v0.13.3

v0.13.3: Azure ND-series GPU coverage (A100/H100)

v0.13.2

v0.13.2: remove inaccurate PR-comment exclusivity claim

v0.13.1

v0.13.1: Azure GPU coverage complete (NCasT4_v3 family)

v0.13.0

v0.13.0: Azure GPU instance support (NCasT4_v3 series)

Standard_NC4as_T4_v3, Standard_NC8as_T4_v3, and Standard_NC16as_T4_v3
(single NVIDIA T4 each) added to the Azure ledger, Scope 2 only, eastus
pricing only. No extractor changes needed - same generic size-field
extraction as the existing AWS GPU work. Surfaces in the dedicated AI
Infrastructure Carbon Impact PR-comment section.

Deferred: Standard_NC64as_T4_v3 (4x T4), Azure ND-series (A100/H100),
Azure ML.

v0.12.0

v0.12.0: dedicated AI Infrastructure Carbon Impact PR-comment section

GPU/SageMaker/Vertex Workbench resources now get a dedicated callout in
the PR comment instead of blending into the general resource table:
combined Scope 2 carbon and cost across all AI/GPU resources, per-resource
embodied-carbon gap flagging, and the managed-AI runtime assumption.

Closes the main remaining gap in the AI differentiation strategy: putting
AI infrastructure carbon cost in front of the engineer before provisioning
is now true in the shipped product, not just the strategy doc.

v0.11.0

v0.11.0: managed AI service detection (SageMaker, Vertex AI Workbench)

v0.10.0

v0.10.0: GPU instance support (g5, p4d, p5), Scope 2 only

v0.9.1

chore: bump to v0.9.1

v0.9.0

New: Kubernetes node group support

Adds carbon, water, and cost analysis for Kubernetes node groups: aws_eks_node_group, azurerm_kubernetes_cluster, azurerm_kubernetes_cluster_node_pool, and google_container_node_pool.

Node groups resolve to the same instance ledger used for standalone VMs. Node count is the new dimension: Scope 2, Scope 3, water, and cost all scale linearly with node count, applied on top of the existing per-instance calculation.

Autoscaling groups are reported at minimum configured size (min_size / min_count / autoscaling.min_node_count), never desired or maximum size. This is a deliberate floor, consistent with the project's existing principle that a tool showing a wrong number is worse than one showing no number. The PR comment notes this explicitly whenever a node group is detected.

ARM upgrade and region shift recommendations apply across the whole node group, using the same per-instance scoring already used for standalone resources.

Also in this release

• New E2E fixture (fixtures/tfplan.eks.e2e.json) and CI step exercising the GitHub Action against a real EKS node group plan.
• 9 new tests (116 total, all passing): extraction across all three providers, the min/desired/max fallback chain, and engine-level linear scaling.
• Fixed a stale claim in METHODOLOGY.md that said memory power draw was not yet included in calculations, despite the documented formulas already including it.
• Fixed stale test/fixture counts in CONTRIBUTING.md and README.md.
• Removed em dashes from all public-facing documentation and from the live PR comment output itself.

See METHODOLOGY.md for the full calculation details and CONTRIBUTING.md for the new fixture.

v0.8.3

chore: bump to v0.8.3 — fix inline suggestions version ref in README