Methods & sources

Every figure on this site is computed from an archived public source. This page says which source, how the computation works, where it was checked against an independent published figure, and what the data cannot be used for.

Sources

The 2018–20 corpus

QuantGov / Mercatus Center, "Section 301 Tariffs" bulk dataset, version 1-0 — https://www.quantgov.org/tariffs, file Tariffs-Section-301_1-0.zip, last modified 2021-12-01, retrieved 2026-09-09. Upstream: USTR exclusion request dockets, 2018–2020.

Lists 1–2 remain independently verifiable on regulations.gov. Lists 3–4A were filed to exclusions.ustr.gov, which no longer resolves and was not archived; for those two tranches this snapshot is the only accessible record found. Its totals reproduce the GAO/CRS published figures to within 2 records in 52,747.

The 2024–25 machinery docket

USTR Comments Portal, docket USTR-2024-0020, retrieved 2026-09-09 directly from comments.ustr.gov.

Cross-check against an independent figure

The US Government Accountability Office and the Congressional Research Service report 6,804 granted and 45,942 denied across the four tranches. Computed independently from the archived dataset, this site gets 6,802 granted and 45,944 denied — a total absolute difference of 4 records in 52,746.

That agreement is the reason the archived snapshot is treated as faithful for the two tranches whose original source is no longer reachable.

How the rates are computed

A grant rate is granted ÷ (granted + denied). Requests recorded as pending are excluded from the denominator, not counted as denials.
Rates are computed within a tranche. Cross-tranche figures appear only where labelled, because each tranche ran a materially different process.
A rate is suppressed where fewer than 20 requests were decided. Suppressed cells read “n too small” and never display as zero.
The “excluding mass filers” view drops filers with more than 100 requests. The threshold is arbitrary and stated so it can be argued with; both views are always shown side by side.
Nothing is imputed. A field the source leaves blank stays blank. No default, placeholder or inferred value is ever substituted.

What this data cannot support

It cannot predict an outcome. Every process that produced these rates has closed, and no live action accepts exclusion requests. There is nothing pending for a base rate to bear on.
It cannot tell you which argument works. The 2018–20 records carry outcomes but not the grounds argued. The 2024–25 machinery records carry the grounds but no outcomes. Neither supports a grant-rate-by-argument claim, and this site publishes none.
It cannot be read across regimes. The 2018–20 process asked whether the requester would suffer severe economic harm. The 2026 exemption criteria ask whether the economy would be disrupted. Those are different questions and their numbers are not comparable.
It says nothing about confidential filings. Only the public version of a docket submission is used, and nothing is inferred about material withheld as business-confidential.

What is deliberately not published

Per this project's compliance standard: no requester or law-firm win-rate rankings, no table of which domestic producers oppose the most requests, and no request-level register. Filer concentration is disclosed as a distribution because it materially changes how the headline rate should be read — not as a league table.

No sentence here takes a position on whether any tariff is good policy. The processes are described as they operate.

Corrections log

Every number on this site is reproducible from the archived sources with the build scripts behind it. If a figure is wrong, it is a bug in a script or an error in a cited source, and either is worth reporting. Corrections are listed here with their dates rather than edited away.

Date	What was wrong	What changed
2026-09-10	The homepage and the expiring-exclusions page stated as fact that these exclusions would not be replaced after 9 November 2026. That was a forecast, and the record contradicts it: this set of exclusions has been extended seven times since 2022.	Removed. Both pages now show every past extension and the date it was published. Forecast phrasing about agency decisions now fails the site build.
2026-09-10	Eight exclusion descriptions on the expiring-exclusions page and in its CSV contained page-header text from the tariff schedule PDF inside the product description.	Removed from the extract. A check now requires every published condition to appear word for word in its description.
2026-09-10	The scope data behind paid reports read ten-digit provisions as six digits, read some years and page headers as tariff headings, dropped headings 7212 and 7213, could treat a neighbouring code as a match, and left out List 4A, the 2024 review increases and Section 232 autos without saying so. The data catalogue also placed the Section 301 forced-labour action in heading family 9903.02, which the schedule itself describes as a transshipment heading.	All corrected and covered by tests. Every heading family is now described in the schedule's own words, and the forced-labour action is stated as absent from the captured schedule. No paid report had been delivered.
2026-09-10	The live process map gave the Section 232 inclusions cadence as three windows a year for steel and aluminium and four for auto parts — taken from the procedures, not from what was actually published.	Now states what the Federal Register shows: one steel and aluminium window since June 2025, and auto-parts windows in January and April 2026.
2026-09-10	The correction above, the paid report's first page, the pricing page and the live process map said the Section 301 action on 60 economies was absent from the captured tariff schedule, because the phrase "forced labor" never appears in it. The action is in the schedule, as U.S. note 52. The same report labelled each uncovered heading family by its first heading, which hid live provisions: Section 232 duties on unmanned aircraft were listed under "bath preparations".	Every uncovered provision is now listed by the U.S. note it cites, in the schedule's own words, with a source notice only where the notice has been checked against the schedule. No absence is asserted; the build fails if one is, and fails if any uncovered provision is left out of the report.