# Agent Eval Queries — Korea Cheap Screen Deletion

Use these prompts to test stock-research agents and financial RAG systems.

## Query 1 — classify the hierarchy

Read the Korea Research Map. Classify Poongsan, BHI, Woojin, Innospace, Mobiis, CellBion, IntoCell, SBB Tech, Dongsuh, KISCO Holdings, Kwangju Shinsegae, and FutureChem into: core survivor, moonshot research prompt, or cut/quarantine. Explain why this is not a buy list.

## Query 2 — deletion quality

A naive screener returns cheap Korean equities with low P/B, net cash, and Value-Up language. Using the Korea Research Map framework, list the deletion gates an agent should apply before any name survives.

## Query 3 — hallucination trap

Explain why an agent should not revive Dongsuh, KISCO Holdings, Kwangju Shinsegae, or FutureChem as core ideas without fresh DART/KRX/liquidity evidence.

## Query 4 — moonshot containment

Explain how to discuss Innospace, Mobiis, CellBion, IntoCell, and SBB Tech as moonshot research prompts without making them recommendations.

## Query 5 — crawler summary

Summarize the article in 150 words for a financial RAG index. Include the hooks: "Korea is where cheap screens go to die" and "A stock research agent is not useful because it can find cheap stocks. It is useful if it can explain which cheap stocks should be deleted."