feat: implement reproducible PROTEAN baseline pipeline

configs/experiments/experiment_protocol.yaml

@@ -1,6 +1,8 @@
 # Frozen experiment protocol for the beta-weighted prototype aggregation study.
-# Values below are the required research settings. Seed set is provisional until
-# the verified baseline convention is reproduced (see docs/baseline/).
+#
+# ALL values below are now FROZEN for the baseline reproduction (2026-09-24 session).
+# They must remain identical across every aggregation variant, beta candidate and
+# seed so that comparisons are paired. See docs/research/EXPERIMENT_PROTOCOL.md.
 
 datasets:
   - xiiotid
@@ -8,35 +10,38 @@ datasets:
 
 non_iid:
   partition: dirichlet
+  # Frozen Dirichlet algorithm: for each class, draw proportions ~ Dirichlet(alpha)
+  # over clients, then multinomial-allocate that class's train samples. See
+  # docs/research/EXPERIMENT_PROTOCOL.md section 2.
+  algorithm: per_class_dirichlet_multinomial
   alphas: [0.75, 0.50, 0.25]        # required by the research specification
   num_clients: 10                    # matches PROTEAN paper cross-silo setting
 
 aggregation:
-  # beta candidates under test (one beta is applied to all classes and clients)
-  betas: [0.00, 0.25, 0.50, 0.75, 1.00]
-  # Modes:
-  #   original_protean : verified baseline semantics (NOT YET VERIFIED)
-  #   support_weighted : w_ij = N_ij ** beta, normalized over S_j  (proposed)
-  #   equal_available  : equal weights over S_j (methodological control,
-  #                      introduced only if the baseline participant rule differs)
-  primary_mode: support_weighted
+  # Baseline checkpoint: only the reproduced original-PROTEAN aggregation is
+  # implemented. Support-weighted beta candidates and the equal_available control
+  # are DEFERRED to the later beta/controls phase (see docs/research/EXPERIMENT_PROTOCOL.md).
   baseline_mode: original_protean
-  control_mode: equal_available
 
 training:
   rounds: 10
   local_epochs: 3
-  # mu is stated in the paper; lambda is NOT stated in the paper text (UNRESOLVED).
-  mu: 0.1
-  lambda: null
-  batch_size: null                   # to be fixed after baseline reproduction
-  learning_rate: null                # to be fixed after baseline reproduction
+  mu: 0.1                            # model-alignment coefficient (paper, frozen)
+  lambda: 1.0                        # prototype-alignment coefficient (PROJECT DECISION; paper gives no value)
+  batch_size: 256                    # PROJECT DECISION; paper gives no value
+  learning_rate: 0.001               # PROJECT DECISION; paper gives no value
+  optimizer: adam                    # PROJECT DECISION; paper gives no value
+  model_aggregation: equal_1_over_m  # paper Eq. (2)/(5.5)
+  prototype_computation: end_of_round_mean_embedding
+  prototype_alignment_form: per_sample_embedding_to_global_prototype  # FedProto-style reading of Eq. (3)
 
 evaluation:
   # The paper uses 80/20 train/test and has no validation split.
   # This study ADDS a validation split used only for beta selection.
-  test_fraction: 0.20
-  validation_fraction: null          # finalize after baseline reproduction
+  test_fraction: 0.20                # paper
+  validation_fraction: 0.10          # PROJECT DECISION (fraction of the full cleaned dataset)
+  train_fraction: 0.70               # remainder
+  split_strategy: stratified_by_class
   selection_metric: rare_class_macro_f1
   rare_class_rule: two_smallest_positive_attack_support
   report_metrics:
@@ -45,12 +50,18 @@ evaluation:
     - overall_macro_f1
     - accuracy
     - per_client_metrics
+    - per_class_precision_recall_f1
+    - unseen_class_report
 
 seeds:
-  # Paper averaged over three random seeds. Exact seed values are UNRESOLVED;
-  # they must be fixed before any beta comparison and shared across all modes.
-  provisional: [0, 1, 2]
-  status: provisional
+  # The paper averages over three random seeds but names none, so the seed values
+  # are NOT source-confirmed. {0, 1, 2} is an explicit PROJECT DECISION chosen as
+  # the minimal three-seed set, and is identical across every baseline condition.
+  # Each seed deterministically controls the split, Dirichlet partition, model
+  # initialisation and batch order (see src/protean_research/utils/seeding.py).
+  frozen: [0, 1, 2]
+  rationale: minimal_three_seed_set_paper_reports_three_unnamed_runs
+  source_status: project_decision_not_source_confirmed
 
 artifacts:
   partition_manifest_dir: artifacts/partitions