Usually the elected invention is prosecuted and the rest is not refiled, which is not the same as losing it. Of 239,936 restricted US utility applications filed 2020 to 2022, 39.5% were patented with no continuing application filed. Reading 107 of those by hand, the examiner had stated that withdrawn claims were rejoined in 25.2% of them. That split was noticed after the results rather than planned, so it is a sample observation and not a measured rate for the class. What it establishes is that this class cannot be read as the non-elected inventions being given up.
A restriction requirement is the examiner saying an application claims more than one invention and asking which one to examine. It is not a rejection, and it is one of the least evenly distributed events in prosecution: whether you ever see one depends almost entirely on your field.
Share of examined applications receiving a restriction, by CPC section
US utility applications filed 2020, of those an examiner acted on. Overall 18.2%. Bars run from zero.
Chemistry and metallurgy (C)
45.6%
Human necessities (A)
29.7%
Textiles and paper (D)
28.5%
Performing operations (B)
20.6%
Mechanical engineering (F)
14.9%
Fixed constructions (E)
13.7%
Physics and computing (G)
7.5%
Chemistry and metallurgy draws a restriction six times as often as physics and computing. The overall 18.2% is an average across that spread and describes neither end of it. Counting against every application filed rather than only those examined moves each of these figures by less than a point, the largest gap being 0.76 points in biotechnology.
Our own counts from the USPTO record, re-verified 2026-11-07.
The same rate, by technology center
US utility applications filed 2020, of those an examiner acted on. Bars run from zero.
1700 Chemical and materials
37.1%
2800 Semiconductors
17.0%
3600 Transportation, e-commerce
9.9%
2100 Computer architecture
2.5%
Biotechnology at 52.9% and computer architecture at 2.5% are the same requirement under the same statute, twenty-one times more likely in one place than the other. Technology center 3900 is left out: it has 170 examined applications in this cohort, too few to report.
Restriction practice follows from unity of invention, so a field whose applications routinely claim a compound, a composition and a method of use will see it constantly, and a field whose applications claim one system will not.
When they arrive
Restrictions are a first-action event, and in biotechnology, where they are commonest, they are nearly all in by the third year.
Biotechnology applications restricted, by months since filing
Cumulative share of technology center 1600 applications that had received a restriction. Bars run from zero.
73% of biotechnology restrictions arrive between 12 and 36 months. After 48 months the line is flat: the last year adds 0.2 points. If an application in this field has not drawn a restriction by its fourth year, it very probably never will.
What happens next
Four outcomes, and together they account for every restricted application.
Outcome of a restricted application
239,936 restricted US utility applications filed 2020 to 2022, status as of 26 October 2026. Every one of them is in the bar.
A continuing application was filed
28.1%
Patented, no continuing application filed
39.5%
Drawn as one whole because these four are one: on this definition they have no overlap and leave nothing unclassified. The segment widths use the unrounded shares so they sum to exactly 100%; the legend rounds to one decimal, which is why the printed figures add to 100.1%. The definition was chosen after the results were seen and assigns each application to the first class it qualifies for, which is what removed a 58 application overlap and 54 miscounted abandonments in the original pass. It describes what was found and tests nothing.
Our own counts from the USPTO record, re-verified 2026-11-07.
The largest class does not mean the rest was abandoned
It is tempting to read "patented, no continuing application filed" as the applicant giving up the non-elected inventions. The record does not support that reading, and this is the finding that changed how this page is written.
On a seeded sample of 236 restricted applications read by hand, the examiner stated that withdrawn claims had been rejoined in 16.1% of restricted applications overall (95% interval 12.0 to 21.3). That measurement was planned in advance, down to the sample and the method.
Splitting the same sample by outcome puts most of the rejoinders in one class: 25.2% of the 107 sampled applications that were patented with no continuing application, against 12.2% where a continuing application was filed and 2.9% among those abandoned. That split was noticed after the results and was not in the plan, so the 25.2% is a sample observation rather than a measured rate for the class, and the planned 16.1% is the firmer figure. It is also computed on the corrected classes, which were themselves settled after the results.
Rejoinder happens when the elected claims are allowable and the withdrawn ones depend on or correspond to them closely enough that the examiner takes them back up. The restricted subject matter is then in the granted patent, with no continuing application anywhere in the record. That is the point these figures support, and it holds on the planned 16.1% alone.
Both figures are lower bounds. They count only cases where the examiner said so in terms, on a sample, so the true share is at least that and the intervals are wide. A superseded earlier pass put the overall figure at 16.5% by a method that read the oldest actions and counted any mention of rejoinder rather than a statement of it.
When a child is filed, how often does it grant
Filed a continuing application, and reached a granted child
Restricted applications filed 2020 to 2022, by technology center. Bars run from zero.
2400 Networks, filed a child
39.8%
2400 Networks, reached a granted child
22.6%
2600 Communications, filed
36.4%
2600 Communications, granted child
23.4%
1600 Biotechnology, filed
26.0%
1600 Biotechnology, granted child
8.2%
These pairs are nested, not alternatives: every application that reached a granted child is also one that filed a child, so the second bar is a subset of the first. Read as a conversion rate, 31.6% of biotechnology applications that filed a child reached a granted one, against 56.7% in networks. Biotechnology both files least often and converts least often.
Restricted applications are granted less often
Share granted, restricted against not restricted
US utility applications filed 2020 to 2022, within the same technology center. Bars run from zero.
1600 Biotechnology, not restricted
59.6%
1600 Biotechnology, restricted
46.4%
1700 Chemical, not restricted
71.5%
1700 Chemical, restricted
60.6%
2400 Networks, not restricted
86.6%
2400 Networks, restricted
81.5%
The gap is 13.2 points in biotechnology, 10.9 in chemical and 5.1 in networks. This is an association and not a cause: an application that claims several inventions may be harder to prosecute for the same reasons it drew a restriction, and the comparison holds the technology center fixed but nothing else.
Our own counts from the USPTO record. These describe what happened across past applications. They say nothing about any particular application, they are associations rather than causes, and none of this is legal advice.