☏ Support automation

The biggest bot here reads every closed support chat and transcribed phone call and writes a claims row: what went wrong, with which product, and what we did about it. The support team now corrects a cell or two instead of typing fifteen. Getting there took more rules than code.

Start with the claims-log post for the pipeline and the edit log that measures it. The twenty-rules post is every rule the bot learned from a human correction. The request-detector post is the other direction: spotting customers who want a document or a number, and why a quarter of keyword hits turned out not to be requests. The polling post is why the bot asks the chat platform once a minute instead of waiting for webhooks, with a live check of chats the platform's list never returns. The phone-call post is what speech-to-text gets wrong, and the photo post is how the bot picks evidence from the order instead of guessing.