When the authoritative collection metadata path is enabled, CollectionCacheRecoverer::recoverCollectionFromDisk rebuilds the routing table from the shard catalog by calling RoutingTableHistory::makeNew. That constructor sets _allowGaps = false, so the ChunkMap validates that the first chunk's min is MinKey and the last chunk's max is MaxKey.

However, the write side (commit_collection_metadata_locally::updateShardCatalogCache and fetchOwnedChunks) intentionally persists only the chunks owned by the local shard, and constructs its in-memory RoutingTableHistory via RoutingTableHistory::makeNewAllowingGaps.

As a result, whenever the local shard does not own chunks at one or both extremes of the key space, authoritative disk recovery uasserts with ChunkMetadataInconsistency (code 431): "Not all elements of { ... } are of type maxKey/minKey".