# ClinicalTrials.gov results deadline vs FDA's March 2026 letters (measured 2026-10-03, not a paper yet)

Claim: 42 CFR 11.44(a) results "no later than 1 year after the primary completion date"; FDA press release 2026-04-13: letters sent 30 Mar 2026 to >2,200 parties / >3,000 trials, "seek voluntary compliance", internal analysis "29.6% ... have no results information submitted". Penalty 21 USC 333(f)(3): $10,000 + $10,000/day after 30-day notice (exhibits/21usc333.html). FDA's own notice page (exhibits/fda_notices_table.json): 8 Notices of Noncompliance 2021-2025, civil money penalty column empty on every row.

Data: ClinicalTrials.gov API v2, all interventional studies with primary completion 2017-01-18..2025-10-02 (216,474; studies.jsonl, pull.py). TrialsTracker (Oxford Bennett Institute) table as of 2026-10-02 (exhibits/trialstracker_trials_2026-10-02.csv; 50,516 tracked, 6,840 overdue, 10,101 reported late, 15,022 on time, 18,543 not yet due). Join by NCT: 31,739 of 31,973 due trials.

FINDING (letter effect): late submissions landing per month (TrialsTracker-derived dates = completion+365+days_late): 2026-03 118, 2026-04 174, 05 111, 06 92, 07 81, 08 86, 09 65; baseline 2025 ≈ 86-139/month. Six months after the letters 632 vs 624 in the same window of 2025 and 568 in the six months before. Net ≈ 70 extra trials against 6,106 overdue trials due before the letters (≈1%). Overdue today by registry: 6,833; 6,099 due before the letters; 134 carry a certified delay. Overdue by sponsor class (due before letters): OTHER 3,252, INDUSTRY 2,564, OTHER_GOV 114, FED 85, NIH 3. Top: Cairo University 70, Ain Shams 37, BTL Industries 31, Mass General 29, Rothman 28, USC 28, Brigham 25.

PITFALL: the API's ResultsFirstSubmitDate is only populated once results are POSTED (0 rows submitted-but-unposted), so registry-side counts for the last ~4 months are depressed by QC lag (2026-07 156, 08 70, 09 17 vs ~450/month). Use TrialsTracker's scraped "submitted" dates for recency; use the registry for anything older than ~6 months.

Band check: no-results share among due trials 6,833/31,739 = 21.5% (TrialsTracker definition) vs FDA 29.6% (their "highly likely ACT" set, not public). FDA's >3,000 letters cover about half of the 6,106 overdue at the time.

Files: pull.py, analyze.py, studies.jsonl (216,474 rows, ~1 GB), letter_effect.json, overdue_before_letters.json, exhibits/ (SHA256SUMS).