# Belief Index Data Dictionary Snapshot date: 2026-07-09 · Schema version: 1.0 Point-in-time datasets of Belief Index levels – rules-based benchmarks of market-implied event probabilities, computed from prediction market data and recorded live at each valuation window since inception. Methodology: https://docs.beliefsystems.xyz/indices/nav-methodology ## Files | File | Contents | | ------------------------------------------- | ------------------------------------------------------------------------------------ | | `belief-index-series.csv` | Series reference – the full universe, including resolved and terminated series | | `belief-index-levels_{TICKER}.csv` | One row per valuation window (native 30-minute cadence) per series | | `belief-index-compositions_{TICKER}.csv` | Basket membership and target weights per composition version, with effective windows | | `belief-index-reconstitutions_{TICKER}.csv` | Completed reconstitutions with chain-linking arithmetic | | `manifest.json` | Machine-readable snapshot metadata: row counts, coverage, SHA-256 checksums | ## belief-index-series.csv | Column | Type | Definition | | --------------------- | --------- | -------------------------------------------------------------------------------------------------------------------- | | `series_id` | string | Stable series identifier (join key across all files) | | `ticker` | string | Series ticker used in filenames | | `name` | string | Series display name | | `status` | enum | `ACTIVE`, `RESOLVED` (all constituents resolved; wind-down), `TERMINATED` (series closed), `ARCHIVED` | | `maturity_type` | enum | `FIXED` (resolves and ends) or `PERPETUAL` (maintained via scheduled reconstitution) | | `inception_at_utc` | timestamp | First valuation window; index level rebased to `base_index_level` here | | `base_index_level` | decimal | Index level at inception (100.00000000 unless stated) | | `inception_raw_nav` | decimal | Raw basket value at inception (current chain-link segment; see reconstitutions) | | `resolved_at_utc` | timestamp | When the series fully resolved (empty if not resolved) | | `terminated_at_utc` | timestamp | When the series terminated (empty if live) | | `mirror_of_series_id` | string | **If non-empty, this series is a mechanically derived inverse of the referenced series.** See "Mirror series" below. | | `composition_version` | string | Current composition version | | `methodology_version` | string | Current methodology version | ## belief-index-levels\_{TICKER}.csv | Column | Type | Definition | | ----------------------------- | --------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `timestamp_utc` | timestamp | Valuation window timestamp (see "Timestamps" below) | | `series_id` / `series_ticker` | string | Series identity | | `computation_id` | integer | Stable identifier of the underlying computation record (citable; quote it in support requests) | | `raw_nav` | decimal | Probability-weighted basket value, bounded [0, 1]. **Discontinuous at reconstitutions** – it re-bases when the basket changes. Use `index_level` for continuous series analysis. | | `index_level` | decimal | Published index level, rebased to 100 at inception and chain-linked across reconstitutions so composition changes never create artificial jumps | | `stale` | boolean | True if **at least one** constituent used a carried-forward price at this window. Coarse flag – use `fresh_markets` / `total_markets` for degree. | | `fresh_markets` | integer | Count of constituents priced from live order book data at this window (empty for early records where the breakdown was not stored) | | `total_markets` | integer | Count of constituents in the basket at this window (empty as above) | | `partially_resolved` | boolean | At least one constituent has resolved; its settlement value is folded into the level | | `fully_resolved` | boolean | All constituents resolved; the level reflects settlement values | | `composition_version` | string | Composition in force at this window – join to the compositions file | | `methodology_version` | string | Methodology in force at this window | ## belief-index-compositions\_{TICKER}.csv One row per (composition version, constituent). | Column | Type | Definition | | --------------------- | --------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `composition_version` | string | Version identifier (joins to the levels file) | | `effective_from_utc` | timestamp | When this composition took effect (series inception for the genesis version) | | `effective_to_utc` | timestamp | When it was superseded (empty = current) | | `condition_id` | string | Constituent market identifier (stable venue-level condition ID) | | `outcome` | enum | Which side the index tracks: `YES` or `NO` | | `question` | string | Constituent market question at composition time | | `weight` | decimal | **Target weight at composition time.** Realized per-window weights drift with prices between reconstitutions; target weights are re-applied at each reconstitution. | ## belief-index-reconstitutions\_{TICKER}.csv One row per completed reconstitution. These fields make the chain-linked splice independently verifiable: within a segment, `index_level = base_index_level x (raw_nav / inception_raw_nav)`; at a reconstitution the anchor level `chain_link_anchor` is carried across so the published level is continuous while `raw_nav` re-bases. | Column | Type | Definition | | ----------------------------------------------------- | --------- | ------------------------------------------------------------------------- | | `event_id` | string | Reconstitution event identifier | | `composition_version_from` / `composition_version_to` | string | Versions before/after | | `gate_activated_at_utc` | timestamp | Start of the reconstitution gate (level publication suspends; see "Gaps") | | `completed_at_utc` | timestamp | When the new composition took effect | | `old_inception_raw_nav` / `new_inception_raw_nav` | decimal | Chain-link segment bases before/after | | `chain_link_anchor` | decimal | Index level carried across the splice | | `base_index_level` | decimal | Base level of the new segment | | `added_markets` / `removed_markets` | list | Constituents added/removed, encoded `condition_id:outcome` joined by `;` | ## Semantics **Gaps.** A missing window means no computation was published – reconstitution gate, integrity halt, or outage. Levels are **never interpolated** and gaps are never backfilled. Reconstitution-gate gaps are attributable via `gate_activated_at_utc` → `completed_at_utc` in the reconstitutions file. **Restatement policy.** As-published values are never rewritten. If a correction is ever required it is published as a new, labeled entry in the `changelog` array of `manifest.json` – the affected rows are identified, not silently replaced. The snapshot pipeline enforces this: generation fails if any historical row differs from the previously published snapshot. **Timestamps.** All timestamps are ISO-8601 UTC with a `Z` suffix. `timestamp_utc` is the actual computation execution time at a ~30-minute cadence; it is **not aligned to a :00/:30 grid**. Align or bucket before joining against grid-aligned data. **Mirror series.** Series with a non-empty `mirror_of_series_id` are mechanically derived inverses of their primary (same constituents, opposite outcomes). **Do not treat a mirror pair as two independent observations** in cross-sectional statistics. **History depth.** Every series is recorded live from its inception – the archive grows at each valuation window and cannot be backfilled, because prediction market order books are not otherwise archived. Per-series coverage (first/last observation, row counts) is in `manifest.json`. **Levels are not execution prices.** Index levels are theoretical values computed from observable order book midprices; they include no spread, slippage, or market impact. See https://docs.beliefsystems.xyz/risk/risk-factors ## Format conventions - UTF-8, comma-separated, header row, RFC 4180 quoting - Decimals serialized as fixed-point strings (8 decimal places) - Booleans: `true` / `false` - Missing values: empty string