Eric Sutton is a senior patent counsel at Oracle and an adjunct professor at Chicago-Kent College of Law. Any views or opinions expressed by him in this article are solely his own and do not necessarily represent those of Oracle Corporation, its subsidiaries or affiliates.
As known to most patent practitioners,[i] a parent provisional does not lock in a priority date for a child non-provisional unless the parent provisional provides a written description and enablement of an invention that is claimed in the child non-provisional. This assessment is conducted on a claim-by-claim basis[ii] to see if the parent provisional adequately supports and enables the precise subject matter of each child non-provisional claim for which priority is desired. If none of the claims in the child non-provisional are fully supported and enabled by the parent provisional, then the priority date for all of the claims is the later filing date of the non-provisional. In this scenario, the provisional has no clear value.
Under our first-inventor-to-file system, references that are published or patent applications that are filed between an applicant’s parent provisional filing and the applicant’s child non-provisional filing open the door for prior art to the claims of the child non-provisional filing, if those claims are not fully supported and enabled by the parent provisional.
Patent applicants, especially for software-related technologies, often file provisionals just before product releases, presentations at conferences, or publications of white papers or marketing materials. In-house patent portfolio managers frequently reassure business leaders that the technologies are protected by the provisionals, and outside patent counsel often similarly reassure in-house patent portfolio managers. However, due to time and budget pressures that often impact provisionals, relying on a provisional for priority may be significantly riskier than relying on a non-provisional.
Consider the hypothetical example below:
In the example, suppose your company files many provisionals and, on average, files many non-provisionals — each about ten months after its corresponding provisional. You file the provisionals ahead of product releases, and your competitor follows up with very similar product releases within a few months. The competitor’s product releases and several publications about the underlying technology occur during the ten-month span between provisional and non-provisional filings. Any aspects of these publications that were not completely derived from your product would serve as potentially invalidating prior art to your non-provisional filings unless your provisional filings satisfy the written description and enablement requirements with respect to your non-provisional claims. This scenario could happen over and over again, and negative effects from this scenario could even be baked into your provisional and non-provisional filing strategies in a fast-paced product development environment.
In a big data analysis of provisional application quality, focused on comparing provisional application content to non-provisional claims using Innography®, LexisNexis® PatentAdvisor®, and self-developed software, I found that many patent attorneys seemingly do rely rather extensively on their provisionals.[iii] On average, applicants rely extensively on their provisionals by allowing an average of ten months of risk for intervening prior art by filing the first-filed child non-provisional ten months after the corresponding provisional. Whether or not that intervening prior art actually shows up is beyond their control, particularly if they have already let the cat out of the bag by sharing their technology with others.
How can we know, with high probability for a large set of provisionals, whether those provisionals provide support for their child non-provisional claims? While there are many approaches available, some are more or less time-consuming, more or less subjective, more or less scalable, and more or less reproducible. I chose to analyze a large chunk of application pairs[iv] to see whether word stems present in the non-provisional claims[v] were also present in the corresponding parent provisionals.[vi] I then further excluded word stems that often appear in patent claims to produce even more interesting results that focus in on the less common word stems that theoretically would be more troubling if missing from the provisionals. The results are objective for what they show, i.e., that provisionals are, on average, missing several word stems from the broadest independent claims of the non-provisionals.
The study does not conclude missing word stems or even missing uncommon word stems necessarily mean that there is no support or enablement in the provisional. On the contrary, the conclusion of this study depends on your personal experience and judgment. Suppose you know nothing about a large set of provisionals in your portfolio other than how many unique word stems and, more specifically, how many uncommon unique word stems from the broadest child non-provisional claims were missing from the provisionals.
My guess is that most people would be very comfortable if all of the word stems or uncommon word stems were present in the provisional (0% miss rate) even though it is still possible they have not supported or enabled the claims in particular scenarios. On the other hand, I think most people would be very uncomfortable if all of the word stems or uncommon word stems were absent from the provisional (100% miss rate) even though it is still possible that they have supported or enabled the claims. Different people will have different comfort levels between 0% and 100% depending on their own experiences and understanding, but at higher percentages one would likely begin to bet against priority rather than for it.
The study evaluated 1905 pairs of parent provisionals and non-provisionals, filed between March 2012 and July 2013.[vii] In the sample, there were 16,739 unique word stems in the first claims of the 1905 non-provisionals. For the parent provisional / child non-provisional pairs, an average of 7.0 unique word stems from claim 1 were missing from the parent provisionals out of an average of 49.8 unique word stems per claim, for a 13.3% unique word-stem miss rate. As mentioned above, applicants waited an average of 10 months between filing the first-filed non-provisional from the provisionals, which means applicants relied on these provisionals by taking on unnecessary risk, despite the significant average miss rates.
The chart below shows percentages of cases at different levels of word stem miss rates.
As shown, about 19.3% of cases were missing 25% or more of the unique word stems, a point that is likely beyond the comfort level of most patent practitioners. That said, 21.5% of cases had a 0% miss rate, indicating that a large group of patent practitioners is avoiding the written description and enablement risk that comes with varied language.
After a careful analysis of stems at different thresholds, it turned out that the top 3% of stems (516 stems) occurred 30 or more times in the sample and included stems that, for the most part, would likely not be considered relevant to a written description analysis, whether or not a synonym was present in the provisional. The choice of a cutoff level was subjective, but the cutoff itself was objectively applied in the sense that all terms above the cutoff were removed from the study and all terms below the cutoff remained in the study. The rationale used for the subjective cutoff level was to pick a level that was just beyond any level including the most common terms occurring in computer-related claims. Although the sample of cases was not limited to computer-related cases, computer-related terms excluded as being part of the top 3% were, for example:
comput, store, storage, devic, service, server, system, machin, network, send, receiv, transmit, communic, provid, transit, instruct, engine, modul, configur, process.
Using the list above as a proxy for familiar common terms was helpful in picking a cutoff point even though a significant portion of the sample did not involve computer-related innovation. Stems beyond the top 3% boundary remained in the study. These less common stems occurred 29 or fewer times in the sample of 1905 claims.
Should the miss rates go up or down after removing the common stems? My hypothesis was that miss rates would go down because terminology such as “compris” (excluded as within the top 3%) might often be added by attorneys down the road as claims are solidified. At a minimum, more cases would have a 0% miss rate when fewer stems are searched. Others predicted that the miss rates would go up for other buckets because common terminology is usually present in the provisional, at least as boilerplate. Therefore, removing the common terminology would reveal the true incompatibility between provisionals and non-provisional claims.
I was not willing to tolerate the same miss rates as I did in the analysis for all stems. Because we are focusing on only the stems that occur relatively infrequently (in 29 or fewer out of 1905 claims), it is significantly more likely (though not certain) that these stems would reflect content that could be relevant to written description or enablement issues.
Others were right to think the results would get worse. Removing the top 3% of stems reduced the average number of unique stems per claim by 27 stems (45.8%) to 22.8 stems. Out of those 22.8 uncommon unique stems remaining, on average, provisionals were missing 5.3 for an average miss rate of 17.7%, up by 4.4%. Personally, I was shocked at this result and felt uncomfortable on behalf of the patent bar.
The chart below shows percentages of cases at different levels of uncommon word stem miss rates.
As shown, about 26.7% of cases were missing 25% or more of the unique word stems. The 26.7% uncommon stems mark is up from the 19.3% mark for all stems, reflecting a delta of 7.4%. If having 25% or more uncommon stems missing is beyond the comfort level of most patent practitioners, then practitioners should be uncomfortable with at least 26.7% of the provisionals being filed. I was equally as shocked to see that 9.8% (up by 5.5%) of cases were missing 50% or more of the uncommon stems. The largest cluster was 24.1% (up by 2.6%) having a miss rate of 0%, although there were a few miss rates even at 100%.
Due to the skewed distribution with miss rates falling at both ends of the spectrum, statistical metrics such as standard deviation did not provide much insight into the dataset beyond the charts shown. Nevertheless:
- a miss rate of 21% (30% of the sample) was greater than one standard deviation away from the peak of 0%,
- a miss rate of 34% (19% of the sample) was 1.645 times the standard deviation away from the peak (considered significantly different than the peak of 0% for a right-tailed distribution, p = .05, if the peak miss rate of 0% is considered to be the target). Note this analysis incorrectly treats the 100% miss rates as deviating from the target, when at least these cases clearly have a different target. This unnecessarily inflates the miss rate required to qualify as significantly different.
- a miss rate of 42% (14% of the sample) was greater than two standard deviations away from the peak.
It turns out that, in the U.S., the patent office does not even check for provisional application priority unless there is intervening prior art. That said, priority issues do come up in courts where litigants are equipped with discovery and additional resources.[viii] Based on the results of this study, it seems that there is some low hanging fruit where the patent office could object to a priority claim that is extremely unlikely to be accurate and, after patent office review, turns out to be inaccurate regardless of whether the patent office has found intervening prior art.
Although I am not qualified to opine on the laws of specific other countries, I do know that many other countries generally have a stricter view on specification support for the purposes of preserving priority. According to Heather McCann, a European Patent Attorney at EIP: “Having a mismatch in the literal terminology of the U.S. provisional and a child PCT application or a child European direct national filing could significantly reduce the possibility of obtaining patent protection and successfully defending a validity challenge, post-grant, if the contents of the U.S. provisional application were published (even by the applicant) prior to the child PCT or child direct national filing.”[ix]
Unfortunately for U.S. applicants, unless they are keen on priority issues, non-U.S. patent offices might be able to reject cases from U.S. applicants more frequently based on a difference between common U.S. practice and foreign written description and enablement laws.
Because inequitable conduct involves intent, which is fact-specific, this article does not delve any deeper into this potential issue. That said, it would be interesting to hear an argument and supporting explanation against such a finding if the applicant incorrectly claimed priority even though 100% of the uncommon word stems were missing from the provisional.
The study is an objective report of terminology disagreement between provisionals and non-provisionals. Individual attorneys will vary on how much terminology disagreement they find comfortable in light of their understanding of 35 U.S.C. 112(a).
With 27% of the provisionals missing a quarter or more of the broadest non-provisional claim’s uncommon unique word stems, there is much room for the patent bar to either improve provisional application quality or avoid inaccurate priority claims. Perhaps this could also be accomplished by prioritizing the conversion to a non-provisional or filing incremental provisionals as more details become available, rather than waiting 10 months for the conversion.
[i] For more information on the written description and enablement requirements, particularly as they apply to software-related innovation, see Sutton, E. Software Patents: A Practical Perspective (ver. 2.2, Sept. 2017), available on Amazon at http://a.co/9xcPNr9 (a software patents textbook in support of a law course offered at Chicago-Kent College of Law).
[ii] Los Angeles Biomedical Research Institute at Harbor-UCLA Medical Center v. Eli Lilly and Company, 849 F.3d 1049 (Fed. Cir. 2017) (inadequate written description in provisional); Phigenix, Inc. v. Genentech Inc., 238 F.Supp.3d 1177 (N.D. Cal. 2017) (inadequate written description in provisional); See Storer v. Clark, 860 F.3d 1340 (Fed. Cir. 2017) (provisional not enabling); Dynamic Drinkware, LLC v. National Graphics, Inc., 800 F.3d 1375 (Fed. Cir. 2015) (“priority claims are not examined by the PTO as a matter of course, and consequently are not entitled to a presumption of adequate written description support in the provisional application,” and “[a] provisional application’s effectiveness as prior art depends on its written description support for the claims of the issued patent . . . .”).
[iii] N=1905, down from an original sample of 2341 after excluding: (a) unparseable provisionals, (b) clearly incomplete claims (for example, a method claim ending at the colon), and (c) later-filed provisionals that paired with the same first-filed non-provisional; provisionals filed Mar. 2012 to July 2013 across all companies, spanning the AIA transition; raw provisional pdfs available from PAIR, provided by Innography® in Jan. 2017; filing dates provided by Innography® directly and through use of their excellent patent portfolio analytics tool; sample spread over TC 1600, 1700, 2100, 2400, 2600, 2800, 3600, and 3700.
[iv] N=1905, same sample as above, comparing claim 1 of the non-provisional to the entire contents of the provisional application pdf, drawings included, based on Official Character Recognition (OCR) performed in batch by Adobe® Acrobat® 9 Standard, version 9.0.0. Provisionals were paired up with their first-filed non-provisionals based on data provided by Innography®. The claims of the non-provisionals were downloaded from Innography®.
[v] The first claims of the first-filed non-provisionals were stemmed using a popular word stemming library to result in unique word stems.
[vi] The text of the corresponding provisionals, as a result of the Adobe® Acrobat® 9 Standard OCR (see above), was searched for any text that would match the unique word stems. Spacing issues and formatting issues were corrected before performing the comparison.
[vii] The provisionals involved were filed between March 2012 and July 2013, bridging the AIA transition. If attorneys changed practice from pre-AIA to post-AIA, those changes were not sufficiently detectable in this sample. Although I personally believe that attorneys have not changed practice post-AIA; to fully test that hypothesis, I would need a sample significantly earlier than the AIA and a separate sample significantly later than the AIA. These samples simply were not available to me when I started the study.
[viii] Los Angeles Biomedical Research Institute at Harbor-UCLA Medical Center v. Eli Lilly and Company, 849 F.3d 1049 (Fed. Cir. 2017) (inadequate written description in provisional); Phigenix, Inc. v. Genentech Inc., 238 F.Supp.3d 1177 (N.D. Cal. 2017) (inadequate written description in provisional); See Storer v. Clark, 860 F.3d 1340 (Fed. Cir. 2017) (provisional not enabling); Dynamic Drinkware, LLC v. National Graphics, Inc., 800 F.3d 1375 (Fed. Cir. 2015) (“priority claims are not examined by the PTO as a matter of course, and consequently are not entitled to a presumption of adequate written description support in the provisional application,” and “[a] provisional application’s effectiveness as prior art depends on its written description support for the claims of the issued patent . . . .”).
[ix] See, for example, European decision T 1983/14 “Purchase of mobile top-up/JANUARY PATENTS LTD.” (Nov. 23, 2015) (“The case law of the Boards of Appeal requires that the subject matter be directly and unambiguously derivable by the skilled person from the disclosure as a whole. The expressions ‘invoiced together’ and ‘charged for together’ . . . disclosed no more than that transactions could take place on the same machine[, not] multiple products being invoiced as one transaction.”); European decision T 2061/13 (June 20, 2017) (“subject-matter of claim 1 is . . . not directly and unambiguously derivable by the skilled person . . . such that . . . claim 1 does not have a right to priority from P1.”).