Skip to content

GROUNDING.md

Latest commit

 

History

History
39 lines (27 loc) · 1005 Bytes

GROUNDING.md

File metadata and controls

39 lines (27 loc) · 1005 Bytes

The Problem: Syntax vs. Semantics

Correct code is not the same as idiomatic code.

LLMs reliably hit:

  • Syntax ✅
  • Functionality ✅

They frequently miss:

  • Idioms ❌
  • Language "feel" ❌
  • Architectural intent ❌

The Uncanny Valley of Code

Consider this Python loop:

for i in range(len(arr)):
    something(arr[i])

It runs. It passes tests. But an experienced Python developer would write:

for item in arr:
    something(item)

The first example fights the language. The second works with it. This instinct—knowing what rejected options look like—is what AI lacks.

The Missing Context

LLMs have training data, but they lack an authoritative map of what experienced practitioners reject. This knowledge lives in:

  • Senior engineers' heads
  • Code reviews
  • Implicit team culture

AI Lint makes this implicit knowledge explicit. It provides a map of "technically correct but wrong" patterns to help AI choose the best expression for a given language.