Style Contract

The post should read like the existing notes: direct, explanatory, and technically adult. It should not read like a LinkedIn essay, a policy memo, a manifesto, or an academic literature review.

Target Voice

  • Plain sentences.
  • Substantial paragraphs with internal flow.
  • Claims stated through mechanisms, examples, and consequences.
  • Few abstractions.
  • No ornate transitions.
  • No excitement words.
  • No rhetorical inflation.

The post can be opinionated, but each opinion must rest on a concrete mechanism.

Local Style From Existing Posts

The nearby technical posts usually do the following:

  • Start with an object or problem.
  • Define the relevant quantities before interpretation.
  • Use examples to make the object concrete.
  • Keep derivations and details available, but do not sell them.
  • Use “For example” often and honestly.
  • Avoid grand claims in the opening.

For this post, the analogous object is not a formula. It is the feedback loop by which AI systems improve: candidate generation, evaluation, and selection.

Sentence Rules

  • Start from the object: verifier, proof, theorem, question, community, student, institution, feedback loop.
  • Vary sentence length.
  • Let each paragraph develop one idea rather than stacking isolated observations.
  • Use transitions that carry logic: “For example”, “In that case”, “The consequence is”, “This matters because”.
  • Prefer “can be checked cheaply” over “verifiable register”.
  • Prefer “questions that matter” over “significance landscape”.
  • Prefer “hard to score” over “open-objective domain” unless the term has already been defined.
  • Prefer “proofs that nobody digests” over “epistemically inert artifacts”.
  • Use “AI” as a broad term, but specify the task whenever possible.

Avoid Staccato Prose

The post should not read like a sequence of aphorisms. Avoid paragraphs made from many short declarative sentences of the same shape. Do not turn the scaffold fields into prose one by one.

Each section paragraph should have a small internal arc:

  1. state the object or mechanism;
  2. give the example;
  3. draw the consequence;
  4. hand off to the next section.

The paragraph should feel like one developed thought, not a bullet list with punctuation.

Avoid

  • “Framework”, unless referring to a named technical framework in a source.
  • “Paradigm”, “lens”, “grammar”, “blueprint”, “taxonomy”, “ecosystem”.
  • “Key insight”, “central idea”, “crucially”, “it is important to note”.
  • “Not merely X but Y” unless X is a real misconception already introduced.
  • “The future of mathematics” as a standalone dramatic phrase.
  • “AI will change everything” and similarly broad sentences.
  • “Humanity”, “civilization”, “epistemic crisis”, unless the sentence is made concrete.
  • Long lists of fields without explaining the feedback cost in each.

Preferred Rewrites

Instead of:

AI changes the epistemic infrastructure of mathematics.

write:

AI can increase the number of proved statements faster than the community can turn them into usable understanding.

Instead of:

The key insight is that taste remains orthogonal to verification.

write:

A proof checker can certify a proof. It cannot certify that the theorem was worth proving.

Instead of:

The verifiable register will be automated while open-objective cognition remains scarce.

write:

Tasks with cheap checks will move faster than tasks where nobody can cheaply say what counts as a good answer.

Paragraph Test

Each paragraph must do at least one of the following:

  • define a mechanism;
  • give an example;
  • state a consequence;
  • state a limitation;
  • connect one section to the next.

Delete or rewrite paragraphs that mainly sound polished. Also rewrite paragraphs that sound mechanically assembled from the blueprint.