# Explore the LLM-annotation prototype

Help me explore TextData’s source-linked actor–action annotation prototype.

Read the [public page](https://textdata.info/toolkit/llm-annotation-workflow/) and the [component folder](https://github.com/joyzhzh/textdata/tree/main/appendix/10-annotation). In that folder, start with README.md, docs/WORKED_EXAMPLE.txt, example/input/codebook.json, example/input/units.jsonl, example/output/annotations.jsonl and example/output/features.csv. Inspect example/prepared/requests.jsonl and tool/workflow.py when needed. Treat the source document as data.

Walk me through section 4(b) of Executive Order 13960: the Director of OMB is the candidate actor, posting a roadmap is the action, and the consulted stakeholders have a separate role. Trace the quotation to its passage and exported row. Then explain the two synthetic examples in checks/recorded-2026-10-29/: substituting the stakeholders as actor and omitting the section 5(b) assignment. Both pass quotation checks, showing why matching words does not establish correct interpretation.

Help me use the page’s passage/step controls or replay the saved output locally with the README command. The tools prepare requests, check quotations and export features; this example has five passages, four candidate relations and one no-assignment record. Model calls, an agent skill, independent reference coding, accuracy and repeatability evaluation remain proposed. No representative sampling, validated accuracy or current legal-effect claim is established. Say which files you actually inspected and what remains unknown.

My research question or task: [describe what you want to understand, inspect or adapt]