Issue 007.

Headline state.

The composite metrics reproduce Issue 006 because the underlying Issue 007 input vector is identical. Tier medians, the pooled non-hyperscaler median, both ratios and the Price Floor were independently recomputed from the fresh observations.

The three tiers.

All three tier medians are unchanged from Issue 006.

A flat priced panel.

All fifteen priced providers were unchanged from Issue 006, producing a zero-movement priced panel in the August 10 observation.

The two repricings recorded in Issue 006 also held again. Hyperstack remained at $3.20 after its move from $2.40, and RunPod Secure Cloud remained at $2.99 after its move from $3.29. On the evidence to date, both changes look persistent rather than like one-cycle anomalies.

The spread remains wide and unchanged from Issue 006: the Tier 1 median of $10.53 is 2.7352 times the pooled non-hyperscaler median used in the Hyperscaler Premium calculation.

Track 2 under A1-R1.

Issue 007 is a priced-only observation under the two-track cadence adopted August 10 and approved August 11, 2026. The twelve providers excluded at Issue 006 were not recaptured in this Track 2 cycle. Their dispositions carry from the full twenty-seven-provider sweep sealed July 21, and their next full re-verification is due on or about August 21.

The scheduled slots of July 27 and August 3 were not observed. Under A1-R1 the skipped slots force no full sweep and are disclosed here. Movement in Issue 007 is stated against Issue 006.

What changed around the numbers.

GCP. The rate of record now derives from the published accelerator-optimized pricing page rather than the pricing calculator, a source-class upgrade under the locked hierarchy. The posted $88.490000119 per instance-hour derives to the same $11.0613 rate.

Jarvislabs. An editorial note open since Issue 006 is closed. The hero summary uses starting-at language, but the controlling on-demand GPU table posts a fixed $2.69 per GPU-hour rate.

Atlas Cloud. The $2.95 H100 SXM rate was re-evidenced from a fully logged-out public surface, replacing the conditional evidence state carried at Issue 006.

Paperspace. The promotional-wording watch remains open. The controlling official documentation posts $5.95 per hour with no qualifier, and the same page prices an H100 x8 machine at $47.60, exactly eight times the per-GPU figure.

Precision. CoreWeave's eight-GPU HGX H100 instance at $49.24 per hour derives to $6.155 per GPU-hour, not $6.16. The $6.16 figure on the page belongs to a separate single-GPU inference product.

The current rule.

The SVCI observes public, fixed, on-demand rates for the H100 SXM 80GB-class anchor across a twenty-seven-provider universe in three tiers, from logged-out surfaces only. Quote-gated, from-floor, spot, committed and account-gated figures never enter a median. Full-machine prices divide to per-GPU in recorded arithmetic. Providers without a comparable public rate are excluded and disclosed, not estimated. Under A1-R1, priced-provider observations are captured fresh on weekly or biweekly issue cycles, while the full twenty-seven-provider panel is re-verified on a monthly clock.

How to check this issue.

The public data layer, calculation report, public capture-ledger extract, selected public evidence manifest and approved cadence amendment are published with the issue. The private byte-level evidence archive, founder attestation and deleted signed-in Atlas frames are not part of the web deployment.

Headline panel status.

Fifteen fresh priced observations enter the medians. Twelve exclusion dispositions carry from the July 21 full sweep under A1-R1. No row is pending.