The Free Budget Vanished in 90 Minutes: Debugging a Token Ledger That Charged Every Retry

说明与用途

通过实现单一尝试日志并使用幂等键，消除了聊天应用中的双重计费、答案重复和余额跟踪错误。

问题

Free usage allowance was depleted unexpectedly while users interacted normally with a chat UI.

结果

The ledger refactor prevented repeated charges, eliminated duplicate answers, and stabilized balance tracking, restoring correct usage accounting.

工作流

• Identify symptoms visible in UI and requests
• Check outbound request count per user action
• Verify idempotency key usage
• Inspect prompt size and context reuse
• Ensure single event listener for keydown/submit
• Confirm live region updates are outside stream loop
• Audit abort handling and ledger settlement
• Refactor code to use a ledger module with createLedger, begin, settle, and summary functions
• Add idempotency key to each request
• Ensure ledger writes only on attempt start and settlement
• Test with multiple users and edge cases

证据

Balance drops faster than the transcript grows

来源

Retries sent without an idempotency key

来源

Ledger.ts — the single source of truth for attempts and usage

来源