From the #1340/#1347 design reviews: providers/aws/ladder.GetUsageBaseline deliberately returns StableUSDPerHour nil (no honest estimator), which makes the allocation engine route ALL core gap to the flex layer (Compute SPs). The base layer (EC2 Instance SPs, the highest-discount instrument) is therefore inert in v1 - safe, but leaves the Compute-SP-vs-EC2-Instance-SP discount delta unclaimed.
Build a real stable-usage estimator: e.g. per-family EC2 usage from CE GetSavingsPlansPurchaseRecommendation for the EC2Instance plan type (AWS already computes stable hourly commitment recommendations), or a family-level low-water percentile over instance-family usage. Output feeds UsageBaseline.StableUSDPerHour with documented semantics (post-buffer-fraction per the pkg/ladder contract).
From the #1340/#1347 design reviews: providers/aws/ladder.GetUsageBaseline deliberately returns StableUSDPerHour nil (no honest estimator), which makes the allocation engine route ALL core gap to the flex layer (Compute SPs). The base layer (EC2 Instance SPs, the highest-discount instrument) is therefore inert in v1 - safe, but leaves the Compute-SP-vs-EC2-Instance-SP discount delta unclaimed.
Build a real stable-usage estimator: e.g. per-family EC2 usage from CE GetSavingsPlansPurchaseRecommendation for the EC2Instance plan type (AWS already computes stable hourly commitment recommendations), or a family-level low-water percentile over instance-family usage. Output feeds UsageBaseline.StableUSDPerHour with documented semantics (post-buffer-fraction per the pkg/ladder contract).