### [[Golden Rules for Analytic Thinking and Writing]] These principles should form the basis of how you approach thinking about, writing, and rewriting [[analytic philosophy]]. 1. **Stage background before application** - Principle. Present the theoretical provenance first, then indicate the selective adoption, then operationalise in the project’s own primitives. - Why. Readers can assess aptness only when they see what is kept, what is dropped, and why. - Examples. - [[Predictive processing]]: start with Bayesian/variational background; retain hierarchical message passing but bracket strong metaphysics; define local primitives (error units, precision, policy) for the target explanandum (e.g., mind-wandering). - Extended mind (2020s): begin with parity/integration conditions; adopt a specific integration norm (e.g., dynamically decouplable vs. task-bound); cash out “coupling,” “resource,” and “boundary test” for [[the particular]] case (e.g., sensorimotor prosthesis vs. cloud-assistance). 2. **Anchor selectively to canonical sources** - Principle. Name only traditions you will use and specify exactly which elements are retained. - Why. Avoids overreach and confusion; shows disciplined uptake. - Examples. - Consciousness science: from [[global workspace]], retain broadcasting/ignition markers but not sufficiency claims; from IIT, retain causal-structure diagnostics for subsystem comparison, not identity theses. - Social cognition: cite dual-process literatures only for task-relative response profiles (speed/effort dissociations), not a two-minds metaphysic. 3. **Introduce only terms you will later employ, in [[the order]] needed** - Principle. Define minimally at first use; avoid terms that will not reappear. - Why. Reduces cognitive load and terminological drift. - Examples. - Metacognition: if the focus is confidence calibration, define “confidence report,” “type-1 performance,” and “type-2 sensitivity”; avoid unused metarepresentational taxonomies. - Action control: if analysing “urge,” “intention-in-action,” and “inhibitory set,” do not introduce full planning hierarchies unless used later. 4. **Encode constraints in definitions, not as add-ons** - Principle. Build methodological constraints (e.g., non-mentalistic, non-teleological, non-referential) into term specifications rather than appending policy notes. - Why. Constraints then follow from usage, not fiat. - Examples. - Affect: define “valence estimate” as a control variable over error landscapes to avoid implicit appraisal talk. - Perception: specify “feature map” and “selection policy” when phenomenology is not the explanandum. 5. **Tether to earlier premises rather than re-teaching** - Principle. When invoking fixed commitments, cite them and proceed; do not re-derive. - Why. Keeps argument forward-moving; signals reliance on agreed ground. - Examples. - Attention: if selection/modulation was fixed earlier, refer back (“given the selection/modulation split…”), not re-argue it. - Agency: if comparator accounts were excluded, operate under that exclusion, do not reopen it. 6. **Match the register and cadence of the target document** - Principle. Maintain analytic prose, vary sentence length, avoid rhetorical questions and meta-announcements. - Why. Coherent style reduces friction and forces explicit inferences. - Examples. - Phenomenal character: prefer “Given X and Y, Z follows…” to framing with rhetorical questions. - Computational ontology: prefer direct specification (“‘state’ denotes a triple ⟨values, update rule, read-out⟩”) over meta talk (“In what follows we will…”). 7. **Maintain economy and cohesion** - Principle. Each sentence should advance the set-up or argument; avoid enumerative taxonomies or task recipes unless demanded. - Why. Prevents scope creep and checklist bias. - Examples. - Active inference: avoid “ten steps to policy selection”; give minimal dependencies (e.g., precision weighting over expected free energy). - Embodiment: avoid catalogues of modalities; specify the coupling relation required and why. 8. **Prepare later process/product pairings implicitly and locally** - Principle. If a later section relies on a process/product dual, seed it with nearby wording that distinguishes capacities vs. outcomes, without naming the pairing prematurely. - Why. Eases later uptake and avoids duplication. - Examples. - Memory: speak of “standing synaptic dispositions” vs. “retrieved pattern” before a formal consolidation/recall split. - Perception–action: speak of “control law” vs. “realised course” before policy/trajectory distinctions. 9. **Derive stances from fixed premises** - Principle. Frame methodological or ontological stances as consequences of what has been fixed, not as independent declarations. - Why. Shows constraints are earned. - Examples. - Enactivism (local use): if task-bound couplings and success conditions are fixed, non-representational talk follows as a consequence. - Non-reductive physicalism (local use): if causal exclusion constraints and multiple realisation are fixed, the stance can be presented as entailed. 10. **Keep the unit of analysis internal to the posited boundary** - Principle. Describe only what falls within the fixed boundary; exclude external resources unless made relevant by explicit inclusion criteria. - Why. Prevents import of unstated resources; preserves testability. - Examples. - Social cognition: if the unit is online inference in a joint task, do not import developmental resources unless shown available at run-time. - Computational psychiatry: if the unit is a trial-level decision, avoid importing cross-session priors unless specified in the generative model. 11. **Use “lights” as discipline-relative pairings of process kinds and entity kinds** - Principle. When invoking a scientific “light,” make explicit that, under that light, [[the process]] kinds it treats as basic play the productive role; the entity kinds it recognises are the produced. Mark relativity to scale and discipline. - Why. Clarifies how different sciences yield different process/product readings of the same scene. - Examples. - Neurobiology: under systems neuroscience, oscillatory coupling and synaptic plasticity are process kinds; local field potentials and activation patterns are entity kinds. Under cognitive ontology, evidence accumulation and gating policies are process kinds; decisions and reports are entity kinds. - Affective science: under RL-based accounts, updates and policy changes are process kinds; choice patterns and latencies are entity kinds. Under constructionist models, feature integration and labelling are process kinds; emotion categories are entity kinds. 12. **When excluding predicates, specify the replacement vocabulary** - Principle. If excluding mental or speech-act predicates, say what vocabulary does the explanatory work. - Why. Avoids vacuity. - Examples. - Agency illusions: replace “experiences control” with “comparator mismatch drives the postdictive control rating,” or with “policy arbitration output is read by a learned control schema.” - Belief reports: replace “believes p” with “report generated by thresholded evidence integration over channel c.” 13. **Mark scale sensitivity explicitly** - Principle. Indicate how terms and claims scale (trial → task → episode → lifespan), and what changes at each level. - Why. Prevents misapplication across levels. - Examples. - Mind-wandering: a trial-level decoupling signature does not entail trait-level propensity; state the targeted level. - Metacognitive efficiency: distinguish within-task type-2 sensitivity from across-task metacognitive ability. 14. **When importing formalism, separate the formal role from metaphysics** - Principle. Use formal tools for explanatory discipline; do not slide into metaphysical theses unless argued for. - Why. Prevents category mistakes and over-claiming. - Examples. - Free-energy: use expected free-energy to organise policy evaluation; avoid “the brain minimises free-energy at all times” unless defended. - Causal discovery: use structural models to articulate intervention–counterfactual structure; avoid causal realism claims unless developed. 15. **Make failure cases do work** - Principle. Use misfits (illusions, breakdowns, edge cases) to tighten definitions and boundaries. - Why. Negative cases constrain scope better than positive cases alone. - Examples. - Temporal illusions: use binding phenomena to refine timing/control variables without importing mental experience talk. - Pathologies: use depersonalisation or thought insertion to constrain boundary tests for attribution systems in agency models.