reported amount is under it are set aside, and a recipient that judged a sub-award below it and never reported it is not in the file at all. A recipient whose own gross income was under $300,000 in the prior tax year is exempt from reporting any of its sub-awards at all, a separate, recipient-level exemption.
• The action date is typed by the filer. Reports whose month is earlier than the sub-award’s month, 8,338, are dropped; a wrong action date inside the plausible range would move a record between on time and late and we cannot see it.
• Pandemic funding is read from the prime award’s disaster emergency fund codes (L, M, N, O, P, U, V); a sub-award under a prime that carried both pandemic and regular money counts as pandemic-funded.
• “State governments carry three-quarters of the late reports” is read from a self-reported business-type field, and the field has at least one known miss among the top 25 recipients: California’s education department is a state agency by name but not self-tagged “State Government”. Fixing that one case alone would push the share higher. 76.9% is a floor, not the exact number.
• One snapshot, 12 October 2026, from one source, USAspending’s bulk download. The same request on another day returns a different file; the hashes of ours are in the exhibits.

Run it yourself

The script uses only Python’s standard library and reads the zips without unpacking them. USAspending builds each file on request, one fiscal year at a time.

curl -s -X POST https://api.usaspending.gov/api/v2/bulk_download/awards/ -H "Content-Type: application/json" \
  -d '{"filters":{"sub_award_types":["grant"],"date_type":"action_date","date_range":{"start_date":"2024-10-01","end_date":"2025-09-30"},"agencies":[{"type":"awarding","tier":"toptier","name":"All"}]},"file_format":"csv"}'
# repeat per fiscal year; poll the status_url until "finished", fetch file_url
curl -O https://markovianprotocol.com/measurements/sm-019/subaward_lag.py
python3 -I subaward_lag.py out/ All_Subawards_*.zip       # prints the summary; writes out/subaward_lag_summary.json and out/subaward_lag_rows.csv

Ran it? Send the number you got and the date to hello@markovianprotocol.com, subject “Rerun SM-019”. We print it on the track record with your name, or just the date if you prefer, and the first outside rerun of each paper gets the same seal a recipient gets for answering or fixing.

What you can do with this

• If you run grants compliance for a state agency or a university with 200 or more sub-awards since 2018, your office’s row is in subaward_lag_summary.json under your UEI; every scored sub-award is in the rows file under your UEI whatever the count (140 rows carry no prime UEI). Your table row gives the median months your office’s late reports run.
• If you are a sub-recipient, a district or a county, the rows file carries your name, your UEI and the sub-award number, so you can see when the money you received was reported.
• If you write about federal spending data, every number above comes from one public download and one script; rerun it on this month’s file before quoting it.

References

1. 2 CFR Part 170, Appendix A, Award Term: Reporting Subawards and Executive Compensation, 89 FR 30111 (22 April 2024) as amended at 89 FR 79732 (1 October 2024). law.cornell.edu/cfr/text/2/appendix-A_to_part_170. Accessed 2026-10-12; eCFR’s own site answered our reads with a block page.
2. U.S. Government Accountability Office. Federal Spending Transparency: Opportunities Exist to Improve COVID-19 and Other Grant Subaward Data on USAspending.gov, GAO-24-106237, 16 November 2023. gao.gov/products/gao-24-106237. Saved copy gao-24-106237.pdf.
3. USAspending.gov. Data Dictionary, sub-award file elements. usaspending.gov/data-dictionary; API copy api.usaspending.gov/api/v2/references/data_dictionary/. Retrieved 2026-10-12.
4. USAspending.gov. Bulk download: awards, api.usaspending.gov/api/v2/bulk_download/awards/, grant sub-awards by action date, one request per fiscal year FY2018 to FY2026 plus 1 to 12 October 2026, built and retrieved 2026-10-12; file names and SHA-256 hashes in dataset_provenance_2026-10-12.txt.
5. Federal Funding Accountability and Transparency Act of 2006, Pub. L. 109-282, as amended, notes to 31 U.S.C. § 6101. law.cornell.edu/uscode/text/31/6101. Section 2(c)(4), “shall be updated not later than 30 days after the award of any Federal award requiring a posting”, is the separate rule for posting prime awards, which the public file carries no first-reported date for and which this paper does not measure. Saved copy usc-31-6101-notes.html.
6. The move of sub-award reporting from FSRS.gov into SAM.gov: Forvis Mazars, FFATA Reporting Transitions From FSRS to SAM.gov, 7 March 2025 (“[GSA] retired … FSRS.gov, on March 6, 2025”), forvismazars.us; ExecutiveGov, GSA Transfers Subaward Reporting Functions to SAM.gov, 13 March 2025, executivegov.com. Saved copies forvismazars_ffata_fsrs_to_samgov.html, executivegov_gsa_subaward_reporting_samgov.html.
7. USAspending.gov. About the Data Download, “Start Date for Disaster Emergency Fund Code (DEFC) Tracking”: the pandemic codes L (3/6/2020) through V (3/11/2021); Z and “1” (11/15/2021) are the Infrastructure Investment and Jobs Act, tracked on the same system. usaspending.gov/data/about-the-data-download.pdf. Saved copy usaspending_about-the-data-download.pdf.
8. Office of Management and Budget. Guidance for Grants and Agreements, 85 FR 49506, published 13 August 2020, effective 12 November 2020: raised the subaward reporting threshold from $25,000 to $30,000 and revised Appendix A to Part 170 accordingly. federalregister.gov/documents/2020/08/13/2020-17468. Saved copy fedreg-2020-17468.html.

Cite as

@misc{markovian-sm019,
  author = {{Markovian Protocol}},
  title  = {Federal grant sub-awards reported after the end-of-next-month deadline in 2 CFR 170,
            every reported sub-award at or above the reporting threshold made October 2017 to October 2026, recomputing GAO-24-106237},
  number = {SM-019},
  year   = {2026},
  month  = oct,
  url    = {https://markovianprotocol.com/measurements/sm-019.html},
  doi    = {10.5281/zenodo.23330729}
}