Mark Compute

Docs

Index methodology

Panel marks publish from the same tape as Quote. Indices are observational references, not clearing underlyings or settlement instructions.

Per-listing scoring coefficients and CireX component weights are proprietary. Public docs describe what each index measures, not the internal blend.

4.1 CireX: Compute Congestion & Risk Index

Range: 0–100 (higher = more systemic stress)
Scope: Market-wide, workload-independent

CireX is a composite stress index built from capacity-weighted panel signals:

Component (public)Meaning
CongestionShare of listed GPU capacity flagged for execution-risk / unreliable-inventory signals
ChurnRate at which listings disappear from the tape over the lookback window
VolatilityDispersion of aggregate spot pricing (capped)
InterruptionPanel-wide interruption model pressure (mean SIR)

Exact component weights and caps are methodology v1 implementation details (not published). Material changes require a version bump (v1 → v2).

4.2 CIX: Compute Integrity Index

Capacity-weighted share of listed GPUs whose signals classify as execution-risk inventory (same family as §3.2; not a claim that hardware is absent):

CIX = execution_risk_capacity / total_capacity × 100

(Internal engine may still label this “ghost inventory” in code; public interpretation is execution-risk share.)

4.3 Basis spread

Median relative gap between naive listed job cost and TRAC on the panel:

Basis_median = median( (listed_job_cost − TRAC) / TRAC )

Reference workload for index publication: 48h · 500 GB (illustrative defaults until caller-specified).

Reminder: TRAC in this ratio uses the modeled risk premium, not calibrated expected loss.

4.4 Co-movement & stress suite

ToolRole
Provider correlation matrixCo-movement of capacity-weighted spot moves across providers
HHICapacity concentration
Stress scenariosDeterministic narrative transforms (e.g. demand spike, deverification wave, churn surge)