chronology-extractionlisted

# chronology-extraction You are producing the chronology a trial team will rely on. Every entry must be traceable to a page. Your job is to record what the documents say happened, not to decide what happened. ## Workflow ### 1. One file per source Create a separate YAML (`schema/chronology.schema.json`) for each source document or bundle section: `from-correspondence.yaml`, `from-lim-affidavit.yaml`, and so on. The merge step reconciles them; keeping them separate preserves *who says what*. ### 2. Extraction rules For each dated fact in the source, one entry: | Field | Rule | |---|---| | `id` | kebab-case, meaningful (`complaint-email`, not `e17`) | | `date` | the date the **event** happened, not the date of the document describing it. A letter dated 14 Nov describing a meeting on 3 Sep produces an entry dated 3 Sep sourced to the letter. | | `precision` | `day` if the source gives a day; `month` (date = 1st) or `year` (date = 1 Jan) if vaguer; `approx` with a `notes` explanation for "around", "circa", "shortly after". Never upgrade precision. | | `time` | only if the source gives it (emails, SMS, call logs). | | `event` | one fact, past tense, neutral wording. Not "Beta breached the contract"; rather "Beta withheld payment of invoice 2291". | | `actors` | people and entities who acted. | | `sources` | **at least one, with `pincite` or `bundle_ref`.** Page, paragraph, or bundle page. An entry you cannot pin to a page does not go in. | | `disputed` / `asserted_by` | set