Recursive Schema Validation

Given a JSON-like data structure and a schema definition, verify that the data conforms (types, required fields, nested shapes). An FAR onsite coding round emphasizing OOP design and edge-case coverage. Requirements Inputs: arbitrary nested data (dict | list | primitive) and a schema describing expected types, required keys, and nested structures. Return whether the data validates; on failure, optionally return the failing path. Support common schema constructs: primitive types, lists with item schema, dicts with required/optional keys, nullable fields. Examples schema = { "type": "object", "properties": { "name": {"type": "string"}, "tags": {"type": "array", "items": {"type": "string"}} }, "required": ["name"] } data = {"name": "foo", "tags": ["a", "b"]} validate(data, schema) -> True Notes Use dataclasses or a small class hierarchy to model Schema variants (ObjectSchema, ListSchema, PrimitiveSchema) and dispatch validation via methods. The original interview was a Principal SDE round — clean class structure was scored explicitly. Candidates routinely run out of time covering all edge cases; budget heavily on additionalProperties, missing required keys, type coercion (int vs float), and nullability. Reference standard: JSON Schema draft-07 semantics are the implicit target. You don't need full coverage — pick a representative subset and document it. Useful class skeleton: a Schema ABC with validate(data, path) -> list[Error] and three subclasses — ObjectSchema(properties, required, additional_properties), ArraySchema(items, min_items, max_items), PrimitiveSchema(type, nullable). Each subclass owns its own edge cases; the dispatcher just routes by type field. Error reporting: return a list of (json_pointer, message) tuples and let the caller decide between "first error" or "all errors". Resist booleanizing — graders penalize loss of failure context. Coverage menu to declare upfront: required keys, additionalProperties, type checks (including int vs float), nullability, array items, nested objects. State which you will implement and which you will stub; explicit scoping signals seniority. Preparation Implement a mini JSON Schema validator on your own machine ahead of the loop — even 100 lines is enough to internalize the recursion shape. Build a 5-edge-case checklist (missing required, extra unknown, wrong type, null in non-nullable, mixed list items). Practice splitting design time vs coding time — Amazon Staff-level rounds penalize a sloppy class hierarchy even when the logic is correct. Write a 100-line subset validator (object + array + primitive + required + nullable) at home, then exercise it against three hand-crafted invalid payloads. The recursion shape sticks after one full implementation. Practice walking through the design on paper before coding — the Principal SDE round scores class-hierarchy decisions and naming as much as it scores running code. Pre-think the follow-up "how do you support $ref?" — answer: a SchemaRegistry keyed on $id, resolved lazily inside ObjectSchema.validate to support cycles.