# DECISIONS — shared (append-only)

## 2026-06-15 — Token representation

Used a dataclass with `kind: str` and `value` (Any). This lets NUMBER store int or float, and other tokens store the character string. Simple and sufficient for the parser phase.

## 2026-06-15 — Number parsing

Integers → int, floats (containing `.`) → float. Handles `.5`, `10.`, `3.14`.

## 2026-06-15 — Advisory: multi-dot number strings

`tokenize("1.2.3")` produces a bare `ValueError` from `float()` rather than a `LexError` because the greedy digit+dot scanner consumes the whole string before conversion. DoD only requires LexError for character-level invalids, so this is not a phase-lex defect. Parser phase should guard against malformed numeric literals if needed.