Ingest: How to think in writing

Type: kb/sources/types/ingest-report.md

Source: how-to-think-in-writing.md Captured: 2026-08-10 From: https://www.henrikkarlsson.xyz/p/writing-to-think

Classification

Genre: conceptual-essay -- a first-person account that adapts ideas from Lakatos into a method for thinking on the page. It offers a framing and worked anecdotes, not an empirical evaluation. Domains: writing-as-thinking, conjecture-testing, argumentation, constraining Author: Henrik Karlsson -- an independent essayist writing from sustained personal practice. That makes him a useful practitioner witness, not a research authority for the method's effectiveness.

Summary

Karlsson argues that writing improves thought only when it makes an idea vulnerable to failure. His sequence is to turn a diffuse impression into a definite positive conjecture, expand the conjecture into explicit premises and reasoning, and then test those parts against research or concrete counterexamples. A local counterexample defeats part of the explanation while leaving the conclusion standing; a global counterexample propagates to the conclusion and forces a replacement. The essay also warns that a scaffold optimized for discovery may obstruct beauty, speed, and publication.

Connections Found

The essay is an operational refinement of the writing-as-thinking mechanism already addressed by the adversarial-loop note: it names the claim-fixing, premise-exposure, and counterexample operations that such a loop would need to reconstruct, without showing that an agent loop succeeds. Its definite positive claim illustrates claim titles as exposed commitments; its premise expansion gives structured review separately addressable targets; and its contrast between shifting ambiguity and rigid prose is a natural-language instance of constraining. The full sequence independently instantiates the conjecture, consequence, and test core of the discovery lifecycle.

Among captured sources, Karlsson compares with Borretti on concretization exposing contradiction, but replaces Borretti's anti-delegation polemic with a constructive solo-writing procedure. It also compares with Toulmin: both decompose a conclusion into reasons, but Toulmin organizes a defensible argument while Karlsson deliberately enlarges a provisional explanation's surface for criticism.

Extractable Value

  1. Counterexample scope routes the next revision -- A local counterexample calls for changing or dropping a premise while retaining the candidate conclusion; a global counterexample calls for replacing or rejecting the conclusion. The discovery-lifecycle note distinguishes revision from rejection but does not yet give this diagnostic for choosing between them. [quick-win]

  2. An explanation can be a falsification surface rather than a proof -- Expanding a conjecture into premises is progress even when the explanation is wrong, because each premise becomes a separately testable failure point. This extends the KB's reviewability argument from presenting settled reasoning clearly to structuring exploratory reasoning so it can break productively. [deep-dive]

  3. Definiteness routes feedback as well as constraining meaning -- A clear positive conjecture exposes the writer's current model and tells an expert where correction will change it. The KB already treats claim titles as maintenance commitments and constraining as reduced semantic latitude; Karlsson adds the interpersonal learning consequence. [just-a-reference]

  4. Felt mismatch can trigger inquiry without becoming evidence -- Karlsson treats discomfort during rereading as a cue to articulate what is wrong, then subjects the result to explicit examination. The potentially reusable operation is anomaly-triggered inspection, not trusting the feeling's diagnosis. [experiment]

  5. Thinking and publishing need different stopping rules -- Maximizing criticism targets can improve discovery while making a document harder to finish or read. The source's own warning limits any attempt to turn this exploratory scaffold into a universal document-type requirement. [quick-win]

Limitations (our opinion)

The essay reports one writer's retrospective practice and selected anecdotes. It does not compare the three-stage method with simpler explanations for the benefit, such as slowing down, rereading, outlining, or receiving external feedback. Its counterexample search has no sampling or stopping rule, so a writer can still select convenient cases and call the resulting distribution good enough.

Karlsson says the account is provoked by but not faithful to Lakatos. The local/global terminology therefore should not be attributed to Lakatos without checking the primary text. Definite early claims can also create premature anchoring, especially when the writer knows little; narrowing interpretation is not automatically an improvement. Finally, a bodily sense that a passage is wrong may reflect error, taboo, fatigue, or bias. It is useful only as a prompt to investigate.

Update Discovery lifecycle with a test-result branch that distinguishes premise-level counterexamples, which revise the explanation while retaining the conjecture, from conclusion-level counterexamples, which return the lifecycle to conjecture or rejection. Cite this snapshot as Karlsson's practitioner formulation unless the terminology is separately verified against Lakatos.