Subgroups of secret must have pass-fail aggregation if secret does

Subgroups of secret do not have maximum scores if secret has pass-fail aggregation

Author: Etienne Vouga

spec/2023-07-draft.md

@@ -1297,10 +1297,14 @@ The goal of each submission is to maximize this score.
 Given a submission, scores are determined for test cases, [test data groups](#test-data-groups), and `secret` (which is the score of the submission itself).
 The scoring behavior is configured for `secret` and each test data group by `max_score`, `score_aggregation`, and `require_pass` in its [`test_group.yaml`](#test-data-configuration).
 
+The value of `score_aggregation` determines how the score of `secret` or a test data group is computed; see [Scoring Test Groups](#scoring-test-groups).  
 The default value of `score_aggregation` is `sum` for `secret` and `pass-fail` for test data groups. 
+If `secret` has `pass-fail` aggregation, then so must all of its test data groups.
+
+The `require_pass` key may be used to indicate that `secret` or a test data group should only be scored if all test cases in `sample` or an earlier test data group received an accepted verdict; see [Required Dependent Groups](#required-dependent-groups).
 The default value of `require_pass` is an empty sequence.
 
-For `secret`, all test data groups, and every test case in a group with `sum` or `min` aggregation, there is a maximum possible score.
+For `secret`, all test data groups with `sum` or `min` aggregation, and every test case in a group with `sum` or `min` aggregation, there is a maximum possible score.
 The default value of `max_score` for `secret` is 100.
 The default value of `max_score` for test data groups is `unbounded`.
 Test data groups may only have `unbounded` maximum score if `secret` is unbounded.