Round

Hot Loop (Round 1+) — 40% of Cold Start targets

Procedure

Round State Ledger

This is the top-level progress tracker. If a round ends with little delta (< 10% improvement on any metric), escalate: the next round should target the weakest metric specifically. </HARD-GATE>

Cold Start (Round 0)

1. Call lit-survey strategy with full iteration budget (10 iterations)
2. (P1) Call gap-analysis strategy — STUB: extract gaps from lit-survey output
3. (P2) Call idea-generation strategy — STUB: placeholder
4. (P1) Call review — STUB: placeholder, return score 0

Hot Loop (Round 1+)

1. Read review feedback from previous round
2. Determine which stages need re-running based on review
3. Call only needed strategies with reduced iteration budgets:
   • lit-survey: 3 iterations (hot)
   • gap-analysis: 2 iterations (hot)
   • idea-generation: 2 iterations (hot)
4. Call review again

State Inheritance

• All state (knowledge, papersRead, urlsVisited) carries forward between rounds
• Each round extends, never replaces, accumulated state

P0 Scope Note

In P0, only lit-survey is fully implemented. Gap-analysis and idea-generation are stubs that return placeholder outputs. The round strategy skeleton is created now so P1/P2 can fill in the real implementations without restructuring.

Output

{
  "roundNumber": 0,
  "surveyResult": {...},
  "gaps": [...],
  "ideas": [...],
  "reviewScore": 0,
  "reviewFeedback": "...",
  "state": { "knowledge": "...", "papersRead": [...], "urlsVisited": [...] }
}