_toon_parse_header returns None for malformed headers, but it calls to_number(ls) without validating that ls is non-empty and numeric. Since to_number raises a runtime error on invalid strings, a malformed header like []: or [abc]: will error instead of cleanly returning None. Consider checking ls with a numeric regex (or similar) before converting, and returning None when it’s not a valid length.

active_delim is passed through _toon_parse_recursive but is never actually used (it’s only forwarded to a nested call). This makes the parsing logic harder to follow; either use it (e.g., as the default delimiter for inline arrays) or remove the parameter and thread the chosen delimiter explicitly where needed.

This assert_eq compares a full JSON string with a specific key order. Because dict iteration order is based on interned Ident ordering (global interner), adding/importing unrelated modules can change key ordering and force churn in this expected string (as seen here). To make the test robust, consider asserting on parsed structure (e.g., json::json_parse(result) then spot-check key paths) or sorting keys in the serializer used for tests.

In _toon_stringify_dict, the array-special-casing concatenates ks + ind + value when value starts with [. For nested dicts ind is non-empty, which will insert extra spaces between the key and the [n]:... header (e.g., key [3]: ...). This breaks round-tripping and formatting. Consider concatenating without ind (or with a single fixed separator if needed) so nested arrays format as key[3]: ... consistently.

_toon_parse_primitive unescapes \\n/\\r/\\t before unescaping \\\\. This mis-parses sequences like "\\\\n" (intended to become \\n) because the \\n replacement will fire on the tail and produce a newline. The unescape logic should handle escaped backslashes first or use a single-pass escape parser to avoid overlapping replacements.

_toon_parse_inline_array infers delim by checking whether the bracket content contains | or a tab, but then still uses the raw bc as the length string (to_number(bc)). For syntax like [3|] / [3\t], to_number will fail because bc includes the delimiter marker. Strip any delimiter marker from bc (similar to _toon_parse_header) before converting and before using it as the reported length.