Non-redundant taxonomic construction-grammar models posit that only fully productive patterns such as let alone (Fillmore, Kay, & O’Connor, 1988), what’s X doing Y (Kay & Fillmore, 1999), or all-clefts (Kay, 2013) qualify as constructions because they license an infinity of expressions. Redundant models claim that, despite subregularities and exceptions, partially productive patterns also count as constructions, providing the overall meanings of such patterns are not the strict sums of their parts (Goldberg, 2006, 2009). Because determining what counts as a construction is a major bone of contention between redundant and non-redundant construction grammar taxinomies, I examine the productivity of a multiple-slot construction in the British National Corpus (XML edition): it be adj to v that. (1) It is important to note that 33% of the sound’s total land area will be protected. (BNC–CRC) (2) It is fair to say that it has the sweep and scope of an encyclical. (BNC–CRK) (3) Even so, it was hard to believe that his birthday in two days’ time would be only his twelfth. (BNC–FRF) There are 2136 tokens of the pattern in the corpus. If we concatenate the two lexical slots (A and Vinf), the type count is 578. There are 162 adjective types and 104 verb types. Presumably, according to Kay (2013), it be adj to v that is not likely to be a “construction” but rather a “pattern of coining” due to its limited type productivity. Contra Kay (2013), and following Author (2015), I demonstrate that even patterns that are not fully productive often have subregularities that are. Quantitative approaches have developed measures to capture the productivity of morphological processes in corpora. Such measures have been adapted to the study of multiple-slot constructions (Zeldes, 2012, Author, 2015). The most obvious measure is the type count V of the members of a category C in a corpus of N construction tokens. A major issue with indexing a measure on type is that it does not discriminate between established forms and new forms. Indeed, types are not distributed uniformly in a corpus, and the larger the corpus, the harder it is to find innovations. To keep track of vocabulary development across a given corpus, one can plot the number of types against the number of tokens at multiple intervals. One obtains a vocabulary growth curve (henceforth VGC) (Baayen, 1993). At first, the curve is expected to rise steeply as most tokens define a new type. Then, as more text is scanned, more and more tokens are included in already defined types, and the curve flattens out gradually. The VGC in Figure 1 shows that a construction can be productive (as indicated by the steep curves of A–Vinf concatenations) even though the productivity of its adjectival and verbal constituents is limited (as indicated by the flat curves of A and Vinf). VGCs rely on a set of measures described in Baayen (1989), Baayen and Lieber (1991), and Baayen (1993). These measures draw on the idea that the number of hapax legomena of a given grammatical category correlates with the number of neologisms in that category, which in turn correlates with the productivity of the rule at work. Thus, productivity is a factor of both a large number of low-frequency words and a low number of high- frequency words. This combination contributes to parsing effects which maintain the affix activated in the minds of speakers. Following Author (2015), I augment these hapax-based productivity measures with a combination of symmetric and asymmetric association measures involving χ2 and ∆P (Allan, 1980). Thanks to this combi- nation, I show that although the productivity of it be adj to v that is limited at its most schematic level, some partially filled subschemas such as it be hard/important/easy/difficult/possible/necessary/reasonable/impossible to v that and it be adj to think/say/suggest/know/assume/realize/see that are arguably productive.