h t t p s : / / w w w . r o u t l e d g e . c o m / T h e - L o g i c - o f - S c i e n t i f i c - D i s c o v e r y / P o p p e r / p / b o o k / 9 7 8 0 4 1 5 2 7 8 4 4 7

Ingest: The Logic of Scientific Discovery

Type: types/ingest-report.md

Classification

A book-length philosophical treatise on the method of the empirical sciences. It argues by logical analysis and examples from physics, not by empirical comparison of methods. The captured text is the Routledge Classics 2002 edition of the 1959 English translation. It keeps the 1934 main text and adds starred footnotes, new appendices i–xii, and 1972 addenda in which Popper corrects, qualifies, or extends the original. Several of those corrections bear on the content used here, so this analysis names the layer (1934 text, starred note, or appendix) where it matters. Author: Karl Popper, stating his own account first-hand. This is the primary source for doctrines the KB already attributes to him.

Summary

Popper replaces verification with falsifiability as the mark of empirical science. A theory is a strictly universal statement: it forbids certain singular states of affairs, and its empirical content is the class of basic statements it forbids (§§15, 21, 31–35). Singular statements and "numerically universal" statements about a finite individual region are a different logical kind, and no process of abstraction or induction leads from them to universals (§§13–14). An explanation deduces a singular prediction from universal laws plus singular initial conditions (§12). Objectivity means inter-subjective testability, so an effect counts only when it is reproducible under stated instructions (§8), and a theory counts as falsified only when a corroborated low-level "falsifying hypothesis" describes a reproducible effect that contradicts it (§22). The basic statements that end any test are accepted by a rule-governed decision, not justified by experience; they stay open to further testing (§§27–30). Since the logic of refutation can always be evaded by ad hoc auxiliary hypotheses, redefinition, or doubting the observer, the empirical method is fixed by methodological rules that forbid these "conventionalist stratagems" (§§19–20). Higher universality or precision means more content and better testability, but when one statement is more universal and another more precise, the two are not comparable (§36). Corroboration is not probability. It is a dated appraisal of how a theory has survived tests relative to the basic statements accepted so far, and it depends on test severity rather than on the number of instances (§§79–85). Appendix *ix adds that evidence counts toward corroboration only when it reports sincere attempts to refute the theory.

Quotes

  • Source extract (verbatim): We say that a theory is falsified only if we have accepted basic statements which contradict it ( cf. section 11, rule 2). This condition is necessary, but not sufficient; for we have seen that non-reproducible single occurrences are of no significance to science. Thus a few stray basic statements contradicting a theory will hardly induce us to reject it as falsified. We shall take it as falsified only if we discover a reproducible effect which refutes the theory. In other words, we only accept the falsification if a low-level empirical hypothesis which describes such an effect is proposed and corroborated. This kind of hypothesis may be called a falsifying hypothesis
  • Source location: Part II, Chapter 4, section 22 "Falsifiability and Falsification", second paragraph
  • Source extract (verbatim): Secondly, if instead of omitting this restriction we restrict the class of the derived accepted basic statements further, by demanding that they should be accepted as the results of sincere attempts to refute the theory, then our definition becomes an adequate definition of ‘positively corroborated’, though not, of course, of ‘degree of corroboration’.
  • Source location: Chapter 10, section 82 "The Positive Theory of Corroboration", starred footnote *1, second paragraph
  • Source extract (verbatim): In opposition to this inductivist attitude, I assert that C ( h , e ) must not be interpreted as the degree of corroboration of h by e , unless e reports the results of our sincere efforts to overthrow h . The requirement of sincerity cannot be formalized—no more than the inductivist requirement that e must represent our total observational knowledge.
  • Source location: Appendix ix "Corroboration, the Weight of Evidence, and Statistical Tests", Third Note, point 14, second paragraph
  • Source extract (verbatim): To a higher degree of universality or precision corresponds a greater (logical or) empirical content, and thus a higher degree of testability.
  • Source location: Chapter 6, section 36 "Levels of Universality and Degrees of Precision", third paragraph (after the diagram of statements p, q, r, s)
  • Source extract (verbatim): If of two statements both their universality and their precision are comparable, then the less universal or less precise is derivable from the more universal or more precise; unless, of course, the one is more universal and the other more precise (as in the case of q and r in my diagram).
  • Source location: Chapter 6, section 36 "Levels of Universality and Degrees of Precision", the derivability rule stated after the definitions of universality and precision for universal conditional statements
  • Source extract (verbatim): The falsifying mode of inference here referred to—the way in which the falsification of a conclusion entails the falsification of the system from which it is derived—is the modus tollens of classical logic.
  • Source location: Part II, Chapter 3, section 18 "Levels of Universality. The Modus Tollens", paragraph introducing the modus tollens
  • Source extract (verbatim): By means of this mode of inference we falsify the whole system (the theory as well as the initial conditions) which was required for the deduction of the statement p , i.e. of the falsified statement. Thus it cannot be asserted of any one statement of the system that it is, or is not, specifically upset by the falsification. Only if p is independent of some part of the system can we say that this part is not involved in the falsification.
  • Source location: Part II, Chapter 3, section 18 "Levels of Universality. The Modus Tollens", paragraph following the symbolized inference ((t → p).p¯) → t¯
  • Source extract (verbatim): The other kind, the numerically universal statements, are in fact equivalent to certain singular statements, or to conjunctions of singular statements, and they will be classed as singular statements here.
  • Source location: Part II, Chapter 3, section 13 "Strict and Numerical Universality", first paragraph
  • Source extract (verbatim): Statement (a) claims to be true for any place and any time. Statement (b) refers only to a finite class of specific elements within a finite individual (or particular) spatio-temporal region. Statements of this latter kind can, in principle, be replaced by a conjunction of singular statements; for given sufficient time, one can enumerate all the elements of the (finite) class concerned.
  • Source location: Part II, Chapter 3, section 13 "Strict and Numerical Universality", paragraph contrasting statement (a) on harmonic oscillators with statement (b) on living human beings
  • Source extract (verbatim): Rather we regard a statement of type (a) as an all-statement , i.e. a universal assertion about an unlimited number of individuals. So interpreted it clearly cannot be replaced by a conjunction of a finite number of singular statements.
  • Source location: Part II, Chapter 3, section 13 "Strict and Numerical Universality", end of the same paragraph
  • Source extract (verbatim): For the verification of a natural law could only be carried out by empirically ascertaining every single event to which the law might apply, and by finding that every such event actually conforms to the law—clearly an impossible task.
  • Source location: Part II, Chapter 3, section 13 "Strict and Numerical Universality", paragraph on views that obliterate the singular/universal distinction
  • Source extract (verbatim): ‘An individual concept is a concept in the definition of which proper names (or equivalent signs) are indispensable. If any reference to proper names can be completely eliminated, then the concept is a universal concept.’
  • Source location: Part II, Chapter 3, section 14 "Universal Concepts and Individual Concepts", paragraph offering definitions (before the 'pasteurized' footnote)
  • Source extract (verbatim): In the same way, any attempt to define universal names with the help of individual names is bound to fail. This fact has often been overlooked, and it is widely believed that it is possible to rise by a process called ‘abstraction’ from individual concepts to universal concepts. This view is a near relation of inductive logic, with its passage from singular statements to universal statements. Logically, these procedures are equally impracticable.3 It is true that one can obtain classes of individuals in this way, but these classes will still be individual concepts—concepts defined with the help of proper names.
  • Source location: Part II, Chapter 3, section 14 "Universal Concepts and Individual Concepts", paragraph beginning "In the same way, any attempt to define universal names"
  • Source extract (verbatim): It is for precisely the same reason that strictly universal statements are not verifiable. Again, we cannot search the whole world in order to make sure that nothing exists which the law forbids.
  • Source location: Part II, Chapter 3, section 15 "Strictly Universal and Existential Statements", paragraph on unilateral decidability
  • Source extract (verbatim): Here one can see very clearly the difference between truth and corroboration. The appraisal of a statement as corroborated or as not corroborated is also a logical appraisal and therefore also timeless; for it asserts that a certain logical relation holds between a theoretical system and some system of accepted basic statements. But we can never simply say of a statement that it is as such, or in itself, ‘corroborated’ (in the way in which we may say that it is ‘true’). We can only say that it is corroborated with respect to some system of basic statements —a system accepted up to a particular point in time. ‘The corroboration which a theory has received up to yesterday’ is logically not identical with ‘the corroboration which a theory has received up to today’. Thus we must attach a subscript, as it were, to every appraisal of corroboration—a subscript characterizing the system of basic statements to which the corroboration relates (for example, by the date of its acceptance).
  • Source location: Chapter 10, section 84 "Remarks Concerning the Use of the Concepts 'True' and 'Corroborated'", paragraph beginning "Here one can see very clearly the difference between truth and corroboration" (footnote marker *2 omitted at end)
  • Source extract (verbatim): Corroboration is therefore not a ‘truth value’; that is, it cannot be placed on a par with the concepts ‘true’ and ‘false’ (which are free from temporal subscripts); for to one and the same statement there may be any number of different corroboration values, of which indeed all can be ‘correct’ or ‘true’ at the same time. For they are values which are logically derivable from the theory and the various sets of basic statements accepted at various times.
  • Source location: Chapter 10, section 84 "Remarks Concerning the Use of the Concepts 'True' and 'Corroborated'", following paragraph beginning "Corroboration is therefore not a 'truth value'"

Connections Found

This book is the primary anchor for the KB's refutation and repair cluster. Until now, that cluster cited Popper through the tertiary Wikipedia ingest, through Conjectures and Refutations, or with no link. The Wikipedia ingest's Recommended Next Action asked for this text. Its two quotations (the verification/falsification asymmetry and the reproducible-effect rule) can now be checked in context. Section 22 read in context also answers that ingest's worry that the reproducible-effect rule licenses dismissing counterevidence. The same passage requires accepting a falsification once a falsifying hypothesis is corroborated, and §20 forbids excluding inter-subjectively testable experiments by doubting the experimenter.

For claim modality is the inference form of the refuter, the book is evidence for both unlinked attributions. The treatment of probability hypotheses as not falsifiable, yet usable as falsifiable through a methodological decision, is in §§65–68. The claim that conceiving a theory calls for no logical analysis is §2. Modus tollens against a whole system (§18), conditioned on a reproducible effect (§22), supplies the universal mode's refuter.

For narrowing bought to survive review is paid for in content and ad hoc explanation can be rational when error is cheap and local, §§19–20 is the original statement of the rescue moves. An auxiliary hypothesis is acceptable only if it increases the system's falsifiability. A changed definition makes the system one that "has to be re-examined as if it were new". Surreptitious changes of usage are forbidden. The auxiliary-hypothesis rule is stricter than the narrowing note's refuter test: the repair must add to what the system forbids, not merely keep some refuter. Domain pricing routes an exception to idealization assessment matches §20's concession that logical form alone cannot tell an empirical system from an immunized one. Only a decision about method can.

Generality bought to avoid counterexamples is paid for in precision cites §36 by title only, and that attribution needs correcting. Popper says higher universality and higher precision both raise content. When one statement is more universal and another more precise (his q and r), he says the two are not comparable by the subclass relation. The note's "content held constant" trade is therefore the note's own extension, not Popper's claim.

For the tentative theory definition, §§82–85 is the primary statement that corroboration does not reduce tentativeness: "every scientific statement must remain tentative for ever". Section 84 adds a point the definition does not yet carry. A corroboration appraisal is relative to a system of basic statements accepted up to a date, so each appraisal carries a "subscript". This is the logical shape of a snapshot-pinned freshness baseline, although that reference document does not claim a Popperian basis.

Two source comparisons hold on named axes. POPPER: automated hypothesis validation compares on how a probabilistic hypothesis becomes refutable: through a methodological decision about which frequency deviations count (§68), or through sequential tests with error control. Sound agentic science requires adversarial experiments compares on severity versus quantity. Section 82 makes corroboration depend on "the severity of the various tests" rather than on the count of instances, and appendix *ix rules out evidence selected to favour a hypothesis.

Learning Claims (our opinion)

Popper's mechanism is conjecture followed by deductive testing. Theories are proposed with no regulated method (§2). Consequences of lower universality are deduced from them and confronted with basic statements that investigators decide to accept (§§18, 29–30). A theory that survives is corroborated, and the appraisal is dated. A theory that fails is replaced, and a well-corroborated theory can only be superseded by a better-testable one that contains it or an approximation to it (§§79, 85). Across rounds, this "quasi-inductive" process raises the level of universality without any inductive inference.

Against the theory-builder conditions:

  • Localized content is strongly supported. §16 requires a system stated definitely enough that "every new assumption" is "easily recognizable for what it is: a modification and therefore a revision of the system". That is a reason for addressable form, stated as a precondition of severe testing.
  • Criticism is the book's core. §18 adds a qualified form of blame assignment: modus tollens falsifies the whole premise set, yet a falsification can sometimes be attributed to a newly introduced higher-level hypothesis. The same localization question appears in the addressability discussion, which already cites the later Conjectures and Refutations treatment.
  • Iteration is supported by §11's rule that a corroborated hypothesis may be dropped only for "good reason" (a better-testable replacement or a falsified consequence) and by §79's rule that a new system must also yield the old corroborated regularities.
  • Consumption is thin in this book. Theories are "applied" mainly as tests (§30). Action guided by theory is left to Popper's later texts, which the definition already cites.

The book does not claim that this process yields improved capacity for future action. It denies that science attains truth or even probability (§85), and its 1972 addendum limits the claim to a rationally justified preference among competing theories. So it grounds the definition's conditions but not the learning hypothesis the definition leaves open. On the fixed-decomposition question in learning inside a fixed decomposition inherits its mistakes, Popper's fixed layer is the accepted basic statements. They are fixed by decision for the current test only and remain testable. This is a methodological stance, not an implemented update space.

Extractable Value

  1. Removing particulars from a Commonplace claim produces a new conjecture, not a derived one. §13 classes statements about a finite, individual region as singular, even when they have universal form ("numerically universal"). §14 defines an individual concept as one whose definition needs a proper name, and holds that no abstraction leads from individual to universal names. A claim about Commonplace is singular in this sense. Deleting "Commonplace" gives a strictly universal claim, but the reference-side evidence does not transfer to it automatically. It corroborates the universal claim only to the extent that it was a severe test of that claim. Two further consequences follow for the move from kb/reference to kb/notes. KB terms defined through Commonplace artifacts are individual concepts until someone defines them without the proper name (compare §14's "pasteurized" example). And by §12, the reference fact becomes an initial condition from which the universal note, together with the system's particulars, predicts behaviour. §12 also warns that some universal premise can always be found to derive any given fact, so a universal note gains nothing from explaining the case it was abstracted from. [deep-dive]
  2. The line Popper draws is between testable and untestable reports, not between internal and external ones. §8 requires inter-subjective testability and dismisses any "occult effect" for which one "could give no instructions" for reproduction. §27 asks a dissenter to state a contradicting assertion "and give us his instructions for testing it". §29 accepts records such as "certificates of tests", which "if the need arises, can be re-examined". On this reading, a retained session trace, report, or ADR record can serve as a basic statement when others can re-examine or reproduce it. An operator's conviction that something worked cannot justify a claim however strong it is (§8). For the occasion, this suggests classifying internal evidence by whether it can be re-examined and reproduced, rather than by its origin. [quick-win]
  3. Corroboration should attach to a claim as a dated appraisal of the tests the claim survived, with severity recorded. §84 makes corroboration relative to basic statements accepted up to a point in time. §82 makes its degree depend on severity, not count. §83 says new instances add little once a theory is well corroborated, unless they come from "a new field of application". A corroboration link in the KB would therefore carry what test was run, against which snapshot, and why that test could have failed. A tally of supporting cases would not do this. [experiment]
  4. Evidence counts as corroboration only when it was gathered as an attempt to refute. Appendix ix: C(h, e) is a degree of corroboration only if e "reports the results of our sincere efforts to overthrow h", and "nothing is easier than to select statistical evidence so that it is favourable". §30 forbids accepting "stray basic statements", meaning ones not gathered in the course of testing a theory. §83 criticizes Keynes for treating a hypothesis proposed before or after the evidence as irrelevant. These passages support discounting evidence that was used to construct a claim. The explicit use-novelty rule and the role of background knowledge in severity, which the occasion names, are in Conjectures and Refutations ch. 10, not here. In this book, "background knowledge" appears only as the formal parameter z in appendix ix's corroboration function. [quick-win]
  5. A falsification is a corroborated low-level hypothesis, not a stray counterexample. §22 requires a reproducible effect described by a corroborated "falsifying hypothesis", which can itself be of low universality (for example, a family of white ravens in the New York zoo). A single failed case in a session therefore justifies logging and a reproduction attempt. A refutation needs the effect stated so that it can recur. [quick-win]
  6. Repairs must raise what the system forbids. §20's rule for auxiliary hypotheses, together with the redefinition rule, gives one guard behind the KB's separate repair-escape notes (narrowing, generality, domain pricing, the ad hoc sense). Popper admits his list of stratagems is incomplete, so a unified guard would still need the per-escape diagnostics. The generality-note attribution needs correcting at the same time (see Connections Found). [experiment]

Limitations (our opinion)

The book argues normatively from logic and selected episodes in physics. It does not compare methods empirically, and its historical cases are chosen to illustrate the rules. Popper corrects some load-bearing examples himself: a starred note to §20 withdraws the claim that the FitzGerald–Lorentz contraction had no falsifiable consequences. Starred notes to §§18 and 35 admit that the 1934 text confused conditionals with entailment. Anyone quoting the book should check whether a starred note qualifies the passage.

The key operations remain underspecified. The rules for accepting basic statements are sketched through the jury analogy and a few rules (§§29–30), not given as a procedure. The degree of corroboration is non-numerical in the main text (§82), and appendix *ix's formal measure depends on a logical-probability metric that Popper says cannot be defined on purely logical grounds. Degrees of falsifiability are only partially ordered (§§33–34), so many pairs of KB claims will not be comparable by content.

Transfer to a KB is our interpretation. Popper's units are strictly universal empirical hypotheses about nature. Many KB artifacts are procedures, definitions, conventions, or design decisions. For these, "potential falsifier" needs a reinterpretation the book does not supply, and a methodological convention is by Popper's own account (§11) not falsifiable in the same way. The treatise also centres on physics with measurable magnitudes, where reproducibility and precision are well defined; for prose claims tested by LLM review, both are open questions. The capture is full, but this analysis read chapters 1–7 and 10, §§65–68, and parts of appendix *ix closely. It did not assess the probability formalism, the quantum-theory chapter, or most starred appendices.

Run cp-skill-ground on tentative theory with Claim needed: corroboration is an appraisal relative to basic statements accepted up to a stated time, whose degree depends on test severity rather than the number of corroborating instances (§§82, 84). This gives the operator's corroboration links a primary-source definition to cite before any claim-level corroboration format is designed.