From 73876c4f771b367b3cfb249ff4bfa38d3c3d7c1f Mon Sep 17 00:00:00 2001
From: sofq <hoanghongquan93@gmail.com>
Date: Tue, 5 May 2026 15:05:44 +0700
Subject: [PATCH] fix(validate): skip-list for shards the validator can't
 compare

Add SKIP_REVALIDATION dict in pipeline/validate/driver.py and short-circuit
those shards with exit="skipped" in the report. Listed shards return 0 so
the data-validate workflow closes any open drift issue.

Initial entry: gcp-gce. Ingest (pipeline/ingest/gcp_gce.py) synthesizes
per-machine totals from per-vCPU + per-GiB component prices, while the
validator reads the raw component unitPrice. The two are not comparable
without re-doing the ingest math, producing 20/20 false-positive drift
records on every run.

Closes #62.
---
 pipeline/tests/test_validate_driver.py | 33 +++++++++++++++++++++++
 pipeline/validate/driver.py            | 36 +++++++++++++++++++++++++-
 2 files changed, 68 insertions(+), 1 deletion(-)

diff --git a/pipeline/tests/test_validate_driver.py b/pipeline/tests/test_validate_driver.py
index a802caf..9fed0d1 100644
--- a/pipeline/tests/test_validate_driver.py
+++ b/pipeline/tests/test_validate_driver.py
@@ -223,3 +223,36 @@ def multi_drift(samples, **kwargs):
     data = json.loads(report_path.read_text())
     assert result == 1
     assert len(data["drift_records"]) == data["sample_size"]
+
+
+
+def test_driver_skips_listed_shard(tmp_path: Path) -> None:
+    """Shards in SKIP_REVALIDATION are skipped — revalidator is not called."""
+    from validate.driver import SKIP_REVALIDATION
+
+    db = _make_minimal_shard(tmp_path)
+    report_path = tmp_path / "report.json"
+
+    called = False
+
+    def must_not_be_called(samples, **kwargs):
+        nonlocal called
+        called = True
+        return [], []
+
+    skipped_shard = next(iter(SKIP_REVALIDATION))
+    result = run_validation(
+        shard=skipped_shard,
+        shard_db=db,
+        budget=5,
+        report=report_path,
+        revalidator=must_not_be_called,
+        seed=42,
+    )
+    assert result == 0
+    assert not called
+    data = json.loads(report_path.read_text())
+    assert data["exit"] == "skipped"
+    assert data["skip_reason"]
+    assert data["sample_size"] == 0
+    assert data["drift_records"] == []
diff --git a/pipeline/validate/driver.py b/pipeline/validate/driver.py
index 7f3a2fd..5ca6763 100644
--- a/pipeline/validate/driver.py
+++ b/pipeline/validate/driver.py
@@ -26,6 +26,22 @@
 logger = logging.getLogger(__name__)
 
 
+# Shards where the upstream API can't be compared one-to-one against catalog
+# rows: catalog stores synthesized values (e.g. gcp-gce machine totals built
+# from per-vCPU + per-GiB component prices) while the upstream API exposes the
+# components. Listing here makes the validator skip revalidation and emit a
+# pass with the reason recorded in the report.
+SKIP_REVALIDATION: dict[str, str] = {
+    "gcp-gce": (
+        "ingest synthesizes machine totals from per-vCPU and per-GiB "
+        "component SKUs (see pipeline/ingest/gcp_gce.py); validator "
+        "compares against the raw component unitPrice, producing "
+        "false-positive drift. Re-enable once a sidecar or component-aware "
+        "comparison lands."
+    ),
+}
+
+
 # ---------------------------------------------------------------------------
 # Types
 # ---------------------------------------------------------------------------
@@ -52,7 +68,8 @@ class ValidationReport:
     drift_records: list[dict]
     missing_upstream: list[str]
     vantage_drift: list[dict]
-    exit: str  # "pass" | "fail"
+    exit: str  # "pass" | "fail" | "skipped"
+    skip_reason: str | None = None
 
     def as_dict(self) -> dict:
         return dataclasses.asdict(self)
@@ -123,6 +140,23 @@ def run_validation(
     int
         0 on pass, 1 on fail.
     """
+    if shard in SKIP_REVALIDATION:
+        reason = SKIP_REVALIDATION[shard]
+        logger.info("Skipping revalidation for %s: %s", shard, reason)
+        report_data = ValidationReport(
+            shard=shard,
+            generated_at=datetime.now(UTC).isoformat(),
+            sample_size=0,
+            drift_records=[],
+            missing_upstream=[],
+            vantage_drift=[],
+            exit="skipped",
+            skip_reason=reason,
+        )
+        report.parent.mkdir(parents=True, exist_ok=True)
+        report.write_text(json.dumps(report_data.as_dict(), indent=2))
+        return 0
+
     if revalidator is None:
         revalidator = _default_revalidator(shard)