The constraints that make philosophy good can be specified as properties of the text itself, not as properties of the author's mental states or post-publication behaviour. This makes "the paper stands on its own" not just a slogan but a methodological thesis with teeth. ## The constraint profile A philosophical artefact is evaluated by whether its text exhibits: 1. **Precision** — explicit scope conditions, clear commitments, no elastic definitions 2. **Cost-accounting** — for each explanatory gain, the costs are named (ontological commitments, revisions to background assumptions, counterintuitive consequences) 3. **Non-ad hocness** — repairs are motivated by independently plausible principles, not introduced solely to block objections 4. **Defeater-sensitivity** — the paper specifies what would undermine the view (conceptual counterexamples, clashes with constraints, empirical findings) 5. **Fair treatment of rivals** — alternatives are presented charitably, not straw-manned These correspond to [[Timothy Williamson]]'s picture of abductive methodology: bolder theories, explanatory potential, simplicity/elegance as part of rational discrimination under noisy evidence. ## Why this matters for LLM evaluation If philosophical goodness is fixed by text-internal properties, then any evaluation of LLM outputs must also be text-internal. The question becomes: does this text exhibit the constraint profile? — not: does the producer have the "right" inner states? This blocks the move from "LLMs are stochastic" to "LLMs can't produce good philosophy." The production mechanism is irrelevant to whether the produced text satisfies the constraints. [[Floridi's critique of LLM abduction|Floridi's "abductive appearance"]] has to be cashed out by pointing to specific text-internal failures—equivocation, illicit premises, ad hoc repairs—not by gesturing at the underlying mechanism. ## Relation to provenance irrelevance If constraints are text-internal, then provenance (who wrote it) cannot rationally affect whether an argument is good. A borderline step remains borderline regardless of authorship. Any resistance to LLM philosophy based on authorship is a refusal to do philosophy, not a philosophical objection. See [[Provenance is not the right kind of variable in philosophical evaluation]]. ## The Williamson connection [[Timothy Williamson]]'s discussion of over-fitting provides principled grounding here. His point: simplicity is a counterbalance to evidential fit because it reduces vulnerability to error ("mistaking noise for signal"). A paper that commits to the simplest view satisfying its explanatory target, and explicitly rejects complexity-adding repairs unless they bring compensating gain, is exhibiting the robustness strategy Williamson defends as methodologically respectable. --- Source: [[Web Clippings/ChatGPT 1|ChatGPT conversation on Floridi, Williamson, abductive competence]]