[Content outdated] "Structured Data" lesson teaches prefill + stop sequences, superseded by structured outputs #161

Course

Skilljar version: "Claude with the Anthropic API"
https://anthropic.skilljar.com/claude-with-the-anthropic-api

Section: "Accessing Claude with the API"
Lesson: "Structured Data"

Problem

The "Structured Data" lesson teaches structured JSON extraction using the
prefilled assistant message + stop sequence technique (prefilling the
assistant turn with ```json and setting ``` as a stop
sequence).

This approach is no longer the current best practice and conflicts with the
feature Anthropic now recommends for this exact use case:

1. Structured outputs are now the purpose-built, recommended solution
   for getting valid JSON, and are not covered by the course. The
   output_config.format parameter with a JSON schema uses constrained
   decoding to guarantee schema-valid output ("Always valid: No more
   JSON.parse() errors ... no retries needed"). It is generally available
   for Claude Opus 4.7, Opus 4.6, Sonnet 4.6, Sonnet 4.5, Opus 4.5, and
   Haiku 4.5.
   Ref: https://platform.claude.com/docs/en/build-with-claude/structured-outputs

2. The technique taught is explicitly incompatible with structured
   outputs. The structured outputs documentation lists "Message
   Prefilling" as incompatible with JSON outputs. So the method the lesson
   teaches cannot be combined with the modern recommended approach, and
   learners are being taught a pattern that the current API design steers
   away from.

Suggested fix

Update the "Structured Data" lesson to teach structured outputs
(output_config.format / json_schema) as the primary method, and present
prefill + stop sequences (and tool use) as historical or alternative
approaches with their current limitations clearly noted.

Related

Part of the broader pattern of course material lagging behind the current
API. See #71 (dynamic model name selection in course notebooks).