S1: integrate EEG methods scope and verify merged CI

.github/workflows/ci.yml

@@ -103,6 +103,44 @@ jobs:
         # covered socket-free by MuseOSCDecoderTests / MindMonitorDecoderTests,
         # which still run. Run the socket integration test locally.
 
+      # The step above names five targets. The package declares nine, and the
+      # app-level fail-closed, runtime-presentation, cancellation, and headless
+      # runtime-factory regressions live in two it never selected:
+      # NeuralComposeAppTests (116 tests) and DialecticSessionTests (5).
+      #
+      # Assert they are discoverable before running them. A `--filter` that
+      # matches nothing exits 0, so a renamed, moved, or emptied target would
+      # otherwise turn into another green step that ran nothing.
+      - name: Assert A2 suites are discoverable
+        run: |
+          swift test list > /tmp/neuralcompose-tests.txt
+
+          # Sentinel classes, one per target.
+          grep -q "AppViewModelRuntimeFailClosedTests" /tmp/neuralcompose-tests.txt \
+            || { echo "::error::AppViewModelRuntimeFailClosedTests not discoverable"; exit 1; }
+          grep -q "RuntimeFactoryClaudeResolutionTests" /tmp/neuralcompose-tests.txt \
+            || { echo "::error::RuntimeFactoryClaudeResolutionTests not discoverable"; exit 1; }
+
+          # Sentinels alone would still pass if the rest of a target vanished,
+          # so floor each target's discovered count too. Floors are set below
+          # the current 116/5 so ordinary test additions never trip them.
+          app=$(grep -c '^NeuralComposeAppTests\.' /tmp/neuralcompose-tests.txt)
+          session=$(grep -c '^DialecticSessionTests\.' /tmp/neuralcompose-tests.txt)
+          echo "discovered: NeuralComposeAppTests=$app DialecticSessionTests=$session"
+          [ "$app" -ge 100 ] \
+            || { echo "::error::NeuralComposeAppTests discovered $app, expected >= 100"; exit 1; }
+          [ "$session" -ge 5 ] \
+            || { echo "::error::DialecticSessionTests discovered $session, expected >= 5"; exit 1; }
+
+      # RuntimeIdentityPresentationTests is deliberately NOT asserted here: it
+      # arrives with PR #32. Add it as a third sentinel when that lands, so the
+      # guard stays true at every revision rather than failing on this base.
+      - name: Test (A2 app and session suites)
+        run: |
+          swift test \
+            --filter NeuralComposeAppTests \
+            --filter DialecticSessionTests
+
   # Tests/eval had no CI at all before this job. Seven modules, and only
   # test_adr_references.py ran (via the direct python3 step above) — the other
   # six executed nowhere, which is how three failing assertions and an
@@ -133,6 +171,15 @@ jobs:
           # contract depends on SciPy's exact/asymptotic behaviour. An
           # unconstrained install would let a major bump change a p-value
           # without any commit here.
+          #
+          # torch and the NeuralComposeEEG requirements come first because one
+          # module here, test_local_open_weight_review.py, imports that
+          # package's test helpers, and those import torch. The CPU wheel is
+          # named explicitly: the default Linux torch pulls a ~800MB CUDA build
+          # for a runner with no GPU. The bounded pins go last so they decide
+          # the numpy and scipy versions.
+          pip install --index-url https://download.pytorch.org/whl/cpu torch
+          pip install -r NeuralComposeEEG/requirements.txt
           pip install \
             'pytest>=8,<9' \
             'numpy>=2,<3' \
@@ -152,9 +199,9 @@ jobs:
           echo "discovered: Tests/eval=${collected:-0} collection_errors=${errors:-0}"
           [ "${errors:-0}" -eq 0 ] \
             || { echo "::error::${errors} module(s) failed to collect"; exit 1; }
-          # Floor sits below the current 63 so ordinary additions never trip it.
-          [ "${collected:-0}" -ge 55 ] \
-            || { echo "::error::Tests/eval collected ${collected:-0}, expected >= 55"; exit 1; }
+          # Floor sits below the current 137 so ordinary additions never trip it.
+          [ "${collected:-0}" -ge 125 ] \
+            || { echo "::error::Tests/eval collected ${collected:-0}, expected >= 125"; exit 1; }
 
       - name: Test (eval suites)
         run: |
@@ -163,21 +210,53 @@ jobs:
           python -m pytest Tests/eval -v \
             --ignore=Tests/eval/test_joint_embedding.py
 
-      # NeuralComposeEEG is not on main yet — it arrives with lane B. Absent is
-      # expected and must be said out loud; present-but-empty is a defect, since
-      # a package whose tests silently collect nothing is the failure this whole
-      # job exists to prevent.
-      - name: Assert NeuralComposeEEG is absent or non-vacuous
+  # NeuralComposeEEG had no CI at all until this job, which is how a test
+  # importing an uncommitted module (neuralcompose_eeg.structured_state) reached
+  # a pull request unnoticed: build-and-test above is Swift-only.
+  #
+  # Linux, not macOS: nothing here is platform-specific, and the runner is both
+  # cheaper and faster.
+  #
+  # This job covers NeuralComposeEEG only. Tests/eval runs in python-eval above,
+  # which bounds the numpy and scipy majors its assertions depend on. Running
+  # it here too would test it twice, the second time in an unbounded
+  # environment.
+  python-contracts:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+          cache: pip
+          cache-dependency-path: NeuralComposeEEG/requirements.txt
+
+      - name: Install Python dependencies
+        run: |
+          # CPU wheel explicitly: the default Linux torch pulls a ~800MB CUDA
+          # build for a runner with no GPU. Installed first so the pinned
+          # requirements below see torch>=2.0 as already satisfied.
+          pip install --index-url https://download.pytorch.org/whl/cpu torch
+          pip install -r NeuralComposeEEG/requirements.txt
+          # pytest is here to count what is collectable, for the guard below.
+          pip install 'pytest>=8,<9'
+
+      # A package whose tests silently collect nothing is the failure this
+      # job exists to prevent, so count before running.
+      - name: Assert NeuralComposeEEG is non-vacuous
         run: |
-          if [ ! -d NeuralComposeEEG ]; then
-            echo "NeuralComposeEEG/ absent on this branch — expected until lane B integrates it"
-            exit 0
-          fi
-          echo "NeuralComposeEEG/ present — its suite must not be vacuous"
           n=$(PYTHONPATH=NeuralComposeEEG/src python -m pytest NeuralComposeEEG/tests \
                 --collect-only -q 2>/dev/null \
                 | grep -oE '^[0-9]+ tests? collected' | grep -oE '^[0-9]+' | head -1)
           echo "discovered: NeuralComposeEEG=${n:-0}"
           [ "${n:-0}" -gt 0 ] \
-            || { echo "::error::NeuralComposeEEG/ is present but collected 0 tests"; exit 1; }
-          PYTHONPATH=NeuralComposeEEG/src python -m pytest NeuralComposeEEG/tests -v
+            || { echo "::error::NeuralComposeEEG/ collected 0 tests"; exit 1; }
+
+      # Offline scientific package: canonical dataset contracts, capture
+      # manifests, and the structured-state shadow bridge. Discovered rather
+      # than enumerated so new contract suites are picked up automatically.
+      - name: Test (NeuralComposeEEG contracts)
+        run: |
+          PYTHONPATH=NeuralComposeEEG/src \
+            python -m unittest discover -s NeuralComposeEEG/tests -t . --verbose

.gitignore

@@ -53,6 +53,10 @@ venv.*.parked/
 __pycache__/
 *.pyc
 
+## Rust utility crates
+target/
+**/target/
+
 ## Synthetic data outputs (regeneratable from Scripts/generate_synthetic_sleep.py)
 Data/*
 !Data/.gitkeep

Evaluation/scripts/analyze_results.py

@@ -14,9 +14,13 @@
 import json
 import math
 import statistics
+import sys
 from pathlib import Path
 
 import numpy as np
+
+sys.path.insert(0, str(Path(__file__).parent))
+from eval_stats import cohens_d
 import pandas as pd
 
 REPO_ROOT = Path(__file__).resolve().parent.parent.parent
@@ -39,38 +43,6 @@ def bootstrap_ci(data, confidence=0.95, n_boot=10000):
     return (lo, hi)
 
 
-def cohens_d(a, b):
-    """Cohen's d effect size (pooled SD)."""
-    a, b = np.array(a, dtype=float), np.array(b, dtype=float)
-    if len(a) < 2 or len(b) < 2:
-        return float("nan")
-    # Degrees-of-freedom-weighted pooled variance. The unweighted
-    # (var_a + var_b) / 2 form coincides with this only when the groups are the
-    # same size, and this function accepts arbitrary lengths — so every
-    # unequal-n caller was dividing by the wrong denominator.
-    n_a, n_b = len(a), len(b)
-    pooled_variance = (
-        (n_a - 1) * a.var(ddof=1) + (n_b - 1) * b.var(ddof=1)
-    ) / (n_a + n_b - 2)
-    pooled_std = math.sqrt(pooled_variance)
-    mean_difference = a.mean() - b.mean()
-    if pooled_std == 0:
-        # PROJECT POLICY, not a universal definition. With both samples constant
-        # the denominator is zero and Cohen's d is strictly undefined. We report
-        # the limiting value — signed infinity — because 0.0 claimed "no effect"
-        # for the most extreme separation representable, and nan would discard
-        # the direction. Identical constants are the one genuinely zero case.
-        #
-        # Callers that PERSIST this must convert at the boundary: JSON has no
-        # Infinity literal (`json.dumps` emits a bare `Infinity`, which strict
-        # parsers reject) and CSV has no convention at all. Decide on a
-        # representation before writing it, rather than assuming the reader copes.
-        if mean_difference == 0:
-            return 0.0
-        return float("inf") if mean_difference > 0 else float("-inf")
-    return float(mean_difference / pooled_std)
-
-
 def safe_mean(values):
     return float(np.nanmean(values)) if len(values) > 0 else float("nan")
 

Evaluation/scripts/embedding_analyze.py

@@ -27,6 +27,9 @@
 
 import numpy as np
 from scipy import stats as sp_stats
+
+sys.path.insert(0, str(Path(__file__).parent))
+from eval_stats import cohens_d
 from sklearn.cluster import KMeans
 from sklearn.preprocessing import StandardScaler
 
@@ -50,37 +53,6 @@ def bootstrap_ci(data, confidence=0.95, n_boot=10000):
     return (lo, hi, len(data))
 
 
-def cohens_d(a, b):
-    a, b = np.array(a, dtype=float), np.array(b, dtype=float)
-    if len(a) < 2 or len(b) < 2:
-        return float("nan")
-    # Degrees-of-freedom-weighted pooled variance. The unweighted
-    # (var_a + var_b) / 2 form coincides with this only when the groups are the
-    # same size, and this function accepts arbitrary lengths — so every
-    # unequal-n caller was dividing by the wrong denominator.
-    n_a, n_b = len(a), len(b)
-    pooled_variance = (
-        (n_a - 1) * a.var(ddof=1) + (n_b - 1) * b.var(ddof=1)
-    ) / (n_a + n_b - 2)
-    pooled_std = math.sqrt(pooled_variance)
-    mean_difference = a.mean() - b.mean()
-    if pooled_std == 0:
-        # PROJECT POLICY, not a universal definition. With both samples constant
-        # the denominator is zero and Cohen's d is strictly undefined. We report
-        # the limiting value — signed infinity — because 0.0 claimed "no effect"
-        # for the most extreme separation representable, and nan would discard
-        # the direction. Identical constants are the one genuinely zero case.
-        #
-        # Callers that PERSIST this must convert at the boundary: JSON has no
-        # Infinity literal (`json.dumps` emits a bare `Infinity`, which strict
-        # parsers reject) and CSV has no convention at all. Decide on a
-        # representation before writing it, rather than assuming the reader copes.
-        if mean_difference == 0:
-            return 0.0
-        return float("inf") if mean_difference > 0 else float("-inf")
-    return float(mean_difference / pooled_std)
-
-
 # Above this per-group size the exact Mann-Whitney null distribution is too
 # expensive to enumerate, so the asymptotic approximation is used. Stated here
 # rather than inherited from SciPy's `auto` threshold, which is a library

Evaluation/scripts/embedding_space_analysis.py

@@ -35,6 +35,13 @@
 
 
 def cka(X, Y):
+    """Biased linear CKA over centered Gram matrices.
+
+    Null values can be substantially inflated when feature dimension approaches
+    or exceeds sample count — independent Gaussians at n=50, d=128/256 score
+    ~0.78, not ~0. Do not compare scores across different n/d regimes without a
+    matched null or permutation baseline.
+    """
     X = np.array(X, dtype=np.float64)
     Y = np.array(Y, dtype=np.float64)
     X = X - X.mean(axis=0)

Evaluation/scripts/eval_stats.py

@@ -32,7 +32,12 @@ def bootstrap_ci(data, confidence=0.95, n_boot=10000):
 
 
 def cohens_d(a, b):
-    """Cohen's d effect size between two samples."""
+    """Cohen's d effect size between two samples.
+
+    Returns NaN where d is undefined rather than a finite value that would rank
+    alongside real effects: under-sized samples, and zero pooled variance with
+    separated means. Only equal constant groups are a true zero.
+    """
     a, b = np.array(a, dtype=float), np.array(b, dtype=float)
     if len(a) < 2 or len(b) < 2:
         return float("nan")

Evaluation/scripts/statistical_analysis.py

@@ -31,6 +31,9 @@
 
 import numpy as np
 from scipy import stats as sp_stats
+
+sys.path.insert(0, str(Path(__file__).parent))
+from eval_stats import cohens_d
 from sklearn.cluster import KMeans
 from sklearn.preprocessing import StandardScaler
 
@@ -55,38 +58,6 @@ def bootstrap_ci(data, confidence=0.95, n_boot=10000):
     return (lo, hi, len(data))
 
 
-def cohens_d(a, b):
-    """Cohen's d effect size with pooled SD."""
-    a, b = np.array(a, dtype=float), np.array(b, dtype=float)
-    if len(a) < 2 or len(b) < 2:
-        return float("nan")
-    # Degrees-of-freedom-weighted pooled variance. The unweighted
-    # (var_a + var_b) / 2 form coincides with this only when the groups are the
-    # same size, and this function accepts arbitrary lengths — so every
-    # unequal-n caller was dividing by the wrong denominator.
-    n_a, n_b = len(a), len(b)
-    pooled_variance = (
-        (n_a - 1) * a.var(ddof=1) + (n_b - 1) * b.var(ddof=1)
-    ) / (n_a + n_b - 2)
-    pooled_std = math.sqrt(pooled_variance)
-    mean_difference = a.mean() - b.mean()
-    if pooled_std == 0:
-        # PROJECT POLICY, not a universal definition. With both samples constant
-        # the denominator is zero and Cohen's d is strictly undefined. We report
-        # the limiting value — signed infinity — because 0.0 claimed "no effect"
-        # for the most extreme separation representable, and nan would discard
-        # the direction. Identical constants are the one genuinely zero case.
-        #
-        # Callers that PERSIST this must convert at the boundary: JSON has no
-        # Infinity literal (`json.dumps` emits a bare `Infinity`, which strict
-        # parsers reject) and CSV has no convention at all. Decide on a
-        # representation before writing it, rather than assuming the reader copes.
-        if mean_difference == 0:
-            return 0.0
-        return float("inf") if mean_difference > 0 else float("-inf")
-    return float(mean_difference / pooled_std)
-
-
 # Above this per-group size the exact Mann-Whitney null distribution is too
 # expensive to enumerate, so the asymptotic approximation is used. Stated here
 # rather than inherited from SciPy's `auto` threshold, which is a library