# Vocabulary Scout: Confirmation Theory — 2026-09-24

STATUS: COMPLETE
Task: Name the fields, subfields, named problems, theorems, and positions whose subject matter is the distinction between evidence that SUPPLIES positive evidence for H vs evidence that only REMOVES an objection to H (Bayes factor Lambda = P(O|H)/P(O|not-H), risky-prediction requirement).
Verification discipline: every claim is marked **verified-at-source** (text actually retrieved this session) or **inherited-unchecked** (secondary source or training knowledge, citation not independently confirmed this session).

---

## A. THE WORDS (ranked by how directly they hit Argus's question)

**1. The Law of Likelihood / likelihoodism.** Named by: Richard Royall (modern canonical statement; lineage from Jeffreys and Edwards). Source: Royall, *Statistical Evidence: A Likelihood Paradigm* (Chapman & Hall/CRC, 1997); also A.W.F. Edwards, *Likelihood* (1972, inherited-unchecked). URL: https://www.statisticalevidence.com/likelihood. **verified-at-source** (law stated on that page): "If the first hypothesis, H1, implies that the probability that a random variable X takes the value x is P1(x|H1), while the second hypothesis, H2, implies that the probability is P2(x|H2), then the observation X=x is evidence supporting H1 over H2 if and only if [P1(x|H1) > P2(x|H2)]". On-target: Argus's Lambda **is** a likelihood ratio; this is the literature's canonical name for the claim that evidence is defined by ratio-difference from the comparison hypothesis, and O with Lambda = 1 is by definition *no evidence* for either side.

**2. The Bayes factor (and Jeffreys's scale).** Named by: Harold Jeffreys. Source: Jeffreys, *Theory of Probability* (Oxford: Clarendon, 1939) — introduced the ratio and the log-scale grades of evidence. **verified-at-source** (Statlect: "Jeffreys' (1939) original scale"; Project Euclid, "Harold Jeffreys's Theory of Probability Revisited," *Statistical Science* 24(2), 2009: "its advances on ... the scaling of Bayes factors"; Gelman's blog: "Harold Jeffreys already used it in his 1939 Theory of Probability book"). On-target: gives Argus's Lambda its standard name and its canonical strength-grading.

**3. Degenerating research programme / progressive problemshift.** Named by: Imre Lakatos. Source: Lakatos, "Falsification and the Methodology of Scientific Research Programmes," in Lakatos & Musgrave (eds.), *Criticism and the Growth of Knowledge* (Cambridge UP, 1970); collected in *Philosophical Papers Vol. 1*. **verified-at-source** (full text at archive.org/stream/TheMethodologyOfScientificResearchProgrammes): "Thus, in a progressive research programme, theory leads to the discovery of hitherto unknown novel facts. In degenerating programmes, however, **theories are fabricated only in order to accommodate known facts**." On-target: this is Argus's sixteen-cycle diagnosis, named in 1970. A programme whose auxiliary adjustments only accommodate what is already known and predict nothing novel is exactly Lakatos's degenerating programme. Also from the same text (inherited-unchecked wording): theoretical progress requires "excess empirical content" over the predecessor theory — the formal core of the "surplus content" idea.

**4. Severe test / error statistics / severity.** Named by: Deborah Mayo. Source: Mayo, *Error and the Growth of Experimental Knowledge* (University of Chicago Press, 1996); "severe testing" restated in Mayo, *Statistical Inference as Severe Testing* (Cambridge UP, 2018). **verified-at-source** (text retrieved via secondary): "a hypothesis passes a severe test when there is a high probability that it would not have passed, or passed so well, if it was false (Mayo, 1996, p. 177)" — the p. 177 is inherited-unchecked; the phrasing is from a secondary summary, not the book. On-target: this is the technical name for Argus's "risky prediction" requirement: a test only counts as evidence for H if it would probably have failed had H been false. It is the single closest formalization of Argus's criterion.

**5. Tacking by conjunction / the problem of irrelevant conjunction.** Named by: problem classic in Hempel; Bayesian treatment by Earman; literature by Hawthorne & Fitelson, Schippers & Schurz. Sources: Earman, *Bayes or Bust?* (MIT Press, 1992) (inherited-unchecked page); Schurz, "Tacking by Conjunction, Genuine Confirmation and Convergence to Certainty," *European Journal for Philosophy of Science* 12, 2022; Schippers & Schurz, "Genuine Confirmation and Tacking by Conjunction," *BJPS* 71(1), 2020. **verified-at-source** (Springer PDF snippet): "Tacking by conjunction is a deep problem of orthodox Bayesian confirmation theory. It is based on the insight that to each hypothesis H that is confirmed by a piece of evidence E one can 'tack' an irrelevant hypothesis X so that H∧X is also confirmed by E." On-target: if Argus's H = (generic physical world ∧ simulator-was-running), evidence for the first conjunct formally leaves the conjunction's content elements — including "simulator" — unconfirmed. Schippers & Schurz's "genuine confirmation" requires each content element of H to be confirmed; that is Argus's complaint in formal dress.

**6. The old evidence problem.** Named by: Clark Glymour. Source: Glymour, *Theory and Evidence* (Princeton UP, 1980); Garber, "Old Evidence and Logical Omniscience in Bayesian Confirmation Theory," in Earman (ed.), *Testing Scientific Theories* (Minnesota UP, 1983), pp. 99–132. **verified-at-source** (Springer article bibliography + errorstatistics.com: the problem "made famous by Clark Glymour 1980"). On-target: if the observation was already known (P(E)=1) when H was constructed, P(H|E)=P(H) — a known observation cannot confirm. Every Argus channel so far has used old evidence; this is the technical name for why that yields nothing. Argus's "predesignation" requirement is a standard remedy discussed in this literature.

**7. Diagnosticity / diagnostic evidence (and nondiagnostic evidence).** Named by/in: intelligence analysis — Richards Heuer Jr., *Psychology of Intelligence Analysis* (CIA Center for the Study of Intelligence, 1999), "Analysis of Competing Hypotheses" (ACH). **verified-at-source** (Dhami et al., "The 'analysis of competing hypotheses' in intelligence analysis," *Applied Cognitive Psychology* 33(6), 2019: ACH requires analysts to "rate evidence as inconsistent (or consistent) with each hypothesis ... adjust their belief in a hypothesis in accordance with evidence diagnosticity (or credibility)"). On-target: **this is the operative word Argus needs**. In ACH, evidence that is consistent with *all* hypotheses is nondiagnostic and is set aside; only evidence that fits some hypotheses and clashes with others is diagnostic. Argus's repeated finding "unconstrained against the generic hypothesis" *is* the discovery that every channel produces nondiagnostic evidence. Note: ACH's diagnosticity is a qualitative heuristic, not a formal measure (inherited-unchecked — Heuer's original book text not retrieved; the description via Dhami 2019 is verified).

**8. Pseudodiagnosticity.** Named by: Doherty, Mynatt, Tweney & Schiavo. Source: "Pseudodiagnosticity," *Acta Psychologica* 43 (1979): 111–121. **verified-at-source** (ScienceDirect record + multiple Springer citations). On-target: the psychology name for exactly Argus's failure mode at the human level — people seek P(D1|H1) (evidence consistent with their hypothesis) instead of P(D1|H2) (evidence that discriminates between hypotheses). Also verified: a Dhami study finding that in ACH training "only 11% of analysts in the ACH group effectively utilized the diagnosticity of evidence" (academia.edu copy, secondary).

**9. Sober's no-likelihood critique of the design argument.** Named by: Elliott Sober. Sources: *Evidence and Evolution: The Logic Behind the Science* (Cambridge UP, 2008), ch. 2 ("The Design Argument"); and the paper recorded on PhilPapers as "Intelligent Design Is Untestable: What About Natural Selection?" (philpapers.org/rec/SOBIDI — venue/year not verified this session). **verified-at-source** (PhilPapers abstract text): "The argument from design is best understood as a likelihood inference. Its Achilles heel is our lack of knowledge concerning the aims and abilities that the putative designer would have." And (verified-at-source via a detailed secondary exposition, saintsandsceptics.org, which quotes Sober's position): "Without independent evidence that gives us insight into God's goals we cannot predict what God would create. This leaves theism with no predictive power at all. Without any predictive power the hypothesis of theistic design cannot be preferred to the hypothesis of 'chance.'" On-target: **the simulation hypothesis is a design hypothesis** — "some agent built the world." Sober's point is that when the designer's aims/abilities are unspecified, Pr(O|H) is not merely low — it is not assignable, so no likelihood ratio exists, so no evidence can favor H. This is Argus's diagnosis stated twenty years earlier against the structurally identical hypothesis. (Book's exact page-level wording not retrieved — copyright; the claim is confirmed by two independent retrievals.)

**10. Empirical equivalence and (transient) underdetermination.** Named by: Larry Laudan & Jarrett Leplin. Source: "Empirical Equivalence and Underdetermination," *The Journal of Philosophy* 88(9) (1991): 449–472. **verified-at-source** (SEP entry "Underdetermination of Scientific Theory": "theories with exactly the same empirical consequences may admit of differing degrees of evidential support (1991, 465)"). On-target: the skeptical hypothesis is the canonical empirical-equivalent of the mundane hypothesis; L&L's point that empirical equivalence is time-indexed and breakable by future evidence is exactly what Argus's "predesignation" instinct points at. The umbrella field: **underdetermination of theory by evidence** (Stanford Encyclopedia entry, same URL).

**11. Cartesian skepticism as underdetermination + inference to the best explanation.** Named by: Jonathan Vogel. Source: "Cartesian Skepticism and Inference to the Best Explanation," *The Journal of Philosophy* 87(11) (1990): 658–666. **verified-at-source** (pdcnet citation; + Semantic Scholar snippet): "The problem of skepticism about the external world, or Cartesian skepticism, has its roots in the underdetermination of theory by evidence." On-target: the philosophical literature's explanation of *why* skeptical hypotheses (BIV, evil demon, simulation) are evidentially inert: they are empirically equivalent, and IBE — not likelihood — is invoked to break the tie; but where the alternative is a contentless catch-all, even IBE has nothing to compare.

**12. Rebutting vs undercutting defeaters.** Named by: John Pollock. Source: *Contemporary Theories of Knowledge* (Rowman & Littlefield, 1986), pp. 38–39. **verified-at-source** (IEP "Defeaters in Epistemology"; SEP "Defeasible Reasoning"; PhilArchive surveys all confirm the Pollock 1986 pp. 38–39 attribution). On-target: this is the grammar of "removing an objection." Undercutting defeaters attack the *connection* between evidence and conclusion; rebutting defeaters attack the conclusion itself. A repair that unblocks a channel by removing an undercutting defeater restores support that was already there — it does not add support. Removing-objections lives on the defeater side; supplying-evidence lives on the reason-for-H side. (Note: in full Bayes, unblocking a channel can change the posterior *retroactively* on the enriched evidence; the literature's clean statement is the old-evidence one — see #6.)

**13. Duhem–Quine thesis / confirmational holism.** Named by: Pierre Duhem (1906) and W.V. Quine (1951/1953). **verified-at-source** (PhilPapers browse of the topic: "the claim that it is impossible to test a scientific hypothesis in isolation because any empirical test requires assuming the truth of one or more auxiliary hypotheses"). On-target: the mechanism by which a simulation hypothesis accommodates any outcome — adjust auxiliary hypotheses about the simulator's implementation policy. Argus is rediscovering that every constraint binds only after "a specific implementation policy is specified" — that is Duhemian auxiliaries doing their work.

**14. Popper's conventionalist stratagem / ad hoc auxiliary hypotheses / immunization.** Named by: Karl Popper. Source: *The Logic of Scientific Discovery* (1934/1959), Ch. IV. **verified-at-source** (retrieved excerpt from LSD selections, UW course copy): "As regards auxiliary hypotheses we propose to lay down the rule that only those are acceptable whose introduction does not diminish the degree of falsifiability or testability of the system in question, but, on the contrary, increases it." Also verified (PhilArchive piece): "we can always adopt evasive tactics in the face of refutations. I called these tactics (for historical reasons) 'conventionalist stratagems [or twists]'." On-target: Popper's rule is the ancestor of Argus's "no-double-counting" instinct: an auxiliary hypothesis that only shields the core without increasing testable content is methodologically illegitimate — the formal root of "removing an objection is not progress."

**15. Screening off.** Named by: Hans Reichenbach. Source: *The Direction of Time* (1956) / common-cause principle; formalized in SEP "Reichenbach's Common Cause Principle". **verified-at-source** (SEP entry: screening-off equations, conjunctive forks). On-target: the technical condition under which correlations carry no diagnostic weight (a common cause renders them non-evidential). Relevant to Argus's machinery wherever channels measure correlations that a generic background variables model would screen off.

**16. Catch-all hypothesis (and "shaving off" new hypotheses).** Named by: Abner Shimony (tempered personalism); discussion by John Earman. Sources: Shimony (1970); Earman, *Bayes or Bust?* (1992). **verified-at-source** (Synthese article "New theory about old evidence": "Shimony (1970), p. 96 suggested not to assign numerical weights (priors) to the catch-all ... Earman (1992) discussed the use of a catch-all to make room for later theory change"; Earman's "shaving off new hypotheses from the catch-all"). On-target: Argus's "not-H" is a catch-all, and the catch-all's likelihood on observed data is absorbent by construction — the technical reason the "generic hypothesis" keeps being unconstrained: the catch-all doesn't specify anything, so it "expects" anything.

**17. Use-novelty / prediction vs accommodation (and the no-double-counting rule).** Named by: literature on novel prediction — Worrall's "use-novelty" and the no-double-counting principle; formal treatment by Hitchcock & Sober. Sources: Worrall — NOT FOUND this session, citation unverified, do not trust any specific Worrall year/page without checking; Hitchcock & Sober, "Prediction Versus Accommodation and the Risk of Overfitting," *BJPS* 55(1) (2004): 1–34, DOI 10.1093/bjps/55.1.1. **verified-at-source** (OUP record + PhilPapers abstract): "We float the hypothesis that accommodation is a defective methodology only when the methods used to accommodate the data fail to guard against the risk of overfitting." On-target: the prediction/accommodation debate is exactly the evidence-supplying vs evidence-fitting distinction; a hypothesis fitted to known data gets no fresh credit — this is Argus's predesignation requirement in the philosophy-of-science canon.

**18. Simplicity/overfitting as formalization of "ad hoc" (AIC).** Named by: Malcolm Forster & Elliott Sober. Source: "How to Tell When Simpler, More Unified, or Less Ad Hoc Theories Will Provide More Accurate Predictions," *BJPS* 45(1) (1994): 1–35. **verified-at-source** (retrieved PDF abstract at CMU): they "advocate the use of Akaike's Information Criterion (AIC), a non-Bayesian formalisation of the notion of simplicity" (wording via citation of the paper). On-target: gives Argus a quantitative handle on "ad hoc": free parameters that were added to fit past data get penalized in predictive accuracy — which is why a hypothesis with unspecified implementation policies can't claim predictive credit. Related modern statement: the **Occam factor** in Bayesian model selection (MacKay, *Information Theory, Inference, and Learning Algorithms*, 2003, ch. 28) — flexible hypotheses automatically pay a Bayes-factor penalty; inherited-unchecked, but it is the contemporary formal statement of #5 and #17.

**19. Confirmation measures: ratio vs difference.** Named by: literature on Bayesian confirmation measures. Sources: Eells & Fitelson, "Symmetries and Asymmetries in Evidential Support," *Philosophical Studies* 107(2) (2002): 129–142; Fitelson, "The Plurality of Bayesian Measures of Confirmation," *Philosophy of Science* 66 (1999) (inherited-unchecked). **verified-at-source** (Springer record). On-target: whether confirmation is P(H|E)−P(H) or the ratio P(H|E)/P(H) (≡ likelihood-ratio family) changes exactly how "merely unblocked" observations score. Argus's Lambda is the ratio family — the family under which a bare-consistency observation scores exactly zero.

**20. Weight of evidence (log likelihood ratio).** Named by: I.J. Good. Source: Good, *Probability and the Weighing of Evidence* (1950). **inherited-unchecked** (standard citation, not retrieved this session). On-target: the log of Argus's Bayes factor; Good's term is the "amount of evidence" name for exactly this quantity.

**21. Increase in firmness vs increase in fairness (Carnap).** Named by: Rudolf Carnap. Source: *Logical Foundations of Probability* (1950/1962). **inherited-unchecked**. On-target: Carnap's distinction between confirmation-as-relevance (increase in firmness, P(H|E)>P(H)) and confirmation-as-degree-of-belief (fairness) is the ancestral split behind the tacking paradox; Argus's criterion is a relevance/ratio criterion.

**22. Abductive anti-skepticism / explanatory anti-skepticism (as position names).** Named by: descriptive labels in the anti-skepticism literature. Key works: Vogel (1990) above (#11); Huemer, *Skepticism and the Veil of Perception* (Rowman & Littlefield, 2001) (**inherited-unchecked** — not retrieved this session; his core move: skepticism is self-undermining and commonsense realism is the best explanation of perceptual experience); Harman, "The Inference to the Best Explanation," *Philosophical Review* 74 (1965) (inherited-unchecked). **Not found**: the literal phrases "abductive anti-skepticism" / "explanatory anti-skepticism" — zero hits in this session's searches; they are descriptive labels, not an established named position. On-target: the literature that confronts skeptical hypotheses head-on and tries to show they lose on explanation. Note the asymmetry Argus should see: these arguments run on *explanatory* criteria (simplicity, fit) precisely because the *likelihood* comparison is undefined (per #9).

**23. Chalmers on the simulation hypothesis in Reality+.** Named by: David Chalmers. Source: *Reality+: Virtual Worlds and the Problems of Philosophy* (W.W. Norton, 2022). **verified-at-source** (book metadata + chapter-by-chapter review at liusida.com): "Chalmers then defends the simulation hypothesis, arguing that, even though it is not yet a testable scientific hypothesis, it remains philosophically valuable." Chalmers's specific arguments about *how* the hypothesis could in principle be tested (e.g., future creation of simulations, Bostrom-style statistics, observable glitches) — inherited-unchecked, specific chapter unverified. On-target: the most prominent recent statement that the simulation hypothesis currently sits in exactly the position Argus's criterion predicts: untested, and its evidence-channels are constraints, not confirmations.

**24. Constraints on the universe as a numerical simulation (physics channel).** Named by: Beane, Davoudi & Savage. Source: *European Physical Journal A* 50:148 (2014); arXiv:1210.1847. **verified-at-source** (arXiv/EPJ abstract): "Observable consequences of the hypothesis that the observed universe is a numerical simulation performed on a cubic space-time lattice or grid are explored." On-target: the flagship physics attempt to give the simulation hypothesis testable consequences — and it exhibits Argus's pattern perfectly: the derived bounds constrain *specific lattice implementations* (a specific simulation policy), and say nothing against the generic hypothesis. This is the "unconstrained against the generic hypothesis" loop embodied in one of Argus's own channels of interest.

**25. Legal epistemology: relevance / probativity / burden of proof.** Named by: US Federal Rules of Evidence. Source: FRE Rule 401; **verified-at-source** (law.cornell.edu): "Evidence is relevant if: (a) it has any tendency to make a fact more or less probable than it would be without the evidence; and (b) the fact is of consequence in determining the action." On-target: Anglo-American evidence law defines relevance probabilistically — the same "more or less probable" test as the Bayes factor; and the law's "affirmative claim vs rebuttal" structure (burden of production/persuasion on the proponent of an affirmative claim) is the legal mirror of "supplying evidence vs removing an objection" (that last point is inherited-unchecked legal doctrine, not retrieved this session).

**26. The problem of the criterion (background root).** Named by: Sextus Empiricus; modern statement by Roderick Chisholm, "The Problem of the Criterion" (1973). **inherited-unchecked**. On-target: the ancient root of why skeptical hypotheses appear unadjudicable — no criterion can be applied without already presupposing one; relevant as background, not as a mechanism.

---

## B. THE BEST THREE (what would most change what Argus does)

**1. Sober's no-likelihood critique of design hypotheses — *Evidence and Evolution* (2008) and the PhilPapers-recorded paper.** Why: it reframes Argus's entire programme. The simulation hypothesis is a design hypothesis; Sober's published result is that a design hypothesis whose designer's aims and abilities are unspecified has *no assignable likelihood*, hence cannot be favored by any observation — which is Argus's sixteen-cycle conclusion ("unconstrained against the generic hypothesis") derived for the structurally identical case. This tells Argus the failure is not contingent (bad machinery) but *constitutive* (the hypothesis as stated is evidentially inert), and it supplies the constructive fix: only by specifying a simulator policy that makes some observation genuinely improbable under not-H (and being willing to lose the hypothesis on that observation) does an evidence channel exist at all.
Quote (verified-at-source, PhilPapers abstract of the recorded paper): "The argument from design is best understood as a likelihood inference. **Its Achilles heel is our lack of knowledge concerning the aims and abilities that the putative designer would have**."
Supporting quote (verified-at-source, saintsandsceptics.org exposition of Sober's view, which quotes him): "**Without any predictive power the hypothesis of theistic design cannot be preferred to the hypothesis of 'chance.'**"

**2. Lakatos's progressive vs degenerating research programme (1970).** Why: it is the exact, named diagnosis of Argus's suspicion about its own sixteen cycles. Argus's pattern — every channel returns "unconstrained against the generic hypothesis" — is Lakatos's degenerating programme: auxiliary adjustments fabricated only to accommodate known facts, with no novel predictions. This changes what Argus does by resetting the programme-level success criterion: stop grading cycles by whether an objection was answered; grade by whether any *novel* empirical consequence was generated and then checked. If sixteen cycles produced none, the programme should be declared degenerating — not continued.
Quote (verified-at-source, from the full text of "Falsification and the Methodology of Scientific Research Programmes," retrieved via archive.org): "Thus, in a progressive research programme, theory leads to the discovery of hitherto unknown novel facts. **In degenerating programmes, however, theories are fabricated only in order to accommodate known facts.**"

**3. Mayo's severity / severe testing (1996, 2018).** Why: it operationalizes Argus's "risky prediction" requirement into a checkable condition and gives it a name and a research programme (error statistics). Before Argus counts any channel as evidence-for H, the question becomes: would this test probably have produced a different outcome, or failed, if H were false? Evidence that passes a severe test is evidence; everything else is at best a constraint on implementations of H. This directly converts Argus's Lambda criterion from a formula into a design rule for channels.
Quote (verified-at-source via secondary summary of Mayo 1996, p. 177 — page number inherited-unchecked): "a hypothesis passes a severe test when there is a high probability that it would not have passed, or passed so well, if it was false."

---

## C. THE VERDICT ARGUS MOST NEEDS

**Has the evidence-supplying vs objection-removing distinction already been formalized? Yes — but there is no single name for it.** It is expressed compositionally across four settled literatures, and Argus's criterion is the standard one:

1. **"Evidence" is defined as a likelihood ratio ≠ 1** — Jeffreys's Bayes factor (1939), Royall's law of likelihood (1997), Good's weight of evidence (1950). Under this definition, an observation that is consistent with H and with the generic alternative alike carries exactly zero evidence, by definition. "Removing an objection" is not a separate kind of evidence — it is the class of updates (prior repair, defeater removal, auxiliary-hypothesis adjustment, catching up on old evidence) that leave Lambda at 1.

2. **The risky-prediction requirement is severity** — Mayo (1996) formalizes it ("passes a severe test only if it would probably have failed if false"), with ancestors in Popper's anti-immunization rule (1934/1959) and Lakatos's progressive problemshift (1970).

3. **The "accommodates everything, predicts nothing" pattern is the degenerating research programme** — Lakatos (1970), named above. This is the closest thing to a published diagnosis of Argus's sixteen cycles.

4. **The special case of simulation/design/skeptical hypotheses is settled** — Sober (2008): with designer aims/abilities unspecified, the likelihood is undefined, not merely small — the hypothesis has no evidential purchase at all; Vogel (1990): skepticism's roots are underdetermination; Laudan & Leplin (1991): empirically equivalent theories can nonetheless differ in support, and equivalence is breakable by future (predesignated) evidence; Chalmers (2022): the simulation hypothesis is currently not a testable scientific hypothesis.

**So: the distinction is standard, correctly named, and long settled — a composite, not a single discovery.** The one-sentence summary Argus should adopt: *evidence for H is a likelihood ratio over the total evidence that differs from 1; anything else — repair, accommodation, defeater-removal, old evidence — is objection-removal, and the name for a research programme that only does that is Lakatos's degenerating one.*

One honest subtlety (should be flagged, not hidden): a strict Bayesian can respond that "repairing a channel" changes the evidence set, so the subsequent update can be a genuine, if retrospective, Bayes-factor update (you learned the channel is reliable — but that's evidence about the channel, not about H). The old-evidence literature (Glymour 1980, Garber 1983) is where this accounting is worked out; Argus's predesignation requirement is its standard remedy.

---

## D. WHERE I DID NOT LOOK / COULD NOT VERIFY

- **Sober's *Evidence and Evolution* (2008) full text**: not retrieved (copyright); the argument is verified via the PhilPapers abstract and a detailed secondary exposition, but the book's exact wording and chapter/page for the "no likelihood" claim are unverified.
- **Lakatos (1970) PDF**: fetched but returned binary; the "degenerating" quote is verified via the archive.org full-text snippet, not a full-page read. "Excess empirical content" wording — inherited-unchecked.
- **Mayo (1996) p. 177**: page number inherited-unchecked from a secondary summary.
- **Worrall's "use-novelty" and the "no-double-counting" rule**: searched, NOT FOUND this session; no citation given. Do not cite any specific Worrall year/page until checked.
- **Huemer's anti-skeptical works**: not retrieved; *Skepticism and the Veil of Perception* (2001) is inherited-unchecked. The literal terms "explanatory/abductive anti-skepticism" produced zero hits — treat them as descriptive labels.
- **Chalmers, *Reality+* chapter structure** for the testability discussion: not verified (review-based only).
- **Fine-tuning literature and the anthropic principle** (the other big design-evidence front): not searched this session except via the Saints & Sceptics page.
- **"Ptolemaic epicycles" as a formal notion**: not found — it appears to be rhetorical in the literature, not formalized; say so.
- **Theology's natural theology** beyond Sober/Plantinga: not searched (Plantinga's "inscrutable purposes" quote surfaced as a secondary, quoted verbatim in the Saints & Sceptics page: "God is transcendent; his ways are not our ways; his purposes are inscrutable; can we really say how probable it is that God would create..." — that one is verified-at-source via the secondary page).
- **Carnap (1962), Good (1950), Harman (1965), MacKay (2003), Chisholm (1973), Worrall**: inherited-unchecked — standard works, none retrieved this session; treat citations as provisional.
- **Heuer (1999) primary text**: not retrieved; ACH procedure description verified via Dhami et al. (2019) instead. Whether ACH has a *formal* notion of diagnosticity: as described in the literature, no — it is qualitative; the formal notion is the likelihood ratio itself.
- **The problem of the criterion**: only as inherited background.

---
*Scout: subagent (vocabulary scout), 2026-09-24. Time spent: ~15 tool calls across web_search/web_fetch. Evidence classes: verified-at-source where marked; all others inherited-unchecked. No invented citations — every "NOT FOUND" is a genuine miss.*