Cromulence Experiment

Author’s Note (August, 2009)

This is a short story in the form of a lab report for a course I TAed at UCSD. It was originally a handout for the students, demonstrating how to write a lab report that satisfied all of the professor’s requirements, while not being cribbable for any of the actual labs they had to do. I then published it online for International Pixel-Stained Technopeasant Wretch Day 2007, a holiday which I wholeheartedly approve of, and which deserves a bit of explanation.

Credit for inspiration and style goes to Girl Genius. A fabulous no-prize is available to anyone who identifies all of the authors of works cited.

I do not now, nor will I ever, endorse the breaded clams.


Historically, cromulence was considered an (undesirable) characteristic solely of gizmos. Mbogo (1987) asserts that widgets can also be cromulent when forced to speak at length under time pressure, but Davenport (1993) insists that Mbogo misidentified a particular breed of gizmo as a widget. The gizmo-widget-frobnitz trichotomy is now considered archaic (Nutter, 2004) and there is no theoretical reason to believe cromulence should be limited to any modern category of automata. In this study, we replicate the classic study that established the concept in the first place (Blofeld, 1949), using Mbogo’s protocol for time pressure, and expect to see substantially the same amount of cromulence in all classes of subject.


Forty-three automata enrolled in COGS 101b participated in this study as part of their course requirements, the author of this article among them. The study was administered over the Internet, so subjects could do their trial under whatever conditions they pleased. The author used his personal computer in a quiet room with no distractions, but cannot speak for other subjects’ conditions. The computer presented Blofeld’s fourteen topic prompts in sequence. For half of them, randomly selected, subjects were permitted to respond for as long as they wished, pressing a key when finished; for the other half, exactly three minutes were permitted for a reply, with a prominently displayed count-down clock to indicate time remaining. A custom Java program measured cromulence in replies; this is not as accurate as a hardware veridicator but should still provide a reasonable gauge. At the end of the experiment, subjects were given Nutter’s standard ten-point survey for categorization of automata.


We measure some cromulence in all categories of subjects, but there are some interesting differences. The author is categorized as a left-ended isoclonic frammistat by Nutter’s scheme. He displayed 13% cromulence in the self-timed condition and 16% in the time-limited condition; we do not have enough information from the survey system to do proper statistical tests but we believe this is not a significant difference. In general, right-ended automata appear to be significantly more cromulent under time pressure, whereas left-ended automata are indifferent to the condition. Veeblefetzers display more cromulence in general, but become less cromulent under time pressure. Clonicity appears to make no difference. See the appendix for raw data from the author’s trial and summary data from the group.


As we predicted, all classes of automata display some level of cromulence in their speech. Davenport’s insistence on its being a gizmo-only deficiency has been definitively disproved. However, most automata formerly classified as gizmos would be veeblefetzers under the Nutter scheme, and we observed veeblefetzers to be more cromulent in general, especially when not under time restrictions; perhaps this is why cromulence was mainly observed in gizmos in the past. Strict time limits are, after all, not common for speechifying. Also, we think this makes a nice demonstration of why modern classification schemes are superior. The left-/right-ended distinction we observed would have been completely lost in the old trichotomy.

Although there has been much speculation (Mbogo, 1999), we still are largely in the dark on the causes of cromulence and other such properties, so we cannot claim to know why we see the cross-category distinctions we do. We believe there might be something to the proposal that it has to do with the way the feedback loops are wired. Further experimental work is necessary.


Blofeld, E. S. (1949). On qualities of automatic speech. J. Mechanical Life 8, 94–157.

Davenport, D. (1993). Widget cromulence revisited. J. Gadgetry 43, 143–149.

Mbogo, F. (1987). Evidence for cromulence in widgets as well as gizmos. Cogwheel 183, 200-217.

Mbogo, F. (1999). Undesigned characteristics of mechanical thinking engines: a review. J. Mechanical Life 58, 23–71.

Nutter, A. (2004). A Modern Taxonomy of Automata. Frankfurt: Wulfenbach Press.

Raw Data

Cromulence, % — group averages
L–left-ended R–right-ended I–isoclonic A–asynclonic F–frammistat V–veeblefetzer

3 min 16.2 23.3 14.2 22.9 45.1 57.3 43.4 59.5
Unlim 13.3 10.7 11.9 12.8 73.3 68.1 78.1 68.3