When a Limit Actually Constrains the Account

Intellectual Practices of Keith Lyons

Two sentences can both sound cautious.

Only one may contain a limitation that actually constrains the account.

That difference became a problem for the whole-corpus study of Clyde Street.

What had to be present before a caveat counted as more than cautious wording?

On 3 July 2015, Keith Lyons published a short post about the penalty shoot-out between Germany and France at the Women’s World Cup. He had reviewed all ten penalties and recorded what he thought he could see about the relationship between each kick and the goalkeeper.

Then he inserted a qualification:

“The video I used to review the shoot out was not of high quality. From what I saw…”

He went on to say that, from what he could see, all ten penalties were goalkeeper independent.

Nothing dramatic happened in the post. Keith did not abandon the analysis. He did not announce a new method. He did not turn uncertainty into the subject of the article.

But the wording matters because the source of the limitation is visible. The video quality was poor. And the limitation had an observable consequence: Keith qualified the inference he was willing to make from that video.

At first glance, that can look like ordinary caution.

Six months earlier, however, Keith had written another post about penalty shoot-outs, this time at the 2015 Asian Cup. Describing several penalties, he wrote that they “appear to be” examples of goalkeeper-dependent kicks.

That sounds cautious too.

Yet the whole-corpus study treated the two posts differently.

In the Asian Cup post, the tentative language is visible, but the source does not identify a particular condition that is doing constraining work on the inference. There is no equivalent of the poor-quality video in the later World Cup post. The phrase “appear to be” signals tentativeness, but the post does not show what, specifically, is limiting the claim.

So one post qualified as an instance of the finding. The other did not.

That small difference gets to the heart of the whole-corpus finding the study called Preserves Limits, Absence and Incompleteness.

The finding was not that Keith sometimes sounded cautious.

It was narrower.

For a case to count, the study had to be able to locate a particular absence, counterexample, failed verification, uncertainty or incomplete condition in Keith’s own account — and then locate what that condition actually constrained.

In reader-facing terms:

Can we locate the specific limit, and can we locate what that limit constrains?

That is our translation of the research boundary. It is not a rule Keith formulated for himself.

Cautious grammar is not enough

The matched penalty-shoot-out cases help because they prevent the finding from drifting into personality.

It would be easy to turn a collection of phrases such as “from what I saw”, “appears to be”, “perhaps” or “I may be missing something” into a portrait of Keith as a cautious person. The study did not do that.

It looked instead for an observable operation inside the inquiry.

In the Women’s World Cup post, the poor video quality bounds the observation. In the Asian Cup post, tentative grammar appears without an equally locatable limiting condition.

That distinction matters because almost any inquiry contains uncertainty. Almost any writer can sound cautious. If caution alone were sufficient, the finding would expand until it meant very little.

The whole-corpus boundary was therefore stricter: a limit had to do something to the account.

Sometimes it restricted the strength of an inference. Sometimes it bounded what a measure could represent. Sometimes it prevented a remembered claim from becoming verified evidence. Sometimes it kept missing representation visible rather than allowing a partial account to stand for the whole.

Those are different forms of the same recurrent operation. They are not a taxonomy and they do not imply that every kind of limitation was handled in the same way.

When memory does not become evidence

An early Clyde Street post shows a different kind of constraint.

In September 2008, during the CCK08 course, Keith was thinking about the diffusion of ideas. While preparing the post, he tried to track down a claim he remembered from an economic-history course he had taken in 1971: that during the English agrarian revolution the planting of turnips had spread at roughly a mile a year.

He could not find a source for the figure.

The post does not quietly convert the recollection into an established historical fact. Keith identified it as a memory, recorded that he could not verify the metric, and then turned to other material he could locate about the diffusion of ideas.

The failed verification remains part of the account.

That does not make the post a general lesson in fact-checking. Nor does it establish that an unverified recollection is useless. Keith still played with the turnip image in the post.

The narrower point is evidential: the remembered statistic and the verified material do not silently acquire the same status.

The limit is specific — he could not find the source — and that failed check constrains what the recollection can become inside the inquiry.

When a measure tells you less than it seems to

In February 2015, Keith used previous-season rankings to follow results across four rugby union competitions.

He described ranking status as a macro indicator. It gave him a simple way to notice whether current results aligned with previous finishing positions and to guide his interest in performance infrastructures.

But he also named what the indicator did not take into account: venue, team selection, referee and weather conditions.

Those omissions do not prove that the ranking indicator is good. Listing missing variables does not validate a measure.

What the exclusions do is mark the scope of the inference.

The ranking measure can organise one kind of comparison. It cannot, on its own, explain the full set of conditions shaping a match result.

Again, the finding lies in the connection between a visible limitation and a visible constraint. The omitted factors are not merely decorative caveats. They bound what the macro indicator can reasonably stand for.

When the missing material is representation

A month later, Keith reflected on a four-week open online course in sport informatics and analytics, #UCSIA15.

He described the course as a “self-consciously modest excursion”. He also acknowledged the remarkable expertise growing in the field that was “not represented (but hinted at)” in the course.

The missing representation mattered because it limited what the course could reasonably stand for.

Keith could describe what the participants had shared. He could value the connections that had formed. He could hope that the course marked the beginning of a longer journey.

But the course itself could not silently become a complete representation of sport informatics and analytics.

That is a different kind of incompleteness from poor video or failed verification. Nothing in the source tells us what the absent people would have contributed, and the study did not infer it.

The constraint is simpler: an incomplete field representation remains visibly incomplete.

The article can therefore say less than a stronger claim might tempt us to say. Keith did not possess the absent expertise merely by acknowledging that it existed. Nor does non-participation become evidence about the views of people who were not there.

The limit governs the reach of the account, not the imagined content of the absence.

A good caveat can still belong to somebody else

The ownership boundary is easiest to see in another 2015 post.

Keith had been reading about digital object identifiers. He drew attention to an explanation by Laurence Horton, including Horton’s warning that a DOI should not be treated as a proxy for data quality.

That is a substantive epistemic caution. It is relevant to the topic. Keith clearly valued it enough to reproduce and discuss it.

But the warning belongs to Horton.

The study therefore did not treat Horton’s caveat, by itself, as evidence that Keith was making the limiting move himself.

This is an important constraint on the public interpretation of the whole corpus.

A limitation stated by somebody Keith was reading does not automatically become evidence of Keith performing the limiting operation himself.

Keith could select a source, connect it to his own interests and make it visible to readers. The intellectual ownership of the source’s warning remains with the person who made it unless the record shows Keith performing a distinct, source-supported operation of his own.

Without that ownership boundary, a caution Keith quoted could easily be mistaken for a limiting move he had made himself. The study did not make that substitution.

What the whole corpus established

The frozen register for this finding contained 85 approved cases across Clyde Street from 2008 to 2020. Seventy-six were classified in the strongest evidence category. The study also retained 72 rivals and near-misses that helped constrain what could count.

The number is evidence of recurrence and temporal persistence. It is not evidence that this was Keith’s most important intellectual practice, his most distinctive characteristic or a continuous feature of everything he wrote.

The cases varied.

Some involved failed verification. Some involved source-quality qualifications. Some identified assumptions or omitted variables in models and indicators. Some kept missing representation visible. Elsewhere in the frozen register, limits could affect which records were admissible for comparison or preserve counterexamples that resisted a summary measure.

But the approved cases were held together by a narrower empirical requirement.

A particular limiting condition had to remain visible in Keith’s own account and have an observable effect on what he was willing to claim, infer, represent or count as evidence.

The negative cases mattered just as much.

Tentative wording alone was not enough.

Merely reporting an exception was not enough.

A caveat supplied by somebody else was not enough.

And unfinishedness by itself was not enough.

Those exclusions are part of the finding. Without them, this practice would collapse into a general story about caution, incompleteness or humility.

The whole-corpus study did not support that story.

It supported a bounded claim: across repeated Clyde Street records, specific Keith-owned limits were sometimes preserved in ways that materially constrained the account being made.

That is an observable intellectual operation, not a personality trait.

What the distinction leaves with the reader

Once a limitation is visible, the practical question of what to do about it comes later. This finding stops one step earlier.

The finding asks something more basic.

Has a specific limitation actually become operative inside the account, or is the writing merely cautious, incomplete or surrounded by exceptions?

The Clyde Street cases do not give readers a procedure for answering that question. They do not establish that every visible limit is important, that every constrained claim is valid, or that preserving a limitation makes an analysis good.

What they make available is a distinction.

A sentence can sound cautious without identifying a consequential limit.

A limitation can be real without belonging to Keith.

An exception can be visible without constraining an inference.

And a specific problem can matter because it restricts the reach of what the evidence is allowed to support.

That leaves a question for the reader — one enabled by the research finding rather than taught by Keith as a method: