quiz · 4 min Memory & State Check Without any memory mechanism, what is true of the agent loop?The loop is entirely stateless; each decision call receives only what you explicitly hand itThe loop automatically retains the last three turns via the runtimeThe model caches prior outputs internally, so the loop is stateful by defaultThe loop stores state in the action layer between turnsWhich statement best describes working memory as defined in the lessons?An external database that persists user preferences across sessionsA vector store used to retrieve relevant facts from previous runsA long-term log shared across all agent invocations in a systemThe running transcript of the current run, scoped to a single invocation, that disappears when the process endsIn the bounded-transcript Memory class from Lesson 19, what happens when the transcript exceeds its budget?The entire transcript is cleared and a fresh one is startedThe oldest turn is evicted and appended to a running summary; the policy only ever sees context()The policy receives the full unbounded log and decides what to dropThe runtime automatically truncates the newest turns to stay within budgetWhat is the correct role of a summary in the memory system described in Lesson 20?Summaries replace the raw turn log once compression is completeSummaries are lossless and can fully reconstruct any discarded turnSummaries are a rendering concern -- they shape what the model sees, while the raw log remains the source of truthSummaries must be stored in long-term memory to avoid context bloatWhat mitigation does Lesson 20 recommend for stale long-term memory facts?Timestamp every long-term write, attach a validity scope, and treat retrieved facts as hypotheses to verifyRe-read the same facts from working memory at the start of each turn to refresh themDelete any long-term fact older than one session to keep memory currentStore facts only in working memory to avoid staleness across runsSubmit Mark complete ← Memory Pitfalls Decompose, Then Act →