A query like ts.aggregate({$group: {_id: "$meta", m: {$min: "$control.min.value"}}}) should be optimized to take advantage of the bucket's control data instead of unpacking the buckets.

When all queries against TS namespace are forced to run in classic, both ts.aggregate({$group: {_id: "$meta", m: {$min: "$value"}}}) and system.buckets.ts..aggregate({$group: {_id: "$meta", m: {$min: "$control.min.value"}}}) generate the following plan:

"stages" : [
	{
		"$cursor" : {
			"queryPlanner" : {
				"namespace" : "my.system.buckets.ts",
				"indexFilterSet" : false,
				"parsedQuery" : {
	},
	"queryHash" : "0D1EB9EE",
	"planCacheKey" : "0D1EB9EE",
	"maxIndexedOrSolutionsReached" : false,
	"maxIndexedAndSolutionsReached" : false,
	"maxScansToExplodeReached" : false,
	"winningPlan" : {
		"stage" : "PROJECTION_DEFAULT",
		"transformBy" : {"control.min.value" : 1, "meta" : 1, "_id" : 0},
		"inputStage" : {"stage" : "COLLSCAN", "direction" : "forward"}
	},
	"rejectedPlans" : [ ]	}}},
	{ "$group" : {"_id" : "$meta", "m" : {"$min" : "$control.min.value"}}}
],

But if SBE lowered is allowed for TS namespaces, the query against the view only lowers the access (PROJECTION_DEFAULT + COLLSCAN) but not GROUP while the query against the backing collection is lowered fully (GROUP + COLLSCAN). This needs to be investigated because not lowering the $group properly is likely to affect performance.