101-04 — Debug a failure

What you will build

Two intentional Err paths: bad JSON from fake_chat and schema-invalid inputs—still fully offline.

Prerequisites

• 101-02, 101-03

Files you will touch

None (inline Python only).

Step 1 — Invalid JSON from fake chat

Reuse the 101-02 scaffold but return ChatResult(True, "{not json") from fake_chat. Run runner.run with the same valid input dict as before.

Expected: Err describing parse / LLM failure (exact type varies by task plumbing).

Step 2 — Schema mismatch input

Keep fake_chat valid JSON, but call runner.run("pw_chunk_classify", "v1", {}) (empty dict).

Expected: Err tied to missing required fields per input_schema.

Step 3 — Read the error surface

• Print r and inspect structured failure payload on Err.
• Cross-link taxonomy: Repair loops, Verification.

Expected output

• Step 1 yields an Err from malformed assistant JSON (offline).
• Step 2 yields an Err listing missing required input_schema keys.
• You can name whether each failure is primarily parse vs schema layer.

Common failures

Verify

You can explain which layer failed (parse vs schema) for each experiment.

What changed

You trust Ok/Err as first-class outcomes—not exceptions—for governed tasks.

Next step

Continue to Tutorial 201 when you need to author tasks, sidecars, operators, and RAG patterns.