Investigate AI-agent analysis report on the load-balancers specification. This ticket tracks findings identified during a systematic review of the spec: missing test coverage, ambiguous normative language, and spec/test inconsistencies.

Missing Tests

• Drivers MUST throw an exception if the connection string contains more than one host with loadBalanced=true
  (§URI Validation)
• killCursors MUST be executed on the same pinned connection when cursor is closed (§Behaviour With Cursors)
• If getMore fails with a network error, the connection MUST remain pinned to the cursor (§Behaviour With Cursors)
• Two concurrent transactions from different sessions MUST NOT share a connection (§Behaviour With Transactions)
• Connection pool MUST track connection purpose in three categories: cursors, transactions, other (§Connection
  Tracking)
• WaitQueueTimeoutError MUST include a detailed message with connection breakdown (§Connection Tracking) — error
  message format not validated
• Pre-handshake errors MUST NOT perform SDAM error handling (§Initial Handshake Errors)
• Pool clearing after handshake errors MUST be scoped to serviceId only (§Initial Handshake Errors)

Ambiguities

• loadBalanced=false vs. omitted: Spec says default is false but doesn't clarify whether driver code
  distinguishes explicit false from omitted.
• SRV lookup timing for validation: Drivers doing async SRV lookup may only discover the host count error during
  first operation, not at MongoClient construction. Spec acknowledges this but doesn't define the required timing.
• logicalSessionTimeoutMinutes in LB mode: Spec says ignore it, but if the server returns a non-null value, should
  the driver warn or silently ignore?
• Concurrent next() and close() on pinned cursor: Spec allows either documenting that it's unsupported OR
  preventing it — drivers may choose differently.

Inconsistencies

• Static topology vs. dynamic pool changes: Monitoring section describes a "static" topology, but Error Handling
  section allows dynamic pool clearing scoped to serviceId. Interaction between a static topology and dynamic pool
  changes is underspecified.
• Cursor pinning vs. transaction pinning: Both sections describe pinning, but the interaction when a transaction
  opens a cursor is not clarified. Does transaction pinning override cursor pinning?
• transactions.yml vs. cursors.yml: Both test pinning but the interaction (transaction that opens a cursor) is not
  tested.

Notes

• 8 unified test files (cursors, event-monitoring, connection-establishment, server-selection, sdam-error-handling,
  transactions, wait-queue-timeouts).
• serviceId isolation for pool clearing is tested in sdam-error-handling.yml but coverage is sparse.
• Read preference: LoadBalancer servers MUST be treated like mongos — tested in server-selection.yml but only basic
  read preference types.