diff --git a/vowpalwabbit/conditional_contextual_bandit.cc b/vowpalwabbit/conditional_contextual_bandit.cc
index f4a4957e949..165e2057450 100644
--- a/vowpalwabbit/conditional_contextual_bandit.cc
+++ b/vowpalwabbit/conditional_contextual_bandit.cc
@@ -44,8 +44,8 @@ void return_v_array(v_array<T>&& array, VW::v_array_pool<T>& pool)
 
 struct ccb
 {
-  vw* all;
-  example* shared;
+  vw* all = nullptr;
+  example* shared = nullptr;
   std::vector<example*> actions, slots;
   std::vector<uint32_t> origin_index;
   CB::cb_class cb_label;
@@ -53,17 +53,17 @@ struct ccb
   namespace_interactions generated_interactions;
   namespace_interactions* original_interactions;
   std::vector<CCB::label> stored_labels;
-  size_t action_with_label;
+  size_t action_with_label = 0;
 
   multi_ex cb_ex;
 
   // All of these hashes are with a hasher seeded with the below namespace hash.
   std::vector<uint64_t> slot_id_hashes;
-  uint64_t id_namespace_hash;
+  uint64_t id_namespace_hash = 0;
   std::string id_namespace_str;
 
-  size_t base_learner_stride_shift;
-  bool all_slots_loss_report;
+  size_t base_learner_stride_shift = 0;
+  bool all_slots_loss_report = false;
 
   VW::v_array_pool<CB::cb_class> cb_label_pool;
   VW::v_array_pool<ACTION_SCORE::action_score> action_score_pool;
@@ -653,7 +653,7 @@ void save_load(ccb& sm, io_buf& io, bool read, bool text)
 
 base_learner* ccb_explore_adf_setup(options_i& options, vw& all)
 {
-  auto data = scoped_calloc_or_throw<ccb>();
+  auto data = VW::make_unique<ccb>();
   bool ccb_explore_adf_option = false;
   bool all_slots_loss_report = false;
 
@@ -698,12 +698,15 @@ base_learner* ccb_explore_adf_setup(options_i& options, vw& all)
   data->id_namespace_str.append("_id");
   data->id_namespace_hash = VW::hash_space(all, data->id_namespace_str);
 
-  learner<ccb, multi_ex>& l = init_learner(data, base, learn_or_predict<true>, learn_or_predict<false>, 1,
-      prediction_type_t::decision_probs, all.get_setupfn_name(ccb_explore_adf_setup), true);
-
-  l.set_finish_example(finish_multiline_example);
-  l.set_save_load(save_load);
-  return make_base(l);
+  auto* l = VW::LEARNER::make_reduction_learner(std::move(data), base, learn_or_predict<true>, learn_or_predict<false>,
+      all.get_setupfn_name(ccb_explore_adf_setup))
+                .set_learn_returns_prediction(true)
+                .set_prediction_type(prediction_type_t::decision_probs)
+                .set_label_type(label_type_t::ccb)
+                .set_finish_example(finish_multiline_example)
+                .set_save_load(save_load)
+                .build();
+  return make_base(*l);
 }
 
 bool ec_is_example_header(example const& ec) { return ec.l.conditional_contextual_bandit.type == example_type::shared; }
diff --git a/vowpalwabbit/learner.h b/vowpalwabbit/learner.h
index 4b7d0f4e00e..a7d92692b68 100644
--- a/vowpalwabbit/learner.h
+++ b/vowpalwabbit/learner.h
@@ -817,8 +817,22 @@ struct reduction_learner_builder
     set_params_per_weight(1);
 
     this->_learner->pred_type = base->pred_type;
-    // TODO add label type as something learner knows about itself.
-    // this->_learner.label_type = label_type;
+    // TODO add label type as something learner knows about itself, this will enable more type checking and better
+    // description of the learner. this->_learner.label_type = label_type;
+  }
+
+  reduction_learner_builder<DataT, ExampleT, BaseLearnerT>& set_prediction_type(prediction_type_t pred_type)
+  {
+    this->_learner->pred_type = pred_type;
+    return *this;
+  }
+
+  reduction_learner_builder<DataT, ExampleT, BaseLearnerT>& set_label_type(label_type_t label_type)
+  {
+    // TODO add label type as something learner knows about itself, this will enable more type checking and better
+    // description of the learner. this->_learner.label_type = label_type;
+    std::ignore = label_type;
+    return *this;
   }
 
   reduction_learner_builder<DataT, ExampleT, BaseLearnerT>& set_params_per_weight(size_t params_per_weight)