So no, the chess analogy isn’t claiming rational discourse is literally a game. It’s forcing a distinction you keep trying to blur, viz. that clarifying the conditions of intelligibility isn't the same thing as arguing for a claim within those conditions or parameters. You can have meta-level norms without turning bedrock conditions into ordinary premises. And pretending otherwise is exactly how the issue of global doubt and endless “improvement” talk becomes performative rather than really answerable. — Sam26

While I can't speak for @J, I can say that it hasn't been my intention to collapse everything into one level. I take it that the distinction between levels has been explicitly granted, and that we're now disputing whether the meta-level is inside or outside of rational normativity as such. For me, it's not about arguing for system-closure, or for some Archimedean stand-point outside of inquiry. It's about acknowledging that reason can come to understand the conditions of its own operation, and that to do so is itself a rational achievement.

For me, it's not about arguing for system-closure, or for some Archimedean stand-point outside of inquiry. It's about acknowledging that reason can come to understand the conditions of its own operation, and that to do so is itself a rational achievement. — Esse Quam Videri

I suspect that what’s at stake here is, at least in relation to Wittgenstein, is to what extent we treat understanding and reason in terms of adequation and conformity vs creation, enaction and becoming.

Thanks for all of the replies. I'm trying to think of another subject for a thread. My philosophical focus tends to be very narrow, but hopefully I'll think of something that's interesting.

I suspect that what’s at stake here is, at least in relation to Wittgenstein, is to what extent we treat understanding and reason in terms of adequation and conformity vs creation, enaction and becoming. — Joshs

Yes, I agree it’s probably the underlying axis. For my part I would tend to side more with . I wouldn't want to deny creation, enaction, or becoming, but my worry is that if we say “normativity is creatively re-established in each use,” we risk collapsing into “norms are whatever we now make them,” which would seem to undercut the possibility of error and the authority of correction.

I'm changing my paper from a conservative defense of JTB into something more innovative. Instead of "I'm clarifying what was always implicit," it now says "I'm developing the guardrails framework, three criteria for evaluating epistemology." I've also changed the title and cut the paper by 20%. The following is the new opening:

Knowledge as Practice-Standing: The Guardrails Framework
By Samuel L. Naccarato

Abstract
Sixty years after Gettier, epistemology remains stuck. The "JTB + X" industry keeps proposing new conditions, but the counterexamples keep coming. I argue the problem is not that justified true belief needs additional conditions, but that we've misunderstood what justification is.

Justification is standing within a practice, not merely having supporting reasons. Drawing on Wittgenstein's later philosophy, I develop a framework that makes this explicit. Three guardrails discipline epistemic standing (no false grounds, practice safety, and defeater screening), and five routes describe how justification proceeds (testimony, inference, perception, linguistic training, and logic's boundary-setting role). Standing requires conceptual competence and rests on bedrock certainties that make justification possible without themselves being justified.

This framework dissolves Gettier cases, which mistake the appearance of support for genuine standing. It also explains why artificial systems lack knowledge. They produce true statements but lack practice-standing. The result preserves JTB's core insight while articulating its grammatical structure.

Introduction
The classical account of knowledge as justified true belief captured something essential. To know is to hold a true belief that meets the criteria our practices of justification require. But something has gone wrong. Fifty years of post-Gettier epistemology has produced an industry of "JTB + X" proposals (adding causal connections, reliability conditions, defeasibility clauses), each attempting to patch the model against counterexamples. The proliferation of patches suggests we've been looking for solutions in the wrong place.

Gettier cases are not counterexamples to justified true belief. They reveal a confusion about what justification is. When we treat justification as merely having supporting reasons, we mistake the appearance of support for genuine justification. The "ten coins" case fails not because JTB is incomplete, but because the belief doesn't satisfy our epistemic criteria. It rests on false grounds, succeeds only by luck, and collapses under scrutiny. These are not missing conditions we need to add. They describe what justification requires.

Justification is standing within a practice, a status conferred when a belief meets the criteria that govern knowledge-attribution in a language-game. This status requires conceptual competence (knowing how to use concepts within a form of life, recognizing what supports what, and responding to challenges). Understanding is not added to justification from outside. It is internal to justification itself.

Moreover, justification operates against bedrock certainties that stand fast without themselves being justified. These Wittgensteinian hinges are not items of knowledge. They make knowledge possible. Doubt presupposes something not in doubt. To question everything is not to extend inquiry but to lose the standpoint from which inquiry proceeds.

I develop three guardrails that discipline epistemic assessment. No False Grounds means support cannot rest on falsehoods that undermine the inference. Practice Safety means the belief must not be true merely by luck. Defeater Screening means the belief must survive relevant challenges. I also distinguish five routes (testimony, inference, perception, linguistic training, and logic's boundary-setting role).

This framework dissolves Gettier cases and explains why AI systems lack knowledge. It provides clear criteria for epistemic assessment in an age of artificial intelligence and information overload. The result is not a new theory but the articulation of a grammar at work in our practices.

[I'm trying to make the paper shorter and more concise.]

I've only just discovered this thread, and I don't suppose for a moment that you'll want to revive it. On the other hand, I can't resist making some comments.

Gettier constructs his examples on the basis of the observation that one can be justified in believing something that turns out to be false and that if one is justified in believing P and P implies Q, then one is justified in believing Q. These two claims underlie all the examples. To bar these cases, we need to undermine those claims. No false grounds will do it, but at the cost of requiring that knowledge that is not backed by conclusive evidence is not knowledge. Which seems far too strict for ordinary use.

The requirement for conclusive evidence is also the obvious way to prevent success by luck in many, if not all, real life cases of knowledge. My guess is that a more specific requirement, close to practice safety, is required for this. My requirement would be that the belief is not simply based on evidence but on competence. So a single success is never sufficient to demonstrate knowledge, but a sustained record of success is. Hence, someone who consistently wins their bets will be credited with knowledge, , whether or not they can produce conclusive evidence based arguments. I think of this as a competence clause.

Defeater screening is also very plausible. But if this clause needs to be satisfied before knowledge is attributed, it will be too strict for ordinary use. Part of the point of the concept of defeaters is that it does not merely create a "prejudice" in favour of a belief but creates a warrant for them - a bit like a decree nisi in divorce proceedings. I can attribute knowledge on grounds that are less than conclusive. If I turn out to be wrong, then I must withdraw the claim. If I do not, then my attribution stands.

There's one factor that I don't think you have thought about, and that is the question of propositional ambiguity. "The man who will get the job..." in Gettier's example turns out to be ambiguous. It can be justified either by Smith or Jones. If we interpreted the claim as "The man who will get the job (viz. Jones)..." or "The man who will get the job (viz. Smith)....." there would be two clearly distinct propositions (beliefs) and no Gettier case. Many of these cases can be dismissed on these grounds, but I'm not sure that all of them can.

The second example in Gettier's original article is one such case. It has a different structure. It assigns two independent propositions to a single disjunction. One alternate is justifiably believed to be true, but turns out not to be. It then turns out that the second alternate is true. Again, the example depends on ambiguity about the proposition that is supposed to be known. The indeterminacy of the concept of a proposition is, imo, the source of the trouble. Sadly, I don't think there is a solution for that.

(All this is from memory, so I don't guarantee that it is accurate.)

PS Just to be clear, I really don't mean to be unhelpful. I don't accept Gettier cases, but, like everybody else, I find it very difficult to prevent them. I suspect it may not be possible to prevent all possible cases that anyone might ever come up with.

Futhermore, for the record, I think that Vogel's paradox of knowledge is a much more interesting problem, even though it has not attracted anything like the same volume of comment. I suspect that this is mainly because it does not have a solution.

This problem was devised by Jonathan Vogel.

1. Someone (call him Al) has parked his car on Avenue A (out of sight now) half an hour ago. Everything is normal, the car is still there, Al has a good memory. Does he know where his car is?

2. Every day, a certain percentage of cars gets stolen. Does Al know, right now, that his car has not been stolen and driven away since he parked it?

3. Meanwhile, in a parallel universe with a similar crime rate, Betty has parked her car on Avenue B half an hour ago. Betty is cognitively very similar to Al (just as good a memory, just as much confidence about the location of her car). Her car, unfortunately, was stolen and driven away. Does Betty, who believes that her car is on Avenue B where she parked it, know that her car is on Avenue B?

I'll dig out some references if you want them.