Evaluate long deliveries separately from long inputs
A four-part task plan and runnable synthetic ledger keep missing runs, retries, truncation and human handoffs in evaluation records.
Read articleBlog
LLM API
Track agent request costs and validate JSON before accepting results
Two offline Python examples show how to preserve unknown tool charges and reject JSON records that parse correctly but disagree with their source.
Read article
Preserve tool-call history across a JSON save and restore
A runnable Python example checks what survives when tool-call history is saved to JSON and restored, including assistant fields and tool result IDs.
Read article
API key concurrency: model one account-wide admission budget
A runnable offline example shows how two API keys can exceed one shared account concurrency budget when each key admits requests independently.
Read article