# Cross-signals — Methodology note

> Type: Analytical framework (not a standalone tradable index)  
> Frequency: With weekly notes / Indicators updates  
> Status: Design draft

---

## 1. Purpose

Read **compute spot indices** and **equity baskets** on one timeline to produce citable narrative signals — not a third price series.

Default inputs:

- $T_t$: CN-Token Spot Median, **USD / 1M tokens** (or weekly $\Delta \ln T_t$)
- $F_t$: CN-Frontier basket price $P_t$ (USD)
- $O_t$: CN-Open basket price $P_t$ (USD)
- $A_t$: CN-Arbitrage Spread (median relative range, %)
- $E^{\text{I}}_t, E^{\text{H}}_t, E^{\text{S}}_t$: Infra / Hyperscale / Software weekly returns or index changes
- Low frequency: CapEx, NVIDIA DC revenue, TTI from Indicators

Compare in USD levels or log changes; converting to index points first is unnecessary.

---

## 2. Core signals

### 2.1 Price–equity divergence (deflation vs supply narrative)

Weekly changes $\Delta \ln T_t$ and $\Delta \ln E^{\text{I}}_t$:

| Pattern | Tag | Reading hint |
|---------|-----|--------------|
| $T$↓ and Infra↑ | `divergence_supply` | Inference cheaper; hardware equities still strong |
| $T$↑ and Infra↓ | `divergence_squeeze` | List prices firm or squeeze narrative vs equities |
| Same direction | `aligned` | Narratives agree (joint deflation or inflation) |

Publish a 4-week rolling correlation plus the week’s tag; avoid over-fitting.

### 2.2 CapEx vs spot lag (buildout vs tariff)

- Leading side: Hyperscaler CapEx, NVIDIA DC (quarterly)
- Lagging side: CN-Token Spot, CN-Open

Report simple correlations and scatter plots at quarterly frequency; **do not claim a fixed “leads by N quarters” in v1** unless out-of-sample evidence is written into a revision.

### 2.3 Frontier–Open gap

Using dollar basket prices $P^F_t$, $P^O_t$:

$$
G_t = \ln P^F_t - \ln P^O_t
$$

Larger $G_t$: closed relatively dearer (or open deflating faster).

### 2.4 Geo / regional premium (optional)

When the panel can split by `region` / provider geography:

$$
\Gamma_t = \frac{\mathrm{median}(p^{\text{CN-ish}})}{\mathrm{median}(p^{\text{global}})} - 1
$$

Cross-check Indicators such as NVIDIA China revenue share. If data are thin, keep as prose observation only — no official level.

### 2.5 Routing value

Rising $A_t$ (Arbitrage Spread) with falling $T_t$: the market is cheaper overall but cross-cloud quotes have not converged → comparison / routing products gain value.

---

## 3. Output shape

A fixed block for ComputeNav weekly notes or News/Market:

```text
week: YYYY-MM-DD
token_median_usd: ...
token_median_wow: ...
infra_wow: ...
divergence_tag: aligned | divergence_supply | ...
frontier_open_gap: ...
arb_spread_pct: ...
note: ≤2 sentences
```

---

## 4. Governance

- Cross-signals **do not replace** single-index methodologies.
- Tag-rule changes are an analysis-framework version bump, not silent tuning.
- Explicit disclaimer: not investment advice; not a guaranteed trading signal.

---

## 5. Relation to GCPI Research

The GCPI white paper covers standardised academic benchmarks and GRP-style structure; this note covers **joint reading of ComputeNav market indices**. Link both; keep series and narrative boundaries clear.