Index Governance

A benchmark is only as useful as the discipline behind its publication. Before an institution cites an index in a risk report, references it in research, or licenses it as an underlying, the questions it asks are not about the formula – the formula is published – but about administration: Who can change the methodology, and how? Are published values ever rewritten? What happens when the data is bad? Will the history still mean the same thing in five years? This page answers those questions in one place.

​

Methodology Versioning

Every published computation records the methodology version it was produced under (currently midprice-v1).

• Revisions are dated and prospective-only. A methodology change applies to future computations, never retroactively to published history. The same rule governs selection and weighting thresholds: revisions apply to future selections, never to existing compositions.
• Changes are announced and documented. Any methodology change is published at least 5 calendar days before it takes effect, with a dated entry in the Governance Log; the version history is published in Index Calculation Methodology.
• Judgment is bounded, and only ever narrows. Where human judgment enters index construction – theme definition and the documented judgment screens – it is applied under published criteria, and approval of any composition change can only narrow a proposal: a candidate that passed the published screens may be excluded with recorded rationale, but a candidate that failed them can never be included. There is no discretionary adjustment of a live series outside the rules-based process below; every proposal, approval, and exclusion is recorded.

​

The Level Calculation

Each index level is computed from constituent market prices and the weights in effect at each computation: level = Σ(weight × price). Weights evolve under published rules – the scheduled review process below and renormalization at settlement exits. Composition downloads carry target weights. The weights each computation actually applied – its effective weights – differ from target when a constituent has no price at that computation and the others are scaled up to cover it, or when a computation priced a composition other than the one it is labeled with; every such computation with a stored per-constituent breakdown is listed, all constituents stated, in the series’ effective-weights file, published with the composition downloads for each series from its first reviewed snapshot (October 2026 at the earliest). Prices are taken from constituent order books at the moment of computation; because those books are not archived by the venue, contemporaneous reproduction is exact, while after-the-fact reproduction depends on the reader’s own price record – see The Point-in-Time Archive. From October 26, 2026, each market in a Perpetual index is priced once per computation. An index holding NO uses one minus the YES price. Composition membership is determined by the governance process on this page – never by execution events.

​

Constituent Selection: Inclusion and Maintenance

Inclusion criteria gate what may be added: order-book depth and bid-ask spread floors, a volume floor, a price band, a resolution-date window, market-state and category screens, and an investability floor – all published and computed from public market data. Candidate discovery is driven by that mechanical, published screen, so a third party can reproduce the quantitative candidate pool; the documented judgment screens are then applied to that pool. Belief Systems generates composition proposals from the screened pool; no change takes effect without named-individual approval recorded against it. Maintenance criteria, effective from the date recorded in the Governance Log, gate what may remain, and are deliberately looser than inclusion criteria so borderline markets do not oscillate in and out: depth and volume floors at 50% of the inclusion levels, a per-side depth minimum, and a spread cap at twice the inclusion cap – always computed on the same measures as the then-current inclusion criteria. There is no price-band test for continued membership: a market pricing toward certainty is the index working. A constituent that fails maintenance criteria at two consecutive evaluated monthly reviews is deleted at the review that records the second failure. A review at which a constituent could not be evaluated neither adds nor clears a strike. Resolved markets are not deletions – they exit through settlement. From 2026-09-30, a constituent whose market has resolved keeps its weight at its settlement value until the next scheduled review. At that review it leaves the composition, the remaining constituents’ weights are scaled up in proportion (a review that also adds markets re-weights the composition), and the level is chain-linked across the change. The exit is part of the review’s approved composition change; unlike a deletion candidate, a resolved constituent cannot be kept. A constituent leaves only once its venue’s final settlement record reports the outcome; a market the venue has closed but not yet settled, or whose data is temporarily unavailable, stays a constituent at its current weight until then; while it waits it is not tested against the maintenance criteria and its strike count is carried unchanged. Missing data never removes a constituent. Continued eligibility. A constituent must also keep passing the published screens that describe the market itself: directional fit (including the status-quo exclusion), binary outcomes only, one horizon per event, resolution clarity, and the category screens. A constituent that no longer passes one is deleted at the next scheduled review, with no two-review wait, because a market’s wording and resolution rules do not fluctuate the way its liquidity does. Removals happen only on published grounds – the two-review maintenance rule or a named eligibility screen – and never on price, performance, or level. Each removal states its ground. The approver may keep a deletion candidate with a recorded reason, but may not remove a constituent on any other ground at a scheduled review; that is an extraordinary deletion, disclosed the same day. Determinations. Each review’s determinations – every constituent evaluated, its result, its strike count, and the ground for any deletion or retained candidate, with each settlement exit recorded as such – are published with the versioned composition downloads after the review. For paired indices that track opposite sides of the same markets, a market is admissible only if both legs pass inclusion criteria, enforced before any change executes. Membership follows the approved change, never the fills: an execution shortfall appears in published weights, not as a silent membership edit. After each pair’s one-time re-weighting at the November 2026 review, both indices in the pair use the same weights. One side’s weights can’t be changed without the other’s. One index in a linked pair cannot be archived or ceased on its own.

​

Composition Changes and Disclosure

The monthly review window. Every Perpetual series is reviewed against its criteria during the first seven calendar days of each month, at the review cadence committed at series creation (default: monthly). Fixed (event-dated) indices have locked compositions and are not subject to review. The commitment is to the review, not to change: a review that concludes no change is the normal case, and a reconstitution executes only when the review produces an approved change. Changes, when they occur, take effect within the window and are published same-day in the composition record: the Holdings table on each index page reflects the change at effectiveness, and the versioned composition downloads carry full before/after composition, weights, and effective timestamps. Routine review results are not written up in the Governance Log, which records methodology changes, corrections, cessations, and any composition change made outside a scheduled review – the same split established index administrators keep between constituent-change notices and their governance record. Scheduled composition changes are disclosed at effectiveness, same-day, rather than in advance, on three grounds. First, no informational asymmetry is created: the quantitative criteria are mechanical and computed from public data, so anyone can run the published screens against the public order books and derive the likely candidate pool independently – non-disclosure of the specific list conveys no private advantage. Second, same-day disclosure is more complete than the traditional-index norm: full before/after composition, weights, and effective timestamps. Third, pre-announcing specific changes in low-liquidity books invites order-flow distortion around the effective date, degrading the very prices the index measures at the moment they matter most. This stance is re-evaluated at each periodic methodology review, and immediately upon any executed data license permitting replication or any publicly disclosed third-party replication.

​

Corrections Policy

As-published values are preserved as published. If an error in a published level is identified, the correction is issued as a new, labeled revision – the original value is not silently overwritten. Every level served by the API carries a revision marker – 0 as first published, 1 or higher once corrected – and corrected levels are published in a level-revisions file with the composition downloads from October 2026, while the level files keep every value as first published. Specifically:

• A level error greater than 10 basis points at any published point is corrected; smaller errors are annotated only.
• Corrections within 5 calendar days of publication are published as labeled revisions of the affected points, with an entry in the Governance Log. After that window, the published level stands and is annotated rather than revised.
• Resolution reversals and venue disputes are handled under this policy.
• No published value is ever changed silently, and no published value is ever removed.

This mirrors the discipline of the underlying accounting, where records are never edited or deleted; adjustments are made through new, labeled entries. The practical consequence for data users: the series you observe at any point in time is the series everyone observed. A backtest run against Belief Index history is a backtest against what was actually published – no silent revisions, no hindsight.

​

Publication Integrity

The pipeline is designed to disclose data-quality problems rather than publish through them:

​

The Point-in-Time Archive

Prediction market order books are ephemeral: quotes are not archived the way equity prices are, and resolved contracts disappear from venue interfaces. A themed probability series reconstructed after the fact is contaminated by hindsight – which markets to include, how to splice across resolutions, which prices to trust are all choices made knowing the outcome. Belief Index levels are computed and recorded as of each valuation window, live, under the versioning and corrections discipline above. The result is a point-in-time record of how markets actually priced each theme – with no lookahead, no survivorship filtering, and no retroactive splicing. Each series’ full computation history is preserved, including staleness flags and methodology versions, so a future reader can know not just the level but the data quality behind it. This archive is treated as the crown jewel of the operation. It compounds with every published window, and no part of it can be rebuilt later.

​

Data Sources

Levels are currently computed from order book data supplied by a single prediction market venue. This is disclosed plainly rather than obscured: see Single Data Source for what it means for data users. The methodology itself is venue-agnostic – selection screens, weighting, and valuation are specified in terms of observable order book quantities, not venue-specific features. Incorporating additional venues as they meet the eligibility screens below is the natural evolution of the methodology, and any such change would follow the versioning discipline above.

​

Venue Eligibility

A market can enter an index, a volatility series, or the graded universe only if its venue meets the screens below. These screens are part of the methodology: adopting a venue for a product family is a methodology change, announced at least five calendar days before it takes effect and recorded in the Governance Log. A venue is eligible when all of the following hold:

1. Regulatory status is public, and we state it wherever we identify the venues behind a published value.
2. Trading rules and settlement procedures are published, including how a market resolves and when that resolution is final.
3. Prices and order books are publicly readable, with identifiers that do not change over a market’s life.
4. The fee schedule is published.
5. No venue influence. No venue has any input into methodology, criteria, composition, or grades; a venue is never a customer of the grades that cover its markets; and no arrangement with a venue is exclusive as to source data.

A venue also needs enough public trading history for us to evaluate it against these screens before any published value depends on it. These are venue-level screens. Whether an individual market is good enough to use is a separate question, answered by the selection criteria and, for market quality, by Market Grades.

​

Conflicts of Interest

Belief Systems is the sole administrator of its indices: composition changes are proposed, approved, and effected within Belief Systems, without external segregation of duties. The compensating controls are structural:

• Discretion is bounded by construction – mechanical published criteria over public data, approval that can only narrow proposals, and a recorded audit trail for every action.
• Execution and abort of composition changes require hardware-key authorization, run inside gated change windows with automated integrity checks, and are recorded against immutable composition versions and append-only internal records.
• Belief Systems holds no positions in the constituent markets it measures – no proprietary, speculative, or replication holdings, and no position expressing a view on any outcome. Index values are calculated from observed market prices alone; no trading by the administrator stands behind any published level. Historical operational holdings from a discontinued internal replication program were fully wound down in August 2026.
• Data-licensing and index-services revenue is fixed by contract and does not vary with the direction or level of any index. Other arrangements are described in the Terms of Service. No fee arrangement is an input to composition or to the published level, both of which are determined solely under the published criteria.

​

Cessation and Continuity

• Fixed (event-dated) indices terminate by design. At resolution, the final level is fixed and archived; the index page, API, and downloads mark the index resolved and continue to serve full history. History of a ceased index remains publicly downloadable for at least five years after cessation, in a machine-readable format that may change, except where prevented by law or events outside Belief Systems’ reasonable control.
• Perpetual indices do not terminate by design. If every constituent of a Perpetual series has resolved, the constituents keep their settlement values, the level stays fixed, the series is shown as resolved, and it is considered for discontinuation under the cessation provisions below.
• Discretionary discontinuation of a live index requires at least 30 calendar days’ notice with published rationale, and history is archived as above. An index referenced by third-party obligations is not discontinued without the re-evaluation process described under disclosure timing.
• Belief Systems is a small administrator with concentrated key-person dependency, stated plainly. Historical data and the full methodology are archived in public, durable form such that users are left with a complete, reproducible record in any circumstance; any future licensing agreement will include a named continuity and notice term.

​

Belief Volatility

This policy governs both index families – Belief Index and Belief Volatility – under one changelog and one notice clock. A second register would be a second place to forget. Belief Volatility measures how much prediction-market probabilities are expected to move over a stated horizon. It is a benchmark data product: there is no fund, no share class and no money path behind it. Its full method is published in the Belief Volatility Methodology.

​

What counts as a methodology change

For Belief Volatility, each of the following is material and carries notice on the same terms as an index methodology change:

• a refit or replacement of the information clock (the release-curve artifact), including a clock_version bump;
• a change to the de-vig convention or the traded-marginal estimator;
• a change to any published gate parameter – the depth floor, the spread and impact tests, the arbitrage tolerance and its size floor, the freshness gates, or the flag and suspend tolerances on the consistency checks. Every one of these is served in the gates block of the public series endpoint, so whether a notice was owed is checkable against the published record rather than taken on trust;
• a change to the tenor set, or to which tenor the bare ticker denotes.

​

Event-forced changes

Some changes cannot carry prior notice because an outside event forces them – a constituent market resolving, or a venue delisting one. These are disclosed same-day, with the reason stated by reference to the published criteria and naming the observable fact that forced the change.

​

Withholding, and why values are never adjusted

A reading that fails a publication gate is withheld, not repaired. There is no clamping, no smoothing and no carrying a stale value forward under a current timestamp. A withheld reading is labeled as withheld; the last good reading is labeled as last published. A reading that publishes with a disclosed condition is labeled with that condition, and the label travels with the value wherever it goes.

​

Retirement

A composite series covers a set of mutually exclusive outcomes. When enough of those outcomes resolve that fewer than three remain live, the composite retires: its combined value is withheld permanently and its per-leg records continue under the same ticker. Retirement is prospective and one-way. It is disclosed when it happens, it is recorded as a cessation event under this policy, and there is no path back – so the condition that triggers it requires a sustained run of computations rather than a single reading, and a briefly illiquid market cannot cause it. Published history is never restated on retirement, or on any other methodology change.

​

Complaints, Review, and Audit

• Complaints about index determinations or this policy can be sent to contact@beliefsystems.xyz. They are acknowledged promptly – in the ordinary course within five business days – and recorded internally.
• The methodology is reviewed periodically, with a published note stating either that no changes were made or what changed. The periodic review also re-evaluates the disclosure-timing stance above and recalibrates selection thresholds against accumulated data.
• As usage grows, Belief Systems intends to establish an independent oversight function consistent with IOSCO Principle 5. No such committee exists today.
• There is no external audit at current scale. The complete audit trail – immutable composition versions, append-only records, execution event records, and the proposal and approval log – is retained and, on reasonable request and subject to confidentiality undertakings, is made available for licensee due-diligence review, redacted for non-public and personal information. External audit will be sought at the earlier of institutional licensing requiring it or material third-party replication.

​

Reference Framework

The administration practices on this page – published methodology, versioned and prospective-only changes, a corrections policy, data-quality disclosure, and documented use of judgment – are informed by the IOSCO Principles for Financial Benchmarks, the international reference framework for benchmark administration. Belief Systems does not claim formal compliance with, or external assurance against, those principles; they serve as the design benchmark the administration practices are built toward as the index family matures. Belief Systems is not an authorized or registered benchmark administrator under the UK or EU Benchmarks Regulation, and Belief indices are not offered for use as benchmarks by supervised entities in the UK or EU. About this policy. This page is a statement of practice, not a contract: it creates no contractual right, warranty, representation, or duty owed to any person, and index data is provided on the terms set out in the Terms of Service and Disclosures. Belief Systems may amend this policy; amendments are dated and recorded in the Governance Log, and material amendments are published at least five calendar days before they take effect. The time periods on this page are ordinary-course commitments – they may be exceeded where compliance is prevented by law, regulation, venue outage, or other events outside Belief Systems’ reasonable control, in which case the reason is published.

​

Further Reading