⛶
minimalist background-lit research brief

Brief: deep research for a background-literature paper on minimalist causal coding#

Produce a standalone literature paper that situates the minimalist / barefoot coding stance (from Minimalist coding for causal mapping) inside the existing scholarly conversation. The paper should set out the strongest opposing positions (steelmen), the views that already agree or run parallel, and the live debates between them, staying as close as possible to the actual arguments and choices made in the minimalist paper rather than wandering into general philosophy of causation.

The downstream aim is to beef up the minimalist paper: to let it anticipate objections, cite the people who already hold its position or the opposite one, and avoid reinventing terms that have settled names in the literature. So the research output is an input to a revision, not a finished garden page.

Each of these is a specific commitment in the paper. The research should attach existing literature to each one: who argues the opposite well, who already argues something similar, and where the debate sits.

  1. Propositions, not variables. Factors are bare propositions ("I started to eat adequately"), not variables with values. Most named causal-mapping traditions would instead code amount eaten -> energy level. The paper claims variable coding is usually "ontologically over-determined" and does not faithfully represent what speakers said.
  2. No polarity by default. Unlike almost everything else called "causal mapping", links carry no +/- sign out of the box.
  3. No strength, weights, or functional form. No magnitudes, no monotonic-relationship assumption, no simulation affordance.
  4. Influence, not determination; no sharp counterfactual. A link means X made a difference to Y "in virtue of its causal power to do so", not that X is necessary or sufficient. Counterfactuals are left where the speaker left them.
  5. Cognition / surface claims, not the world (the "Janus problem"). The object coded is the speaker's expressed causal thinking, with reality bracketed off. Stay on the surface; do not infer hidden meaning.
  6. Presence/absence asymmetry; no coding of absences. Not mentioning a factor is not evidence against it; proportions are suspect because the denominator is unclear.
  7. Evidence volume is not effect size. Citation and source counts measure corpus evidence, not causal magnitude.
  8. Deferred judgement / code willy-nilly into bundles. Code all co-terminal claims first, judge quality/relevance later. Contrast with deferred-to-mapping evaluative judgement (Jewlya Lynn; contribution rubrics; contribution analysis more broadly).
  9. Multiple sources, combined later. Per-source maps then aggregated, vs single-expert maps or consensus maps.
  10. The transitivity trap and thread-tracing. It is usually invalid to stitch A->B (source 1) and B->C (source 2) into a chain; only within-source paths are safe.
  11. Context and mechanism get no special machinery. Both are coded as ordinary factors/links; the paper declines the realist CMO apparatus and the "what is a mechanism" debate.
  12. The 90% rule and the residual 10%. Enablers/blockers (second-order causal claims), causal packages/conjunctions, and idioms that pack a causal story into one phrase ("went viral") are acknowledged as out of reach and left by choice in the write-up, not the data structure.
  13. Scale + auditability + AI clerk. The minimalism is justified pragmatically: it is easy to apply, easy to check (quote + source provenance), and easy to automate, with the human as architect and the AI as clerk.

For every cluster: name the strongest opposing steelman, the closest allies/precursors, and the central disagreement, tied to the numbered claims above.

Not just for-and-against. Each cluster is rarely a single two-sided argument. Expect several overlapping positions, partial agreements, and sub-debates that cut across the simple pro/anti minimalist line: people who agree with the minimalist conclusion for different reasons, people who share its premises but reach the opposite practice, and intermediate positions that take some of the machinery and not the rest. Map that structure (a small cluster of related positions and how they relate), not a debate with two seats.

A. Variables vs propositions in causal/cognitive mapping#

  • The cognitive-mapping and cause-mapping traditions the paper names: Axelrod (structure of decision, signed cognitive maps), Eden & Ackermann (cause maps, problem structuring, Decision Explorer), Laukkanen (comparative causal mapping, standardisation), Maule, Markóczy & Goldberg, Hodgkinson. Steelman why they code variables/polarity and what is lost without it.
  • Fuzzy Cognitive Maps (Kosko and successors): weights, feedback, simulation. The clearest opposite of minimalism.
  • Whether "factor as proposition" has named precedents (event / situation semantics, Davidsonian events, propositional vs variable representations of belief).

B. Theories of causation behind the "influence, not determination / causal power" language#

  • The paper repeatedly uses "causal power". Map this to the dispositional/powers tradition (Mumford & Anjum; Cartwright on capacities) and to Patricia Cheng's causal power theory in psychology (this is a near-exact terminological match and must be addressed).
  • Contrast traditions the paper implicitly declines: counterfactual (Lewis; and the interventionist Woodward / Pearl line, causal Bayesian networks and DAGs), regularity (Hume), probabilistic, and mechanistic/process theories.
  • Whether "influence not determination" and "left-open counterfactual" can be defended (or is open to attack) against the interventionist orthodoxy that a causal claim is a counterfactual/intervention claim.

C. Force dynamics, linguistics and the psychology of causal language#

  • Talmy's force dynamics (already cited in Despite-claims); Wolff's force/dynamics model of causation and the CAUSE/ENABLE/PREVENT distinction; Dowty and the linguistics of causatives; lexical causatives and the cause/enable/prevent triad. This directly underpins claims 4 and 12 (enablers/blockers) and the "going viral" idiom point.
  • Pragmatics of causal explanation and causal selection: Hilton's conversational model, Hesslow, and the philosophical cause/condition distinction (Hart & Honoré, Causation in the Law) — directly relevant to why speakers name one cause and not others, and to claim 6 (absence) and claim 12 (enablers).

D. Necessity, sufficiency, conjunctural causation (the 10% the paper sets aside)#

  • Mackie's INUS conditions; Rothman's causal pies (epidemiology); QCA / set-theoretic methods (Ragin, already nodded to) and calibration; the necessity/sufficiency machinery the paper refuses. Steelman: when configurational/conjunctural coding is worth the cost.
  • Distinguish this fairly from the minimalist claim that such structure is rare in ordinary narrative.

E. Realist evaluation: context and mechanism#

  • Pawson & Tilley CMO; the Lemire et al. review of "what is a mechanism"; analytical-sociology mechanisms (Hedström & Swedberg; Machamer, Darden & Craver). Steelman the realist insistence that context and mechanism are not just more factors. Map the minimalist counter (claim 11).

F. Causal inference from narrative in evaluation (the method neighbours)#

  • Contribution analysis (Mayne); process tracing and Bayesian process tracing (Beach & Pedersen; Befani & Stedman-Bryce); outcome harvesting (Wilson-Grau); QuIP (Copestake); contribution rubrics (Aston). Where each places the evaluative judgement relative to coding — this is exactly the "deferred judgement" debate (claim 8).
  • The transitivity/aggregation problem (claim 10): how these and the cognitive-map-aggregation literature handle combining evidence across people; any precedent for within-source path constraints; ecological-inference style cautions.

G. Coding theory and QDA#

  • Saldaña (note: he defines a "Causation Coding" method explicitly — must engage), Miles & Huberman, Braun & Clarke (thematic analysis), Mayring (qualitative content analysis), grounded theory (Glaser & Strauss; Charmaz), Lincoln & Guba. The presence/absence asymmetry of coding (claim 6) and the "coding is not analysis" critique (already pre-empted in the paper). Steelman the Big-Q objection that minimalist coding decontextualises.

H. Participatory / consensus mapping vs multi-source aggregation#

  • Participatory Systems Mapping (Barbrook-Johnson & Penn), group model building (Vennix), system dynamics / causal loop diagrams and DAGs as "model of the world" approaches. Steelman the consensus-map and whole-system-model positions against claim 9 and the cognition-not-world stance (claim 5).

I. Auditability, scale, and AI/NLP causal extraction#

  • Causal relation extraction / causality mining in NLP and argument mining: does the minimalist "extract candidate links with quotes" stance line up with how the NLP field frames causal extraction? Any standards for auditable extraction. Supports claim 13.

A single markdown literature paper, roughly 4,000–7,000 words, structured by the clusters A–I above. For each cluster:

  • Steelman: the strongest opposing position, stated fairly and in its own terms, with the best one to three citations.
  • Allies / precursors: who already holds the minimalist position or a near-neighbour, so we can cite rather than reinvent.
  • The live debate: what is actually contested, and where the minimalist paper lands.
  • Hook back: one line on how this should change or strengthen the minimalist paper (a claim to soften, a term to adopt, an objection to pre-empt, a citation to add).

End with a short synthesis: the three or four places where the minimalist paper is most exposed, and the three or four where it is on stronger ground than it currently claims.

  • No invented citations, quotes, page numbers or findings. Every reference must be real and verifiable. Where a claim cannot be sourced, say so rather than filling the gap. Prefer works already in the Causal Map Zotero library / MyLibrary.bib where they exist, and flag new ones to add.
  • Use the keys already in the minimalist paper where the same work recurs (e.g. axelrodStructureDecisionCognitive1976, raginMeasurementCalibrationSetTheoretic2008, pawsonRealisticEvaluation1997, lemireWhatThisThing2020, talmyForceDynamicsLanguage1988, milesQualitativeDataAnalysis2014, saldanaCodingManualQualitative2015, braunThematicAnalysisPractical2021, mayringQualitativeContentAnalysis2000, lincolnNaturalisticInquiry1985, charmazConstructingGroundedTheory2014, barbrook-johnsonParticipatorySystemsMapping2022).
  • British English; no boosterism, no em/en dashes; follow the garden writing rules.
  • Stay tied to the paper's own arguments. Background context only where it changes how the minimalist paper should be written. Resist a general philosophy-of-causation survey.
  • Distinguish clearly throughout between modelling cognition/claims and modelling the world (the Janus distinction), since most cross-tradition disagreement traces back to which of these the author thinks a map is for.
  • Re-explaining the Causal Map app, filters, or pipeline mechanics (covered in A formalisation of causal mapping and Causal mapping as causal QDA).
  • General LLM/NLP capability claims beyond the narrow causal-extraction framing.
  • Defending minimalism: the job is to map the field fairly, including where minimalism is weak.