Use of AI tools

Coding assistants (Claude and OpenAI Codex) supported the Python implementation, source research, model checks, tests and written reports. Their contributions covered the data adapters, Dash views, forecasting code, Pydantic schema, extraction workflow and documentation.

DeepSeek deepseek-flash extracted economic events from 14 archived ISW/CTP reports. The run covered 520 narrative paragraphs in 71 chunks. Automated semantic review checked economic relevance and attribution after schema and evidence validation. This is an extraction experiment rather than a human-labelled accuracy benchmark.

Verification includes offline replay, live source acquisition, baseline comparisons, timing and schema tests, independent forecast recalculation, browser checks and PDF review. The executed notebook, reports/validation.json and the Q4 report contain the results and methods needed to explain and reproduce the work.