A water utility replaces its published tariff PDFs and needs the resident bot to stop citing outdated rates. What should the grounding team do to the search index?

When a water utility replaces published tariff PDFs, the grounding team must monitor and rebuild or refresh the search index so the resident bot stops citing outdated rates. Stale vectors and chunks continue to retrieve old prices until the corpus and index are updated to match the newly published tariffs. Reconfiguring the semantic ranker and scoring profile for tariff PDF queries only reorders existing results and does not remove stale rate vectors from the index, so the bot can still retrieve outdated prices and erode trust in municipal rate answers. Ingesting the new tariff PDF to the index with revised metadata filters only changes how results are filtered and leaves both the old and new rates retrievable side by side unless the superseded document is removed, so the bot can still cite the outdated figure even after the refresh. Increasing vector search top-k and enabling semantic captions for rate queries retrieves more candidates and better summaries but still surfaces stale tariff passages instead of fixing the retrieval source of truth. Exam caveat: corpus change for published municipal rates requires index monitor-and-refresh—not semantic ranker and scoring retuning, top-k and caption tweaks, or ingesting with revised metadata filters. Operational check: after each tariff PDF publish, invalidate or reingest affected documents, rebuild or incremental-refresh the index, query known old versus new rate strings side by side, and gate the resident bot on a freshness check that confirms only current tariff document versions are retrievable before answering rate questions.