{System name}

Type: note · Status: current · Tags: related-systems

{One-paragraph summary — what it is, what it's for, who built it.}

Repository: {URL}

Core Ideas

{What the system does and why. Focus on design choices, not feature lists. Each idea should be a bolded phrase followed by explanation — makes scanning easier.}

Comparison with Our System

{Where we align, where we diverge, and why the divergences matter. Tables work well for dimension-by-dimension comparisons. Name the trade-offs honestly — neither system is uniformly better.}

Borrowable Ideas

{Concrete things we could adopt — structures, patterns, framings, mechanisms. For each, note what it would look like in our system and what would need to change. Mark whether it's ready to borrow now or needs a use case first.}

Curiosity Pass

After drafting the sections above, re-read the report with fresh eyes:

  • What surprises you? What triggers your curiosity?
  • Where is the cost/benefit of a design choice not obvious — what's the simpler alternative that achieves the same result?
  • For each strong claim, ask: what could this mechanism actually achieve, even if it works perfectly?

Then, for each bolded phrase in Core Ideas, apply these questions:

  1. What property does this claim to produce? Name the benefit explicitly (constraining, verification, isolation, etc.).
  2. Does the mechanism transform the data, or just relocate it? If the input format equals the output format (e.g. markdown in, markdown out), the claimed property (codification, compilation, distillation) may be naming, not mechanism. Check the source.
  3. What's the simpler alternative that achieves the same result? If a simpler mechanism produces identical behaviour, the complexity is pointless.
  4. What could this mechanism actually achieve, even if it works perfectly? Some mechanisms have a ceiling — e.g. symbolic checks cannot meaningfully verify freeform prose, regardless of implementation quality.

Investigate mechanistically. Update Core Ideas and Comparison with what you find.

{This step catches claims that sound impressive but don't survive inspection. The first pass is a broad generator (curiosity, cost/benefit, oracle-strength). The second pass forces systematic coverage — each Core Idea, not just the ones that seem interesting — and checks for representation-change illusions (mechanisms that claim transformation but just move data). See kb/work/curiosity-prompts/experiment-report.md for the experiment behind this step.}

What to Watch

  • {How might this system evolve in ways that affect our design?}
  • {What experiments are they running that we'd learn from?}

Relevant Notes: