I analysed the JSONL session logs from 8 Claude Code sessions across a day of building a side project. I expected planned sessions to be more efficient, but not 3xhttps://seanfloyd.dev/blog/claude-code-post-mortem
42 validation errors in one run. Claude apologising instead of writing HTML. OAuth tokens expiring mid-digest.Then I fixed the constraints. Eight days, zero failures, zero intervention.The secret wasn't better prompts... it was treating the LLM as a constrained function: schema-validated tool calls that reject malformed output and force retries, two-pass architecture separating...