# ComputeNav / GCPI Index Handbook

> Status: design draft (not all series published)  
> Data sources: ComputeNav weekly inference price panels + Firms daily snapshots  
> Ownership: GCPI Research holds **index research assets** (scripts, series, handbook); ComputeNav holds panels and market UI  
> Build: `python3 scripts/build_cn_indices.py` → `data/processed/cn_*.csv`, `api/v1/indices.json`  
> Live (v0.3): Token Spot · Frontier (seat-role) · Open · Arbitrage · QA deferred

Concise methodologies for planned compute price indices and equity baskets. Each file can be cited on its own.

## Reading order

### Compute price indices (priority)

| File | Index | Headline unit | One-liner |
|------|-------|---------------|-----------|
| [compute/CN-Token-Spot.md](compute/CN-Token-Spot.md) | CN-Token Spot | USD / 1M tokens | Market-wide chat token spot Floor / P25 / Median |
| [compute/CN-Frontier-Basket.md](compute/CN-Frontier-Basket.md) | CN-Frontier Basket | USD / 1M tokens | Fixed frontier closed-model basket blended cost |
| [compute/CN-Open-Basket.md](compute/CN-Open-Basket.md) | CN-Open Basket | USD / 1M tokens | Mainstream open-weight hosting median |
| [compute/CN-Arbitrage-Spread.md](compute/CN-Arbitrage-Spread.md) | CN-Arbitrage Spread | % | Cross-host spread for the same model (friction) |
| [compute/CN-Quality-Adjusted.md](compute/CN-Quality-Adjusted.md) | CN-Quality-Adjusted | USD / capability unit | Quality-adjusted effective compute cost |

### Equity indices

| File | Index | Frequency | One-liner |
|------|-------|-----------|-----------|
| [equities/CN-Infra.md](equities/CN-Infra.md) | CN-Infra | Daily / weekly | AI hardware and manufacturing supply chain |
| [equities/CN-Hyperscale.md](equities/CN-Hyperscale.md) | CN-Hyperscale | Daily / weekly | Hyperscaler cloud / platform demand side |
| [equities/CN-Software-AI.md](equities/CN-Software-AI.md) | CN-Software AI | Daily / weekly | AI software and application layer |

### Cross

| File | Contents |
|------|----------|
| [cross-signals.md](cross-signals.md) | Token spot vs equity baskets, CapEx lag, geo premium |

## Shared conventions

1. **Auditable**: Constituents, rebalance rules, and missing-data handling are fixed in writing; same evidence-chain habit as the ComputeNav weekly checklist.
2. **Units (important)**
   - **Compute price indices**: Headline in **USD** (typically USD / 1M tokens blended; Arbitrage in %; Quality-Adjusted in USD / capability unit). Pages, API, and notes cite dollar levels by default.
   - **Optional companion series**: Index points with base `2026-06-20`, $I_t = 100 \times P_t / P_{t_0}$, only for multi-series comparison or equity co-plots; **does not replace** the USD headline.
   - **Equity indices**: Price index, base close on `2026-06-20` = 100 (industry habit; not total return unless stated).
3. **Blended quote**: Unless stated otherwise, $p = 0.7\cdot\text{input} + 0.3\cdot\text{output}$, USD / 1M tokens.
4. **Non-goals (v1)**: Sub-minute realtime feeds, futures simulation, complex volatility products, investment advice.
5. **Relation to GCPI**: Academic GCPI (`docs/methodology_outline.md`) is a standardised research benchmark; handbook CN-\* series are **spot / basket statistics** on ComputeNav panels. Cite both if needed; do not treat them as one series.

## Suggested release cadence

1. **v1**: CN-Token Spot (Median + Floor + P25) + this handbook  
2. **v1.1**: CN-Infra, CN-Hyperscale (equal weight)  
3. **v2**: Frontier / Open baskets + cross-signal weekly note  
4. **Deferred**: Quality-Adjusted official levels, volatility indices