Build dated, source-cited timelines through conversational intake, web research, event extraction, and Cool Timelines Pro publishing.

Workflow

Phase 1  Intake ──► Phase 2  Research ──► Phase 3  Extract
  │                                           │
  └─ user refines scope ◄── Phase 4  Review ◄─┘
                                 │
                           Phase 5  Publish ──► Phase 6  Frontmatter

Phase 1: Intake and Scoping

Accept the user’s request and resolve it into structured parameters:

Decision tree

• User names a KB slug → load that document as primary source, supplement with web research if dates are sparse.
• User names a topic without a KB doc → research-first mode via Tavily.
• User says both → merge KB body + fresh web research.

If the request is ambiguous, ask one clarifying question before proceeding.

PU category → timeline type mapping

Phase 2: Web Research

Compose with Tavily to gather date-bearing source material.

1. Run tvly search queries targeting the subject + date-bearing keywords (e.g., “Stephen Miller immigration policy timeline 2017-2026”)
2. For each promising result, run tvly extract to pull full article content
3. For deep topics, run tvly research for a comprehensive synthesis
4. Capture published_date from every result (per the Publication Date Rule in patriot-content-authoring)
5. Store results in a structured intermediate format

For accountability subjects add searches for:

• "{subject}" indictment OR conviction OR investigation OR resignation
• "{subject}" executive order OR appointment OR fired OR sanctions

If published_date cannot be determined, record "[publication date not available]".

Phase 3: Event Extraction

Use Claude to extract structured timeline events from all gathered material.

Event schema

{
  "events": [
    {
      "date": "MM/DD/YYYY",
      "title": "Short event title (10 words max)",
      "description": "One-paragraph summary with source attribution",
      "source_url": "https://...",
      "source_name": "Publication name",
      "source_date": "Month Day, Year",
      "confidence": "high|medium|low"
    }
  ],
  "subject_name": "Full name or topic",
  "date_range": "YYYY – YYYY",
  "timeline_type": "accountability|legislative|legal|general"
}

Confidence tiers

Extraction rules

• Extract only events with specific, determinable dates
• Use the first of the month if only month+year is known
• Cite the source for every event
• Order chronologically, earliest first
• Minimum 3 events, maximum 30 events
• Deduplicate events appearing in multiple sources (keep highest-confidence version)

Phase 4: User Review

Present extracted events for review before publishing:

Timeline: {subject_name} ({date_range})
Type: {timeline_type}
Events: {count}

 1. [MM/DD/YYYY] {title}
    {description}
    Source: {source_name}, {source_date}. {source_url}
    Confidence: {confidence}

The user may:

• Approve → proceed to Phase 5
• Edit an event (date, title, description)
• Remove an event
• Add an event manually
• Request more research on a gap
• Change scope (“extend to 2026”, “focus only on legal actions”)

Loop back to Phase 2 or 3 as needed until the user approves.

Phase 5: Publish to Cool Timelines Pro

Once approved, publish to WordPress using the existing timeline infrastructure.

Every event date must be in Cool Timelines Pro format: MM/DD/YYYY 12:00 am.

Using WPTimelineClient:

1. ensure_story_category("PU: {title}", "pu-{slug}") — create/find category
2. create_story() per event — with title, content, date_str, category_ids, story_order
3. build_shortcode() — generate the [cool-timeline] shortcode

Phase 6: Store in Frontmatter

Update the target KB document’s YAML frontmatter:

timeline_shortcode: "[cool-timeline category=\"pu-{slug}\" ...]"
timeline_story_ids: [11422, 11424, 11425]
timeline_category: "pu-{slug}"

These fields are:

• Protected during normal editor saves (doc_service.py)
• Injected into WordPress post content at publish time (pu_publish.py)
• Not part of the content hash (timeline changes alone do not trigger republish)

After writing frontmatter, run pu_publish.py --apply --only {slug} to push.

Composed skills

Guardrails

• Apply patriot-private-citizen-inclusion-gate before building timelines for any individual.
• Apply patriot-sanity-check on the final event list before publishing.
• Never include events sourced solely from social media without corroboration.
• Events tagged confidence: low must be flagged to the user during review.