Every change to the CRI methodology, index scope, data sources, or governance framework is classified, documented with rationale, and versioned. Historical index values computed under a prior methodology version are never restated. This log is the public record of that discipline.

Patch  ·  x.x.0 → x.x.1
5 Business Days
Model string corrections, collection logic fixes, new model variants. No material impact on published rates.
Minor  ·  x.0.x → x.1.0
30 Calendar Days
Source additions or removals, Silver filter threshold adjustments, new index series. Comment period runs for full notice period.
Major  ·  1.x.x → 2.0.0
90 Calendar Days
Computation method, window length, unit of measurement changes, index retirement. Formal consultation period with published response required.
2026
CRI v1.0.0 — Initial Publication
First publication of the CRI index family under Methodology v1.0.0. Five spot indices and six cluster indices across two tiers. Full governance stack published alongside rate data.
  • Spot indices: CRI-H100, CRI-A100, CRI-H200, CRI-L40S, CRI-B200
  • Cluster indices (MM): CRI-H100-CLS-MM, CRI-A100-CLS-MM, CRI-B200-CLS-MM
  • Cluster indices (ENT): CRI-H100-CLS-ENT, CRI-H200-CLS-ENT, CRI-B200-CLS-ENT
  • Primary rate: seller_ask_noninterruptible — on-demand spot, no preemption risk
  • Sources: RunPod, Shadeform, Lambda Labs, TensorDock, Nebius (spot); RunPod Cluster, Lambda Cluster (MM); CoreWeave (ENT)
  • Computation: Flat median (percentile_approx 0.5) across all Silver-eligible observations in trailing 7-day window
  • Pipeline: Azure Databricks, Delta Lake medallion architecture, nightly audit_pipeline + weekly audit_sources
Methodology v1.0.0 Governance v1.0.0 Change Management v1.0.0 Fallback Guidance v1.0.0 Complaints Procedure v1.0 Design Rationale v1.0.0
Future patch changes will appear here
Model string corrections, collection logic fixes, and new GPU model variants added to existing indices are classified as Patch changes. Each will be logged here with a 5 business day advance notice before taking effect.