Claim Tree + Cumulative Annotation Process

Three core objects:

1. Claim Tree — hierarchical decomposition of a top-level claim (or question) into sub-claims (or sub-questions). No scoring, no attribution. Just structure.
2. Source Document — any external text: paper, blog post, comment thread. The raw material that exists elsewhere in the world.
3. Annotations — extracted fragments from source docs, mapped onto specific nodes in the claim tree, with a link back to origin. AI does the mapping; user confirms/corrects.

The flow:

[Source doc OR seed claim]
       ↓
  [Claim Tree]        ← AI generates from input
       ↓
[New source doc]      ← user brings in external material
       ↓
[AI analysis]         ← "which parts of this doc are relevant to which nodes?"
       ↓
[Annotated Tree]      ← tree nodes now have attached quotes/links
       ↓
[Repeat with more docs] ← cumulative sense-making

Key insight worth naming: The claim tree is a stable scaffold. Documents come and go, but the tree is the persistent structure that discourse accumulates against. That's the cathedral-building idea.

Toy Example Options

Need something: (a) simple enough to grasp fast, (b) substantive enough to be interesting, (c) has existing real material to annotate with.

Candidates:

┌────────────────────────────────────────────┬─────────────────────────────────────────────────────┐
│                   Claim                    │                      Why good                       │
├────────────────────────────────────────────┼─────────────────────────────────────────────────────┤
│ "Coffee improves cognitive performance"    │ Simple, decomposable, tons of studies               │
├────────────────────────────────────────────┼─────────────────────────────────────────────────────┤
│ "Remote work increases productivity"       │ Relatable, contested, lots of blog posts + research │
├────────────────────────────────────────────┼─────────────────────────────────────────────────────┤
│ "Social media harms teenage mental health" │ Hot, decomposable, real papers exist                │
├────────────────────────────────────────────┼─────────────────────────────────────────────────────┤
│ "Reading fiction increases empathy"        │ Surprising, decomposable, studies exist             │
└────────────────────────────────────────────┴─────────────────────────────────────────────────────┘

Suggested: "Does remote work increase productivity?" — relatable to any knowledge worker, genuinely contested, decomposable into clear sub-questions (for whom? what tasks? short vs long term?), and real material (Stanford studies, company blog posts, HBR articles) exists to annotate with.

Visualization Shape

Single scrollable page. Sections reveal as user scrolls down.

Full vision (animated):

• Section 1: Seed question, large and centered
• Section 2: Question animates / reconfigures into tree (nodes appear one by one)
• Section 3: Two-column layout begins — tree locks to LHS (sticky), source doc slides in on RHS
• Section 4: Still two-column — highlighted passages in doc, annotation cards appear on tree nodes as user scrolls

Key UX idea: once the tree is built, it stays visible on the left for all subsequent sections. The right side is where new material arrives and gets processed against the tree. This mirrors the mental model: tree = stable scaffold, right side = incoming world.

Simpler fallback (implemented): Same layout but without the question-to-tree animation. Scroll-triggered fade-ins for tree nodes, doc, and annotations. Sticky tree on LHS during doc + annotation sections.

Toy Example: "Does remote work increase productivity?"

Issue Tree

Does remote work increase productivity?
├── For whom? (what types of workers/roles?)
│   ├── Individual contributors vs managers?
│   └── Deep work roles vs collaborative roles?
├── For what tasks?
│   ├── Focused/solo tasks?
│   └── Creative/collaborative tasks?
├── Over what time horizon?
│   ├── Short-term (first months)?
│   └── Long-term (1+ years)?
└── Under what conditions?
    ├── Home setup quality?
    └── Managerial practices?

Source Document

Nick Bloom (Stanford, 2015) — study of call centre workers at Ctrip.

Key findings:

• 13% productivity gain for remote workers
• Gain was on solo call-handling tasks specifically
• Gain partially reversed when workers chose to return to office

Annotations (AI-generated, mapped to tree nodes)

• "13% productivity gain" → root question + "focused/solo tasks" node — supports remote productivity
• "gain partially reversed on return" → "time horizon" node — complicates (short-term vs long-term distinction matters)
• "call centre workers only" → "for whom" + "what tasks" nodes — limits scope (may not generalise to knowledge workers)