mirror of
https://github.com/vitorpamplona/amethyst.git
synced 2026-10-06 19:53:08 +00:00
docs: add Verify, Don't Guess standing instruction to CLAUDE.md
This commit is contained in:
@@ -20,6 +20,28 @@ nostrnests reference relay. The IETF code is kept as a reference + unit-test
|
||||
implementation for any future IETF target; see
|
||||
`nestsClient/plans/2026-04-26-moq-lite-gap.md`.
|
||||
|
||||
## Verify, Don't Guess (standing instruction)
|
||||
|
||||
A plausible-sounding explanation is cheap; being right is not. Before
|
||||
asserting what a problem is or how something behaves:
|
||||
|
||||
1. **State hypotheses as hypotheses.** If you haven't run it, say "I'm
|
||||
guessing" or "haven't verified" — never dress an untested guess up as a
|
||||
diagnosis. Use "I verified X by running Y" only when you actually did.
|
||||
2. **Reproduce before diagnosing.** If a claim is checkable in under a
|
||||
minute, check it before stating it. This repo gives you the means:
|
||||
`./gradlew test`, the per-module tests, and `amy` (the CLI exists partly
|
||||
to drive `quartz`/`commons` for interop checks). Write the failing case
|
||||
first, watch it fail, *then* explain. For non-trivial bugs use `/bugfix`
|
||||
(reproduce-first) or `/investigate` (competing hypotheses + refutation).
|
||||
3. **Predict, then run.** Before running a command, state the output you
|
||||
expect. A mismatch is the cheapest signal that your model is wrong.
|
||||
4. **Don't commit to one cause.** A single immediate explanation stops you
|
||||
from looking. Hold 2–3 candidates and a discriminating test for each.
|
||||
|
||||
If you find yourself writing paragraphs to defend a theory, that effort
|
||||
almost always should have been one test.
|
||||
|
||||
## Architecture
|
||||
|
||||
```
|
||||
|
||||
Reference in New Issue
Block a user