> ## Documentation Index
> Fetch the complete documentation index at: https://docs.beliefsystems.xyz/llms.txt
> Use this file to discover all available pages before exploring further.

# How It Works

> How a Belief Index is produced: from raw prediction market data to a published, verifiable index level.

Every Belief Index level is the output of a fixed production pipeline: define a theme, screen the eligible markets, weight the basket by published rules, value it at regular NAV windows, and publish a chain-linked index level. This page walks through that pipeline end to end. Each stage links to the full methodology document that governs it.

<Note>
  If you are familiar with how equity index providers operate – a published rulebook, an eligible universe, systematic weighting, scheduled rebalances – the Belief Index pipeline will feel familiar. The differences are the underlying instruments (prediction market contracts rather than equities), the valuation cadence (every 30 minutes rather than end-of-day), and one thing a paper benchmark cannot offer: each series is physically replicated, and the constituent holdings backing it are verifiable on a public blockchain.
</Note>

## The Production Pipeline

<Steps>
  <Step title="Define the theme">
    Each series begins with a coherent question – "how is the market pricing the 2026 easing cycle?" – and a universe of prediction markets that bear on it. Themes are chosen so the index level reads as an answer to one category-level question.
  </Step>

  <Step title="Screen for eligibility">
    Candidate markets pass published screens: minimum liquidity and order book depth, resolution clarity (an unambiguous settlement source), thematic relevance, and duplication checks. The criteria and thresholds are documented in [Market Selection & Weighting](/indices/selection-and-weighting).
  </Step>

  <Step title="Weight the basket">
    Constituent weights follow the published formula, with concentration caps so no single event dominates the series. Weighting is systematic – the same inputs always produce the same weights.
  </Step>

  <Step title="Value at NAV windows">
    At each NAV window – currently every 30 minutes, subject to change – every constituent is priced from its observable order book midprice. If a fresh price cannot be fetched, the system falls back to the last known good price and **flags the computation as stale** – the flag is published with the level, never hidden. The full computation is specified in [NAV Methodology](/indices/nav-methodology).
  </Step>

  <Step title="Compute the index level">
    The basket valuation (Raw NAV) is rebased so each series starts at an index level of **100** at inception and moves proportionally with the underlying basket. This is the published, citable number.
  </Step>

  <Step title="Handle resolutions">
    When a constituent market resolves, its settlement value (\$1 or \$0) folds into the level – the discrete, binary event is absorbed by the rest of the basket. Fixed-term series run until all constituents resolve; [Perpetual Series](/indices/perpetual-series) replace resolved markets on a scheduled review cadence.
  </Step>

  <Step title="Reconstitute with chain-linking">
    When a Perpetual series changes composition, the old and new baskets are chain-linked at the reconstitution point – the standard index technique that ensures composition changes never create artificial jumps in the published series. Every reconstitution leaves a full audit trail.
  </Step>
</Steps>

## NAV Windows: The Heartbeat of the System

NAV windows are the periodic checkpoints at which every series is revalued and its index level published. Think of them as an index provider's calculation schedule – but running every 30 minutes rather than once at the close.

| Property            | Detail                                                                                                       |
| ------------------- | ------------------------------------------------------------------------------------------------------------ |
| **Frequency**       | Currently every 30 minutes (subject to change)                                                               |
| **Inputs**          | Observable order book midprices for every constituent, captured at the window                                |
| **Staleness flags** | Any level computed on a fallback price is flagged stale in the published data – data quality is never hidden |
| **Continuity**      | Resolved constituents contribute their settlement values; reconstitutions are chain-linked                   |
| **Auditability**    | Every published level is reproducible from its window's inputs and the published formulas                    |

<Warning>
  Index levels are **theoretical values computed from midprices**, not execution prices. They assume constituent positions could be valued at the mid of each order book, without spread, slippage, or market impact. This is the standard caveat for any published index – and it matters more here because prediction market order books can be thin. See [Risk Factors](/risk/risk-factors) for the full discussion of theoretical-versus-executable limitations.
</Warning>

## What Moves an Index Level

<AccordionGroup>
  <Accordion title="1. Market movements">
    The underlying prediction markets move as new information arrives – polling data, economic releases, policy announcements, and other catalysts shift the implied probabilities of event outcomes. Each NAV window reprices all constituents using the latest order book data.

    For example, if a strong jobs report suddenly prices down the probability of rate cuts, all rate-related constituents in a rates series adjust – and the index level reflects those movements at the next window.
  </Accordion>

  <Accordion title="2. Market resolutions">
    When an event resolves, the constituent settles to \$1 (if the tracked outcome occurred) or \$0 (if it did not). This is immediately reflected in the level.

    Resolution events are the defining feature of event markets: they create discrete, binary value changes. A diversified basket absorbs these events across multiple constituents, which is what turns binary settlement noise into a continuous, readable series.
  </Accordion>

  <Accordion title="3. Reconstitution (Perpetual series)">
    On a scheduled review cadence, Perpetual series retire resolved or ineligible constituents and admit new ones under the published criteria. Chain-linking ensures the level is continuous across the change: the index measures the theme, not the turnover. See [Perpetual Series](/indices/perpetual-series).
  </Accordion>
</AccordionGroup>

<Tip>
  Index levels are computed gross of any fees. Fees exist only inside the private Alpha Program, where they affect participants' NAV per share – they never touch the published benchmark. This is the same distinction as an equity index (fee-free, theoretical) versus a fund tracking it (fee-bearing, investable).
</Tip>

## Data Integrity Controls

The same production system that computes index levels enforces a set of integrity controls, all designed to prevent a bad number from ever being published:

| Control                     | Purpose                                                                                                                       |
| --------------------------- | ----------------------------------------------------------------------------------------------------------------------------- |
| **Staleness flags**         | Levels computed on any fallback price are transparently flagged stale in the published data                                   |
| **On-chain reconciliation** | Constituent holdings are continuously reconciled against on-chain custody balances                                            |
| **Integrity halts**         | Automated halts if reconciliation detects any discrepancy – the series pauses rather than publishing through an inconsistency |
| **Resolution verification** | Market resolutions are confirmed against on-chain settlement state, not a single API                                          |
| **Audit trail**             | Every level, composition change, and reconstitution is recorded and reproducible                                              |

These controls may occasionally pause a series. That is by design – the system prioritizes integrity of the published data over continuous availability.

<CardGroup cols={2}>
  <Card title="NAV Methodology" icon="calculator" href="/indices/nav-methodology">
    The full computation, step by step – replicable with a spreadsheet.
  </Card>

  <Card title="Market Selection & Weighting" icon="filter" href="/indices/selection-and-weighting">
    The eligibility screens and weighting formula.
  </Card>

  <Card title="Worked Examples" icon="pen-to-square" href="/indices/worked-examples">
    Full numerical examples of every computation.
  </Card>

  <Card title="Perpetual Series" icon="infinity" href="/indices/perpetual-series">
    Reconstitution, chain-linking, and review cadence.
  </Card>
</CardGroup>