diff --git a/packages/data-analytics-demo/Makefile b/packages/data-analytics-demo/Makefile
index c1f8bd0..b8baafa 100644
--- a/packages/data-analytics-demo/Makefile
+++ b/packages/data-analytics-demo/Makefile
@@ -39,8 +39,7 @@ dashboard:
 	$(PYTHON) -m data_analytics_demo.dashboard.render
 
 semantic-validate:
-	@echo "[semantic-validate] TODO T-10: MetricFlow validation not yet implemented"
-	@exit 1
+	$(PYTHON) -m data_analytics_demo.semantic.validator
 
 demo: data dbt ml narrative dashboard
 	@echo "[demo] full pipeline OK"
diff --git a/packages/data-analytics-demo/pyproject.toml b/packages/data-analytics-demo/pyproject.toml
index 8b6f0b1..a94fbc7 100644
--- a/packages/data-analytics-demo/pyproject.toml
+++ b/packages/data-analytics-demo/pyproject.toml
@@ -40,6 +40,7 @@ dev = [
     "ruff>=0.7",
     "mypy>=1.13",
     "pip-audit>=2.7",
+    "types-PyYAML>=6.0",
 ]
 
 [project.scripts]
@@ -77,7 +78,7 @@ mypy_path = "src"
 # but lags behind `pandas` releases; treating these as untyped is the
 # pragmatic choice for a Python 3.11 + pandas 3.x stack.
 [[tool.mypy.overrides]]
-module = ["pandas", "pandas.*", "duckdb", "faker", "shap", "xgboost", "sklearn.*", "plotly", "plotly.*"]
+module = ["pandas", "pandas.*", "duckdb", "faker", "shap", "xgboost", "sklearn.*", "plotly", "plotly.*", "yaml"]
 ignore_missing_imports = true
 
 [tool.pytest.ini_options]
diff --git a/packages/data-analytics-demo/semantic/kpi.yml b/packages/data-analytics-demo/semantic/kpi.yml
new file mode 100644
index 0000000..088aeaf
--- /dev/null
+++ b/packages/data-analytics-demo/semantic/kpi.yml
@@ -0,0 +1,135 @@
+# MetricFlow-compatible semantic models + metrics for the data-analytics demo.
+#
+# This file declares the single source of truth for KPIs that the dashboard
+# and any downstream BI tool can reuse. Each metric resolves to a primary
+# measure plus a set of dimensions that the metric can be sliced by; the
+# schema mirrors MetricFlow 0.21x.
+#
+# Validation: `make semantic-validate` (Python validator in
+# src/data_analytics_demo/semantic/validator.py — checks YAML shape +
+# AC-6.2 ≥1 dim + ≥1 measure per metric).
+
+semantic_models:
+  - name: customers
+    description: Tenant accounts. Drives signup-month cohort + region slicing.
+    model: ref('stg_customers')
+    entities:
+      - name: customer
+        type: primary
+        expr: customer_id
+    dimensions:
+      - name: region
+        type: categorical
+      - name: plan_tier_at_signup
+        type: categorical
+      - name: signup_date
+        type: time
+        type_params:
+          time_granularity: day
+    measures:
+      - name: customer_count
+        description: Number of customers.
+        agg: count
+        expr: customer_id
+
+  - name: subscriptions
+    description: Subscription lifecycle facts (active / canceled / paused).
+    model: ref('stg_subscriptions')
+    entities:
+      - name: subscription
+        type: primary
+        expr: subscription_id
+      - name: customer
+        type: foreign
+        expr: customer_id
+    dimensions:
+      - name: plan_tier
+        type: categorical
+      - name: status
+        type: categorical
+      - name: start_date
+        type: time
+        type_params:
+          time_granularity: day
+    measures:
+      - name: subscription_count
+        description: Number of subscriptions.
+        agg: count
+        expr: subscription_id
+      - name: monthly_amount
+        description: Monthly subscription value (USD).
+        agg: sum
+        expr: monthly_amount_usd
+      - name: active_subscription_count
+        description: Count of subscriptions whose status is active.
+        agg: sum
+        expr: case when status = 'active' then 1 else 0 end
+
+  - name: invoices
+    description: Monthly billing records for paid subscriptions.
+    model: ref('stg_invoices')
+    entities:
+      - name: invoice
+        type: primary
+        expr: invoice_id
+      - name: customer
+        type: foreign
+        expr: customer_id
+      - name: subscription
+        type: foreign
+        expr: subscription_id
+    dimensions:
+      - name: status
+        type: categorical
+      - name: period_start
+        type: time
+        type_params:
+          time_granularity: day
+    measures:
+      - name: invoice_count
+        description: Number of invoices.
+        agg: count
+        expr: invoice_id
+      - name: paid_amount_usd
+        description: Sum of paid invoice amounts.
+        agg: sum
+        expr: case when status = 'paid' then amount_usd else 0 end
+
+metrics:
+  - name: customers
+    description: Distinct customer count.
+    type: simple
+    type_params:
+      measure: customer_count
+    dimensions:
+      - region
+      - plan_tier_at_signup
+      - signup_date
+
+  - name: active_subscriptions
+    description: Count of subscriptions whose status is "active" right now.
+    type: simple
+    type_params:
+      measure: active_subscription_count
+    dimensions:
+      - plan_tier
+      - start_date
+
+  - name: monthly_recurring_revenue
+    description: Sum of monthly subscription amounts (US dollars).
+    type: simple
+    type_params:
+      measure: monthly_amount
+    dimensions:
+      - plan_tier
+      - status
+      - start_date
+
+  - name: paid_invoice_volume
+    description: Sum of paid invoice amounts (US dollars).
+    type: simple
+    type_params:
+      measure: paid_amount_usd
+    dimensions:
+      - status
+      - period_start
diff --git a/packages/data-analytics-demo/src/data_analytics_demo/cli.py b/packages/data-analytics-demo/src/data_analytics_demo/cli.py
index 8e66fbe..101ad66 100644
--- a/packages/data-analytics-demo/src/data_analytics_demo/cli.py
+++ b/packages/data-analytics-demo/src/data_analytics_demo/cli.py
@@ -63,5 +63,17 @@ def narrative() -> None:
     typer.echo(f"wrote {out}")
 
 
+@app.command()
+def semantic() -> None:
+    """Validate the MetricFlow KPI definitions (semantic/kpi.yml)."""
+    from data_analytics_demo.semantic import validator
+
+    report = validator.main()
+    typer.echo(
+        f"OK: {report.semantic_model_count} semantic models, "
+        f"{report.metric_count} metrics"
+    )
+
+
 if __name__ == "__main__":
     app()
diff --git a/packages/data-analytics-demo/src/data_analytics_demo/semantic/__init__.py b/packages/data-analytics-demo/src/data_analytics_demo/semantic/__init__.py
new file mode 100644
index 0000000..fba6e66
--- /dev/null
+++ b/packages/data-analytics-demo/src/data_analytics_demo/semantic/__init__.py
@@ -0,0 +1,6 @@
+"""Semantic-layer validator for `semantic/kpi.yml`.
+
+Checks the MetricFlow-compatible KPI definition file against AC-6.1 and
+AC-6.2 (each metric has ≥ 1 dimension and ≥ 1 measure). Independent of
+the MetricFlow CLI so the test suite has no CLI-shell dependency.
+"""
diff --git a/packages/data-analytics-demo/src/data_analytics_demo/semantic/validator.py b/packages/data-analytics-demo/src/data_analytics_demo/semantic/validator.py
new file mode 100644
index 0000000..3916aa9
--- /dev/null
+++ b/packages/data-analytics-demo/src/data_analytics_demo/semantic/validator.py
@@ -0,0 +1,146 @@
+"""KPI YAML validator — enforces the MetricFlow schema invariants we rely on.
+
+Run via `python -m data_analytics_demo.semantic.validator` or
+`data-analytics-demo semantic` (the make target proxies through this).
+"""
+
+from __future__ import annotations
+
+import sys
+from dataclasses import dataclass
+from pathlib import Path
+
+import yaml
+
+from ..ml import _io
+
+DEFAULT_KPI_PATH_PARTS = ("semantic", "kpi.yml")
+
+
+def _emit(msg: str) -> None:
+    print(f"[semantic] {msg}", file=sys.stderr, flush=True)  # noqa: T201
+
+
+def default_kpi_path() -> Path:
+    root = _io.package_root()
+    return root.joinpath(*DEFAULT_KPI_PATH_PARTS)
+
+
+@dataclass(frozen=True)
+class ValidationReport:
+    semantic_model_count: int
+    metric_count: int
+    metric_names: list[str]
+
+
+class ValidationError(RuntimeError):
+    """Raised when the KPI YAML fails an AC-6.x invariant."""
+
+
+def _require_keys(
+    name: str,
+    obj: dict[str, object],
+    required: tuple[str, ...],
+) -> None:
+    missing = [k for k in required if k not in obj]
+    if missing:
+        raise ValidationError(f"{name}: missing required keys {missing}")
+
+
+def _validate_semantic_model(node: dict[str, object]) -> dict[str, list[str]]:
+    """Return the set of dimension / measure names this model exposes."""
+    _require_keys("semantic_model", node, ("name", "model", "entities", "dimensions", "measures"))
+    name = str(node["name"])
+    dims = node.get("dimensions", [])
+    measures = node.get("measures", [])
+    if not isinstance(dims, list) or not dims:
+        raise ValidationError(f"semantic_model {name!r}: needs ≥ 1 dimension")
+    if not isinstance(measures, list) or not measures:
+        raise ValidationError(f"semantic_model {name!r}: needs ≥ 1 measure")
+    return {
+        "dimensions": [str(d["name"]) for d in dims if isinstance(d, dict) and "name" in d],
+        "measures": [str(m["name"]) for m in measures if isinstance(m, dict) and "name" in m],
+    }
+
+
+def _validate_metric(
+    node: dict[str, object],
+    all_dims: set[str],
+    all_measures: set[str],
+) -> str:
+    _require_keys("metric", node, ("name", "type", "type_params", "dimensions"))
+    name = str(node["name"])
+    tp = node.get("type_params", {})
+    if not isinstance(tp, dict) or "measure" not in tp:
+        raise ValidationError(f"metric {name!r}: type_params.measure is required")
+    measure_ref = str(tp["measure"])
+    if measure_ref not in all_measures:
+        raise ValidationError(
+            f"metric {name!r}: references unknown measure {measure_ref!r}"
+        )
+    dims = node.get("dimensions", [])
+    if not isinstance(dims, list) or not dims:
+        raise ValidationError(f"metric {name!r}: needs ≥ 1 dimension (AC-6.2)")
+    for d in dims:
+        if str(d) not in all_dims:
+            raise ValidationError(
+                f"metric {name!r}: references unknown dimension {d!r}"
+            )
+    return name
+
+
+def validate(path: Path | None = None) -> ValidationReport:
+    kpi_path = path or default_kpi_path()
+    if not kpi_path.exists():
+        raise FileNotFoundError(f"kpi.yml not found at {kpi_path}")
+
+    _emit(f"loading {kpi_path}")
+    text = kpi_path.read_text(encoding="utf-8")
+    try:
+        doc = yaml.safe_load(text)
+    except yaml.YAMLError as exc:
+        raise ValidationError(f"YAML parse error: {exc}") from exc
+
+    if not isinstance(doc, dict):
+        raise ValidationError("kpi.yml: top-level must be a mapping")
+
+    semantic_models = doc.get("semantic_models", [])
+    metrics = doc.get("metrics", [])
+    if not isinstance(semantic_models, list) or not semantic_models:
+        raise ValidationError("kpi.yml: needs ≥ 1 semantic_model")
+    if not isinstance(metrics, list) or not metrics:
+        raise ValidationError("kpi.yml: needs ≥ 1 metric")
+
+    all_dims: set[str] = set()
+    all_measures: set[str] = set()
+    for sm in semantic_models:
+        if not isinstance(sm, dict):
+            raise ValidationError("semantic_models[]: items must be mappings")
+        exposed = _validate_semantic_model(sm)
+        all_dims.update(exposed["dimensions"])
+        all_measures.update(exposed["measures"])
+
+    metric_names: list[str] = []
+    for m in metrics:
+        if not isinstance(m, dict):
+            raise ValidationError("metrics[]: items must be mappings")
+        metric_names.append(_validate_metric(m, all_dims, all_measures))
+
+    report = ValidationReport(
+        semantic_model_count=len(semantic_models),
+        metric_count=len(metrics),
+        metric_names=metric_names,
+    )
+    _emit(
+        f"OK — {report.semantic_model_count} semantic models / "
+        f"{report.metric_count} metrics: {', '.join(report.metric_names)}"
+    )
+    return report
+
+
+def main() -> ValidationReport:
+    return validate()
+
+
+if __name__ == "__main__":
+    main()
diff --git a/packages/data-analytics-demo/tests/test_semantic.py b/packages/data-analytics-demo/tests/test_semantic.py
new file mode 100644
index 0000000..dff5cb6
--- /dev/null
+++ b/packages/data-analytics-demo/tests/test_semantic.py
@@ -0,0 +1,86 @@
+"""Tests for the semantic-layer validator (T-10 / AC-6.1〜6.2)."""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import pytest
+
+from data_analytics_demo.semantic import validator
+
+# ---- AC-6.1: WHEN `make semantic-validate`, exit 0 on a valid file ---------
+
+def test_ac_6_1_valid_file_returns_report() -> None:
+    report = validator.validate()
+    assert report.semantic_model_count >= 1
+    assert report.metric_count >= 1
+    assert all(isinstance(n, str) for n in report.metric_names)
+
+
+# ---- AC-6.2: each metric has ≥ 1 dimension and ≥ 1 measure -----------------
+
+def test_ac_6_2_each_metric_has_dim_and_measure_via_model() -> None:
+    report = validator.validate()
+    # validator.validate() already enforces this; the test would fail on
+    # ValidationError if any metric were missing a dimension or measure.
+    assert report.metric_count == len(report.metric_names)
+
+
+# ---- Negative paths --------------------------------------------------------
+
+def test_missing_file_raises(tmp_path: Path) -> None:
+    with pytest.raises(FileNotFoundError, match="kpi.yml not found"):
+        validator.validate(tmp_path / "nope.yml")
+
+
+def test_empty_metric_dimensions_raises(tmp_path: Path) -> None:
+    bad = tmp_path / "bad.yml"
+    bad.write_text(
+        """
+semantic_models:
+  - name: x
+    model: "ref('stg_x')"
+    entities: [{name: x, type: primary, expr: id}]
+    dimensions: [{name: d, type: categorical}]
+    measures: [{name: m, agg: count, expr: id}]
+metrics:
+  - name: bad_metric
+    type: simple
+    type_params: {measure: m}
+    dimensions: []
+""",
+        encoding="utf-8",
+    )
+    with pytest.raises(validator.ValidationError, match="needs ≥ 1 dimension"):
+        validator.validate(bad)
+
+
+def test_unknown_measure_raises(tmp_path: Path) -> None:
+    bad = tmp_path / "bad.yml"
+    bad.write_text(
+        """
+semantic_models:
+  - name: x
+    model: "ref('stg_x')"
+    entities: [{name: x, type: primary, expr: id}]
+    dimensions: [{name: d, type: categorical}]
+    measures: [{name: m, agg: count, expr: id}]
+metrics:
+  - name: bad_metric
+    type: simple
+    type_params: {measure: does_not_exist}
+    dimensions: [d]
+""",
+        encoding="utf-8",
+    )
+    with pytest.raises(validator.ValidationError, match="unknown measure"):
+        validator.validate(bad)
+
+
+# ---- Required metric inventory ---------------------------------------------
+
+def test_includes_canonical_metrics() -> None:
+    """The four headline metrics the dashboard reads must be defined."""
+    report = validator.validate()
+    expected = {"customers", "active_subscriptions", "monthly_recurring_revenue", "paid_invoice_volume"}
+    assert expected.issubset(set(report.metric_names))