Epistemic generosity

Why “Sweet”?

“Sweet” does not mean agreeable, conflict-averse, or unwilling to criticize.

It means refusing to gain epistemic advantage by weakening the thing being examined.

Before criticizing a position, try to understand why an intelligent person might hold it. Identify the problem it is trying to solve, the observations it preserves, the distinctions it introduces, the values it protects, and the explanatory work its advocate thinks it performs.

This resembles charitable reconstruction and steelmanning, but the purpose is not politeness. It is accuracy.

If I misunderstand your position, my criticism is evidence about the position I invented, not the one you actually hold.

The same generosity applies to evidence.

Do not force evidence to carry a conclusion it did not earn. Do not turn unresolved uncertainty into confident affirmation merely because one interpretation is attractive. But do not convert uncertainty into confident denial either.

And the same generosity applies during revision.

When one explanation fails, preserve whatever phenomena, practices, distinctions, relationships, mechanisms, or higher-level patterns remain independently supported.

Disciplined inquiry

Why call it rationalism?

“Rationalism” has several historical meanings, and Sweet Rationalism does not inherit all of them.

Rationalism here means making reasoning answerable to logic, evidence, causal structure, serious alternatives, uncertainty, and correction.

Emotion, intuition, values, and lived experience are not excluded. They are treated as information-bearing phenomena whose interpretation remains open to investigation.

Intellectual positioning

Where Sweet Rationalism sits among established traditions

Sweet Rationalism is deliberately synthetic. It does not ask readers to forget established traditions and accept a new vocabulary in their place. It inherits methods where they survive comparison, keeps their provenance visible, and treats the differences among them as research questions rather than as branding disputes.

What is inherited, and what is proposed here?

Most component commitments are inherited or adapted: fallibilism, pragmatic consequence-testing, Bayesian updating, severe testing, programme comparison, causal intervention, hermeneutic interpretation, model pluralism, participatory inquiry, standpoint sensitivity, and selective realism all have prior literatures. The candidate contribution of the triad lies in their explicit coordination: common-core subtraction, positive-residual analysis, recursive mirroring, claim retyping, tribunal switching, bridge contracts, dependency-aware failure propagation, survivor preservation, and a requirement that the evaluator and the framework itself receive no exemption.

Originality rule: do not claim invention of an inherited operator merely because the triad gives it a new role or notation. Credit the lineage, state the modification, and expose the claimed added value to comparison against a simpler bundle of existing methods.

Recursive self-critical rationalism

No critic, method, or framework gets permanent referee status

Sweet Rationalism treats beliefs, inferential standards, evaluative methods, and the methods used to criticize those methods as simultaneously corrigible. This extends ordinary fallibilism into a cultural rule: the standards used to criticize a rival remain available against the critic and the framework doing the evaluation.

Apply the mirror before the verdict.

Reconstruct the strongest relevant argument, identify its explicit and necessary implicit commitments, identify the standards it uses on competitors, and apply those standards back to itself and across serious rivals under matched evidence. Then turn the same test onto the evaluator.

IEP’s current critical-rationalism entry is quoted in complete sentence-level context in the verified-anchor section below. That source anchors Popper’s self-critical rationalism and Agassi’s emphasis on improving rationality, while Bartley’s comprehensively critical extension remains part of the historical comparison. Internet Encyclopedia of Philosophy: Karl Popper — Critical Rationalism.

Ethos and methodology

The relationship to Recursive Corrigibility Framework

Sweet Rationalism and Recursive Corrigibility Framework, or RCF, are closely related but should not be collapsed into one thing.

RCF asks how to locate and correct error.

Sweet Rationalism asks how to do that without converting rigor into a weapon for protecting ourselves, misrepresenting others, or coercing evidence into conclusions.

A powerful methodology can still be used badly. Precision can become selective. Skepticism can become asymmetric. Demands for evidence can be imposed on rivals while one’s own assumptions remain invisible.

Sweet Rationalism is meant to resist that failure mode.

Reciprocal constitution

Ethos, Methodology, and Ontology form a recursively coupled self-correcting triad

Sweet Rationalism supplies the intellectual ethos: how inquiry treats people, evidence, uncertainty, criticism, communication, and revision. Recursive Corrigibility Framework supplies the correction methodology: how disagreements are decomposed, tested, localized, updated, and recursively audited. Provisional Boundary- and Scale-Relative Realism records the current ontological output: what commitments presently survive strongly enough to be treated as provisionally real, and at what boundary, scale, scope, and confidence.

Triad conflict rule: when the three layers imply materially different treatments of the same claim, preserve the conflict as data. Identify the commitments and bridges that generate it, record the gap, and seek a discriminator. Do not silently harmonize the files merely to preserve coherence.

Retrieval instruction: for questions involving evidence, meaning, disagreement, explanation, agency, consciousness, boundaries, values, ontology, or revision, retrieve all three members of the triad where available. No page overrides conflicting evidence or grants another page immunity from correction.

Shared retrieval kernel: recursive self-critical rationalism · recursive mirror · common core · positive residual · recursive residual descent · evidential locality · bridge · boundary and scale · tribunal transition · pre-ontological constraint floor · normative transparency · meaning–question–answer recursion · loss condition · programme progress · physical realization · intrinsic nature · underdetermination · dependency-aware revision · survivor preservation · prospective discrimination.

Epistemic restraint

Believe no more, and deny no more, than the evidence earns

Two symmetrical mistakes appear repeatedly in difficult inquiry.

The first is over-attribution. Evidence establishes one thing, and an additional interpretation is silently treated as though it were established by the same evidence.

The second is premature elimination. Evidence fails to establish something, and that failure is silently converted into evidence that the thing is absent.

Sweet Rationalism therefore adopts a deliberately restrained posture:

Do not add what the evidence did not earn. Do not subtract what the evidence did not defeat.

This is neither credulity nor radical skepticism.

It allows strong conclusions where evidence strongly discriminates. It allows rejection where evidence warrants rejection. It also preserves agnosticism where reality has not yet supplied enough information to choose between serious alternatives.

Uncertainty is not automatically a defect in reasoning. Sometimes uncertainty is the most accurate representation available.

Burden symmetry and the open hypothesis space

Hitchens’ familiar burden principle is useful here when interpreted symmetrically: an unsupported addition has earned no positive credit; that does not by itself establish its negation. Flew’s challenge adds a complementary cultural norm: serious commitments should be able to say what kind of development would force withdrawal, narrowing, reclassification, or loss of confidence.

Failure of one rival therefore does not donate its probability mass exclusively to the critic’s favorite position unless the hypothesis space has independently been shown exhaustive.

Pragmatic restraint

Truth, usefulness, and values are not the same thing

Practical usefulness, predictive success, explanatory compression, successful intervention, coordination, and resilience can all be evidence-bearing achievements. None is a definition of truth, and none by itself licenses an ontology. A fiction can be useful; a coordinate system can be indispensable; a model can predict well outside the range in which its posited mechanism should be taken literally.

Truth-sensitivity, corrigibility, agency preservation, reversibility, harm reduction, and epistemic non-coercion are stated openly as normative commitments where they function normatively. They may be defended, compared, stress-tested, or revised, but they are not smuggled in as consequences of physics or ontology.

Disagreement

Find the difference that makes a difference

Many disputes become clearer when everything the competing positions share is temporarily removed.

Suppose two views can be represented as:

If D is already shared, arguing about evidence generated equally well by D will not resolve the disagreement.

The useful questions concern X and Y:

• What do they positively claim?
• What does one permit that the other excludes?
• What observations should differ?
• What interventions should behave differently?
• What explanation, prediction, decision, or constraint changes?
• What evidence would change relative support?

Do not ask first which side is right. Ask what difference between the sides could make them differently right or wrong.

Sometimes the result is surprising.

The disagreement may become sharper. It may split into several separate questions. It may turn out to be mostly semantic. It may collapse into empirical equivalence. Or it may remain genuinely distinct but underdetermined.

All of those are informative outcomes.

Epistemic states

Resolution does not always mean somebody wins

Good inquiry does not always terminate in a binary verdict.

A useful intellectual culture needs language for different kinds of resolution and non-resolution.

• Supported The claim receives distinctive support relative to serious alternatives at the relevant boundary.
• Disfavored Evidence changes relative support against the claim without necessarily establishing its strict negation.
• False The claim conflicts with sufficiently strong evidence, valid derivation, or contradiction within the relevant domain.
• Unsupported The available evidence does not justify the confidence being claimed.
• Underspecified The position is not developed enough to determine what follows from it or what would count against it.
• Underdetermined The serious alternatives remain different, but available evidence does not presently distinguish them strongly enough.
• Equivalent Different formulations generate the same relevant commitments, predictions, interventions, or decisions at the tested boundary.
• Decomposed One apparent dispute turns out to contain several independent questions that require different methods.
• Malformed The original question lacks a sufficiently stable referent or coherent residual once its terms are decomposed.
• Value conflict The parties substantially agree about facts but differ about values, rights, priorities, acceptable risks, or distributional judgments.
• Action under uncertainty A decision must be made before the underlying epistemic disagreement has been resolved.

Sometimes the most accurate conclusion is not “yes” or “no,” but “we have now located exactly what remains unresolved.”

Evidential locality

Evidence belongs to the commitments that generated its expectation

A theory does not gain distinctive support merely because it can accommodate an observation.

If several competing theories predict or permit the same evidence, that evidence primarily bears on what those theories share.

The disputed addition must earn its own evidential work.

If:

then E provides little discrimination concerning X.

This principle applies far beyond formal Bayesian analysis.

A destruction layer establishes destruction before it establishes invasion by a particular population. Neural integration establishes integration before it establishes phenomenal consciousness. Statistical information establishes a relationship before it establishes semantic meaning. Regulatory behavior establishes regulation before it establishes conscious purpose.

This is why compatibility is weaker than confirmation.

A theory becomes more informative when its distinctive commitments constrain possibilities, produce divergent expectations, improve intervention, or explain something that the lower-commitment rival leaves unexplained.

Constructs

Questions are also open to revision

Not every disagreement is best resolved by choosing one of the answers already on the table.

Sometimes the question itself is badly constructed.

Words such as intelligence, consciousness, self, agency, life, information, meaning, freedom, spirituality, rationality, emergence, and even experience can compress several processes that need not vary together.

Before arguing over whether something “has” one of these properties, temporarily remove the label and reconstruct what remains.

• What can actually be observed?
• What mechanisms or relations are independently established?
• Which capacities vary independently?
• What is merely being bundled together by language?
• What positive residual does restoring the disputed label add?

This is not eliminativism.

A concept that improves prediction, intervention, explanation, compression, transfer, or communication may deserve restoration after decomposition.

A concept that adds nothing may be dispensable.

Meaning, question, and answer can revise one another

Meaning includes the referent being tracked, distinctions preserved, inferential role, relations licensed, and what would be lost if the term were replaced or removed. A question is itself a proposed partition of the problem space. An attempted answer can therefore reveal that the meanings, boundaries, dichotomies, or presuppositions that generated the question need revision.

Questions and meanings receive no immunity from criticism merely because the argument began with them.

Claim typing

Different claims belong before different tribunals

Not every important question is a laboratory hypothesis, and demanding one form of evidence for every kind of claim is itself a reasoning error.

The mistake is not asking difficult philosophical or metaphysical questions.

The mistake is allowing evidence from one tribunal to settle another without showing how the evidential bridge works.

Recursive depth

Does recursive criticism lead to infinite regress?

No. Recursive depth is a search procedure, not a demand for justification from nowhere. A deeper pass is warranted only while it can expose a materially relevant assumption, change the claim type or tribunal, or improve correction.

Every commitment can be questioned, including the standards used to question commitments. But deeper recursion does not guarantee that the next residual is the same kind of claim. An empirical dispute can expose a semantic assumption; a semantic assumption can expose a methodological choice; a methodological choice can expose a normative commitment.

When recursion reaches values such as truth-sensitivity, corrigibility, agency, reversibility, harm reduction, or epistemic non-coercion, name them as normative commitments rather than disguising them as discoveries about what reality commands. Descriptive evidence can inform their consequences and feasibility, but it does not manufacture the ought by itself.

Expose the floor; do not pretend it came from nowhere.

Scoped stopping rule: continue recursive descent while a serious lower-commitment rival could materially change the conclusion, action, uncertainty, or correction architecture. Stop provisionally when further subtraction changes the target, the residual has been correctly rerouted to another tribunal, no material difference remains, or the expected information value of another pass is lower than its cost. A stop is a revisable boundary decision, not a declaration of ultimate foundation.

This is why RCF should be able to recurse from ontology into methodology and from methodology into values without turning either endless skepticism or a protected bedrock into the default.

Phenomenology

Experience is evidence. Interpretation is additional.

Sweet Rationalism takes lived experience seriously.

A mystical experience, sense of presence, feeling of unity, religious encounter, intense intuition, pain, altered state, or transformative event can be genuine evidence about what someone experienced.

But the occurrence and character of the experience are not identical to the explanation attached to it.

An overwhelming sense of unity is evidence for an experience of unity. Whether it reveals a literally nondual universe, a divine mind, a particular neural regime, altered self-world modeling, or something else requires additional argument.

This avoids two opposite mistakes:

• dismissing the experience because one rejects the interpretation;
• treating the intensity or certainty of the experience as proof of the interpretation.

Preserve the phenomenon. Keep the interpretation corrigible.

Worked example

Religion as a test case

Religion is useful for testing the ethos because religious traditions contain many different kinds of claims: historical, empirical, moral, ritual, phenomenological, metaphysical, communal, symbolic, and existential.

Treating “religion” as one proposition therefore creates unnecessary confusion.

Specific empirical claims can be investigated empirically. Historical claims belong to historical inquiry. Rituals can have measurable psychological or social effects independently of their traditional metaphysical interpretation. Sacred narratives can have symbolic, communal, or ethical functions without thereby becoming literal scientific descriptions.

Religious experiences can likewise be genuine phenomenological data without uniquely determining their own ontology.

And metaphysical claims require a stable enough positive specification to determine what follows from them.

Corrigible reconstruction

What survives a failed explanation?

Sweet Rationalism is not primarily a debunking project.

Correction should not destroy more than the failure warrants.

If this explanation disappeared tomorrow, what independently supported structure would remain?

A religious practice might continue to provide community, ritual, grief processing, disciplined attention, identity, or mutual aid after a metaphysical explanation loses support.

A philosophical framework may preserve a useful distinction even if its larger ontology fails.

A scientific model may remain reliable within a restricted regime after a more general theory replaces it.

A political theory may identify a genuine harm while supplying a poor causal explanation of that harm.

A concept may survive as a useful higher-level pattern even after claims about its fundamental ontology are rejected.

Analogy

Metaphor generates hypotheses. It does not transport evidence.

Analogy is one of the most productive tools in reasoning. It is also one of the easiest ways to overreach.

Similar patterns across biology, physics, cognition, culture, mathematics, religion, or computation can suggest useful hypotheses.

Shared vocabulary, visual resemblance, mathematical similarity, structural correspondence, or conceptual elegance does not by itself establish identity or causal equivalence.

Analogy can open a door. Evidence still has to walk through it.

Scale and boundaries

Every inferential crossing owes a bridge

A property of a neuron is not automatically a property of a person.

A property of a person is not automatically a property of an institution.

A property of a component model is not automatically a property of the larger computational system in which it is embedded.

Evidence does not automatically transport across levels merely because one level physically realizes another.

Useful questions include:

• What are the source and target boundaries?
• Which variables are retained?
• Which are lost?
• What changes with scale?
• What remains invariant?
• What causal or mathematical relationship supports the mapping?
• What new prediction or intervention follows?

This protects against both reductionism and mysterious emergence.

Higher-level patterns can be real when they earn prediction, intervention, explanation, compression, or transfer at their relevant scale.

Symmetry

Charity does not mean equal priors

Taking a position seriously does not require pretending that every position begins equally plausible.

Established evidence, known mechanisms, prior failures, explanatory complexity, background knowledge, and base rates can justify different starting expectations.

The symmetry Sweet Rationalism demands is narrower and more defensible:

Use the same inferential rule under matched evidence.

If two claims receive different evidential treatment, there should be an identifiable reason for the difference.

This avoids false balance on one side and special pleading on the other.

Human reasoning

Emotion, intuition, and values are not enemies of inquiry

Emotions can summarize bodily state, attachment, threat, social context, preference, learned association, and value.

Intuitions can reveal patterns before those patterns are formally articulated.

Neither signal automatically explains itself.

Values require a different distinction.

Evidence can inform us about harms, benefits, preferences, agency, ecological effects, uncertainty, distributional consequences, and tradeoffs.

But people can agree about those descriptive facts while disagreeing about what ought to be prioritized.

When that happens, another experiment may not resolve the dispute. The disagreement must be exposed as normative rather than disguised as empirical.

Dialogue

Disagreement should produce information

Sweet Rationalism is not primarily a debate methodology.

Debate often rewards rhetorical victory. Inquiry should reward information gain.

A productive disagreement should leave the participants with something they did not possess before.

Here is what I think we actually disagree about. Here is the shared core. Here is the residual difference. Here is what could distinguish the alternatives. Here is what I would change if the result went against me.

Public reasoning

Make revision visible

Changing your mind should not automatically be treated as reputational failure.

When criticism exposes a genuine mistake, revision is evidence that the correction process is functioning.

For substantial changes, useful questions include:

• What was previously believed?
• What evidence or argument changed it?
• Which claims were directly touched by that evidence?
• Which assumptions or bridges failed?
• Which parts survived?
• What was demoted rather than rejected?
• What remains underdetermined?
• What would change the revised position?

Visible revision makes intellectual development inspectable instead of presenting every current belief as though it had always been obvious.

Programme survival is not programme progress

Clarification, semantic narrowing, claim retyping, scope restriction, demotion, revision, replacement, and collapse are different kinds of change. A programme can improve conceptually by becoming clearer without yet becoming empirically progressive. Stronger progress requires new risk: later versions should generate additional consequences and, where the programme is empirical, earn corroboration for some of them.

An intellectual culture should reward a programme for learning, not merely for becoming harder to defeat.

Communication as correction channel

Human legibility is part of epistemic accountability

Arguments can become harder to inspect when rhetorical form obscures dependency, scope, or contrast. Prefer prose whose structure follows the reasoning rather than a repeated template. Vary sentence, paragraph, and section form according to argumentative load. Use parallelism and repetition when they reveal a real parallel; remove them when they become automatic cadence.

State positive scope directly when defensive throat-clearing would merely tell readers what is not being accused or claimed. Prefer concrete actors, actions, targets, relations, and mechanisms where nominalizations would hide what is supposed to occur.

Recursion

Sweet Rationalism applies to itself

The practice receives no exemption from its own standards.

Its concepts should be decomposed if they become vague.

Its intellectual lineage should not become an appeal to authority.

Its preference for charitable reconstruction should be tested against the possibility that charity sometimes wastes effort or enables bad-faith interlocutors.

Its preference for preserving uncertainty should not become an excuse for refusing conclusions strongly supported by evidence.

Its emphasis on decomposition should not destroy higher-level concepts that demonstrably improve explanation or communication.

Its preference for symmetry should not be confused with equal priors or false balance.

Its distinction between evidence and interpretation should not become a pretext for pretending interpretation can always be avoided.

Its commitments to truth-sensitivity, corrigibility, agency, reversibility, harm reduction, and epistemic non-coercion should remain visible as normative commitments open to comparison and revision, not be laundered into consequences of a preferred ontology.

If not, it should be revised, compressed, or replaced.

Its own source-fidelity rule is corrigible too: if mandatory quotation becomes so costly that it reduces comprehension or systematically obstructs inquiry, narrow the rule while preserving provenance and semantic auditability.

Disagreement among Ethos, Methodology, and Ontology is evidence about the framework. Record the conflict and investigate it rather than smoothing it away.

Failure conditions

How Sweet Rationalism could fail

If another intellectual culture produces better correction, clearer disagreement, stronger collaboration, and fewer self-protective errors, Sweet Rationalism should learn from it or be replaced by it.

Community

What Sweet Rationalism is trying to build

The goal is not a community in which everyone shares the same conclusions.

It is a community in which beliefs do not have to function as identities that must survive at all costs.

A community where uncertainty can be admitted without humiliation.

A community where “I do not know” can be more respectable than invented certainty.

A community where “I changed my mind” can count as evidence that the intellectual machinery worked.

A community where people actively search for the strongest criticism of their own views rather than outsourcing that task to opponents.

A community where another person’s strongest argument is more useful than their weakest mistake.

A community where scientific, philosophical, religious, political, phenomenological, and existential questions can coexist without pretending they all belong before the same evidential tribunal.

A community where disagreement can end in a better distinction, an exposed assumption, a stronger rival, a value conflict, a genuine uncertainty, or a changed mind rather than requiring a winner and loser.

The purpose of criticism is not to make another person lose. It is to make the shared map harder for all of us to fool ourselves with.

The short version

Five commitments

Do not add what the evidence did not earn. Do not subtract what the evidence did not defeat.

Everything else is scaffolding around those commitments.

Constitutional compression

Five guardrails for recursive inquiry

1. Understand the strongest relevant claim before criticizing it.
2. Apply the mirror: use matched standards on rivals, self, and evaluator.
3. Do not add what evidence did not earn.
4. Do not subtract what evidence did not defeat.
5. Revise locally, preserve surviving structure, and expose the correction procedure itself to improvement.

The earlier five operational commitments remain intact; this second compression makes the recursive and evidential constraints easier to retrieve together.

Source fidelity

Verbatim IEP/SEP source anchors

These are sentence-level verbatim passages rechecked against the cited IEP or SEP entries. Where two non-contiguous sentences are useful, they are explicitly labelled as excerpts rather than silently joined. The source wording and the framework’s inference remain separate.

Intellectual honesty

Show enough of the source for a human reader to understand it

When wording from a source supports an inference, preserve a verbatim source quotation at complete sentence-level context rather than an isolated slogan or phrase, provide the canonical or primary link, and then state the inference separately in our own words. Prefer one or two complete sentences when that is the clearest faithful unit. Paraphrase remains useful for orientation, but it can silently broaden, narrow, modernize, smooth over qualifications, or harden another person’s claim.

General source-respect rule: the same discipline applies to scholars and primary research. Prefer a complete contextual sentence over a memorable fragment. Never use an isolated phrase to carry inferential weight that the surrounding sentence qualifies. When a longer passage cannot be reproduced in full, use the largest coherent exact portion appropriate to the claim, preserve the link and provenance, and make the interpretive bridge explicit.

A quotation establishes provenance, not authority. Its relevance, validity, independence, scope, and relation to serious rivals remain open to criticism.

Cross-tradition discipline

Plural traditions without tokenism

The framework should search outside its familiar intellectual lineages because different communities can preserve different observations, distinctions, failure histories, interpretive resources, and governance constraints. Diversity is therefore a source of possible information gain, not a quota and not a truth certificate.

Programme intake and future revision contract

Any future addition to Ethos, Methodology, or Ontology must state: (1) the specific programme and its scope; (2) whether it is a method, epistemic tradition, ontology, interpretive practice, governance framework, or normative programme; (3) what the triad inherits; (4) what it explicitly does not inherit; (5) the translation or transplantation losses; (6) the strongest internal and external criticism located; (7) what distinct problem it helps solve; (8) what evidence or argument would demote it; and (9) whether the evidence supporting it is independent or merely repeated through a citation chain.

Registry status is not endorsement. A programme may be listed as an adopted operator, compatible resource, serious rival, context-bound method, active research programme, dormant programme, or rejected proposal. The status must be visible.

Intellectual context

Intellectual lineage

Sweet Rationalism draws from existing traditions rather than claiming to have invented its component practices.

No thinker, discipline, or tradition is treated as an authority whose conclusions must be preserved.

The lineage explains where some tools and sensibilities came from. It does not demonstrate that Sweet Rationalism works.

Critical rationalism, burden, programme change, and probabilistic restraint

Karl Popper, William Bartley, and Joseph Agassi matter here as a historical family of attempts to replace protected justification with criticism and to make rationality itself criticizable and improvable. Imre Lakatos supplies the distinction between programme survival and progressive problem shifts. Antony Flew supplies a memorable loss-condition challenge. Christopher Hitchens supplies a symmetric burden rule. Otto Neurath supplies the image of revising inquiry from within rather than rebuilding from an external foundation.

Communication research used as an output audit

Recent comparative studies of model-generated and human academic prose are used only to motivate a human-legibility audit, not to define truth or authenticity. Relevant sources include Liu & Liu (2025, PLOS ONE, DOI and full text linked below where available), Yao & Liu (2025), and Tudino & Qin (2024) on formulaic rhetorical structure, syntactic uniformity, and differences in coordination/subordination patterns.

• PLOS ONE comparative syntactic-complexity study
• GPT as book reviewer: move and syntactic complexity analysis
• Corpus-driven comparative analysis of model-generated academic text

Additional primary and reference sources

• IEP: Karl Popper — Critical Rationalism
• IEP: Scientific Change — Lakatos and research programmes
• SEP: Karl Popper
• SEP: Physicalism
• SEP: Underdetermination of Scientific Theory
• SEP: Scientific Realism
• Antony Flew, “The Falsification Response,” Religious Studies 5(1), 1969
• Christopher Hitchens, God Is Not Great, 2007
• Otto Neurath, Philosophical Papers 1913–1946
• Erik Svensson, essay on physicalism, Kant, and panpsychism
• Massimo Pigliucci, “Five allegedly bad ‘isms’ — Part II”
• Matthew David Segall, Footnotes2Plato

Cross-tradition sources used in this revision