Why did weekly operational reporting take two days?
Weekly operational reporting took most of two days to assemble, and by the time it landed the operators had already noticed the same problems on the floor. The reporting wasn't wrong — it was late, and it described what happened without saying why.
How do you explain a deviation without the model inventing numbers?
The deviation detection itself is statistical rather than generative: thresholds and trend breaks computed over the operational tables. The language model does the part it is genuinely good at — pulling together the related signals behind a flagged deviation and writing the explanation a shift lead can read in ninety seconds.
How does the briefing reach the operations team?
The briefing pipeline reads from the existing PostgreSQL replica and the warehouse management exports, posts the finished briefing into the operations Slack channel, and links each figure to the Grafana panel it came from, so anyone reading it can open the underlying series and check the numbers themselves.
How is every figure in the briefing verified?
Every number in the briefing is traceable to a query, and the queries live in version control and are reviewed like any other code. The model never computes a figure — it only explains figures the pipeline computed, which keeps the worst failure mode at 'unhelpful narrative' rather than 'wrong number in a report'.
What changed for the operations standup?
Preparation for the weekly operational report dropped to roughly an hour, down from an assembly job that took the better part of two days. The daily briefing also changed the standup itself: it now starts from the explanation instead of from someone reading numbers off a screen. Deviation detection stays statistical, with thresholds and trend breaks computed over the operational tables, while the language model only assembles the related signals behind a flagged deviation and writes the ninety-second explanation a shift lead can act on. Every number in the briefing traces back to a query, the queries live in version control and are reviewed like any other code, and each figure links to the Grafana panel it came from. The model never computes a value, which keeps the worst failure mode at an unhelpful narrative instead of a wrong number in a report. That discipline also governs the tender pipeline, and it shapes how we scope document and data work.