AI for QA / Quiz

Probe the Retrieval Seams

Diagnose boundary splits, embedding-space drift, rank cutoffs, and lexical blind spots with targeted fixtures.

Difficulty
Hard
Format
Quiz
Points
200
Estimate
10 min

// MISSION BRIEF

Your Mission

A RAG pipeline misses a policy answer even though the source was ingested. Use the stored chunks and retrieval trace to distinguish source presence from retrievability, then choose probes that expose failures before production.

// FIRST CONTACT

Battle teaser

Stored with chunk_size=80 and overlap=0: Chunk 41 ends: "After the fifth failed sign-in attempt, the account" Chunk 42 begins: "is locked for 15 minutes. Administrators cannot shorten it." Query: "lockout duration after five failures". What risk is shown?

  1. AAny overlap value above zero would delete both chunks
  2. BA boundary split separated the triggering condition from its duration, weakening each chunk's standalone meaning
  3. CThe source contains no lockout rule
  4. DThe model must always answer from training instead of retrieval
Answers, scoring, hints, and the full battle stay sealed.

// SKILL TAGS

rag-builderchunkingembeddingsretrievalai-for-qa