feat: integrate advanced math solver and complete automated testing pipeline" -m "- Task 1: Wrote unit tests to verify the internal math translator.

.github/workflows/ci.yml

@@ -0,0 +1,76 @@
+name: CI
+
+on:
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - main
+
+jobs:
+  test:
+    name: Run all test suites (Python ${{ matrix.python-version }})
+    runs-on: ubuntu-latest
+
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version: ["3.10", "3.11", "3.12"]
+
+    steps:
+      # ── 1. Checkout ────────────────────────────────────────────────────────
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      # ── 2. Python setup ────────────────────────────────────────────────────
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+          cache: pip
+
+      # ── 3. Install runtime dependencies ───────────────────────────────────
+      - name: Install runtime dependencies
+        run: pip install -r requirements.txt
+
+      # ── 4. Install test dependencies ───────────────────────────────────────
+      # pytest     — test runner
+      # httpx      — required internally by starlette.testclient.TestClient
+      # These are not in requirements.txt (they are dev/test only).
+      - name: Install test dependencies
+        run: pip install pytest httpx
+
+      # ── 5. Unit tests ──────────────────────────────────────────────────────
+      # Tests tokenizer/slang_serializer.py round-trip.
+      # No network, no solver, no env vars needed.
+      - name: Run unit tests
+        run: pytest tests/unit/ -v --tb=short
+        env:
+          GROQ_API_KEY: ""
+
+      # ── 6. Integration tests ───────────────────────────────────────────────
+      # Tests POST /solve against the live Starlette app using TestClient.
+      # GROQ_API_KEY must be empty so FallbackSolver is always selected.
+      - name: Run integration tests
+        run: pytest tests/integration/ -v --tb=short
+        env:
+          GROQ_API_KEY: ""
+
+      # ── 7. Regression tests ────────────────────────────────────────────────
+      # Loads every .json fixture from tests/regression/fixtures/ and
+      # runs solver.solve() against it, comparing to the golden output.
+      # GROQ_API_KEY must be empty so FallbackSolver is always selected.
+      - name: Run regression tests
+        run: pytest tests/regression/ -v --tb=short
+        env:
+          GROQ_API_KEY: ""
+
+      # ── 8. Full suite run (all three together) ─────────────────────────────
+      # Runs all suites in one pytest process to catch any cross-suite
+      # singleton leakage that the per-suite runs would not detect.
+      # conftest.py's autouse fixture handles singleton reset between tests.
+      - name: Run full test suite
+        run: pytest tests/ -v --tb=short
+        env:
+          GROQ_API_KEY: ""

api/_shared.py

@@ -136,18 +136,31 @@ def get_solver():
             _solver_error = str(exc)
             print(f"[CalculusSolver] Groq load failed: {exc}", flush=True)
 
-    # 2. Fall back to deterministic solver (pure Python, no model)
-    from inference.fallback_solver import FallbackSolver
-
-    _solver = FallbackSolver()
-    _solver_mode = "fallback"
-    if not _solver_error:
-        _solver_error = "No GROQ_API_KEY provided. Falling back to deterministic solver."
-    print(
-        "[CalculusSolver] Running in FALLBACK mode — "
-        "supports diff, partial, integrate, gradient, tangent_line.",
-        flush=True,
-    )
+    # 2. Load B's real solver
+    try:
+        from inference.solve import CalculusSolverInference
+        _solver = CalculusSolverInference()
+        _solver_mode = "inference"
+        _solver_error = None
+        print(
+            "[CalculusSolver] B's solver loaded successfully.",
+            flush=True,
+        )
+    except Exception as exc:
+        _solver_error = str(exc)
+        print(f"[CalculusSolver] B's solver load failed: {exc}", flush=True)
+        # Final fallback — always available
+        from inference.fallback_solver import FallbackSolver
+
+        _solver = FallbackSolver()
+        _solver_mode = "fallback"
+        if not _solver_error:
+            _solver_error = "B's solver failed to load. Using deterministic fallback."
+        print(
+            "[CalculusSolver] Running in FALLBACK mode — "
+            "supports diff, partial, integrate, gradient, tangent_line.",
+            flush=True,
+        )
     return _solver, _solver_mode
 
 
@@ -220,7 +233,7 @@ def fraction_to_latex(expr: dict) -> str:
 
 def normalize_solver_result(result: dict, mode: str) -> dict:
     """Normalize/unwrap solver output into the standard API response format."""
-    if mode == "neural":
+    if mode in ("neural", "inference"):
         output = result.get("output") or {}
         if isinstance(output, dict) and "expr" in output:
             expr = output["expr"]
@@ -235,10 +248,10 @@ def normalize_solver_result(result: dict, mode: str) -> dict:
             "steps": steps,
             "latex": latex,
             "confidence": float(result.get("confidence", 0.0)),
-            "verified": result.get("verified"),
+            "verified": bool(result.get("verified")),
             "warning": result.get("warning"),
             "rule": result.get("rule"),
-            "mode": "neural",
+            "mode": mode,
         }
     else:
         # Fallback and Groq solver results already have the correct structure

conftest.py

@@ -0,0 +1,45 @@
+"""
+Root conftest.py — shared fixtures for all test suites.
+
+Provides:
+  reset_solver_singleton — autouse fixture that clears the api._shared
+  solver singleton before every test. This prevents solver state from
+  leaking between test modules when pytest runs all suites in one process.
+
+  Explicitly removes GROQ_API_KEY from the environment so that
+  get_solver() always falls through to FallbackSolver in CI and locally.
+"""
+
+import os
+import pytest
+
+
+@pytest.fixture(autouse=True)
+def reset_solver_singleton():
+    """
+    Clear the solver singleton and remove GROQ_API_KEY before every test.
+
+    autouse=True means this runs automatically for every test in every
+    test file, with no need to import or reference it explicitly.
+
+    Yields control to the test, then does nothing on teardown — the
+    singleton stays reset for the next test because the next invocation
+    of this fixture will reset it again before that test runs.
+    """
+    # Remove Groq key first so that any import of api._shared triggered
+    # below does not accidentally instantiate GroqSolver
+    os.environ.pop("GROQ_API_KEY", None)
+
+    # Import here (not at module level) to avoid circular import issues
+    # if conftest is loaded before the package is fully on sys.path
+    try:
+        import api._shared as _shared
+        _shared._solver = None
+        _shared._solver_mode = "unloaded"
+        _shared._solver_error = None
+    except ImportError:
+        # api._shared not importable yet (e.g. during unit-only runs
+        # where the api package is not needed). Safe to ignore.
+        pass
+
+    yield