Scientific Analytics Alliance · Evidence
Validation recordRealized backtestMethodologySample resultsMonitor status
Methodology: what is calculated, what is not
Version: 2026-11-17
Annual figures for crypto
From 2026-11-14 (UTC), new Factor & Dependence and Portfolio What-if calculations use 365 calendar days for applicable annual figures, matching Portfolio Risk. Crypto trades every day.
Saved results using 252 trading days remain as calculated. They are not converted or recalculated. Compare annual figures only when their scale, method and observation period match.
Changing the annual scale changes annual volatility even when the daily returns are unchanged. This change does not demonstrate a change in the asset's risk.
01 · Defining the problem
A risk figure is only useful if an analyst can explain what it measures, where it came from, and what it does not prove.
- Scope: for a selected set of digital-asset exposures, where is vulnerability concentrated, how did the composition behave under historical stress, which conclusions does the evidence support
- State of the art: blockchain-analytics tools are built for tracing and attribution; general-purpose risk models were built for equities and treat missing data as zero
- Both deliver numbers without basis, coverage or limits
02 · The solution
Existing research capabilities, one bounded workflow.
- Asset analysis and portfolio risk attribution
- Historical crisis replay over fixed windows, coverage disclosed
- Observed liquidity depth from a licensed source, with observation times
- Documented ecosystem relationships, kept separate from price co-movement
- Calculation basis, observation period, data coverage and limitations shown with every result
The objective is not trading signals: it is to separate a measured vulnerability from a model assumption, a statistical relationship, or an unanswered question.
03 · Historical stress
Crisis windows: the change at the end of each window against the deepest drawdown inside it.
- Fixed historical windows, same composition, same reconstruction method
- Coverage shown, including positions without sufficient history
- Reconstruction of past stress, not crisis prediction
04 · Liquidity
Observed order-book depth within 2% of the mid price, across the admitted venues of a licensed source.
- Observation time shown for every figure
- Share of observed depth a given position would use
- Cost inside the band: not established
- A depth figure is not an executable exit
Market data: CoinGecko
05 · Ecosystem
Documented relationships are kept separate from statistical price co-movement.
- A link to a company or fund does not establish the amount of its exposure or the loss it would suffer
- Structural evidence and price relationships are shown as two different things
06 · The evidence
From a headline finding to its basis: dates, sources, methodology, stated limits.
- Missing observations remain visible as gaps
- Insufficient evidence is not presented as low risk
- No causal, multi-step contagion is calculated
- Historical stress reconstruction is not crisis prediction
07 · Technical approach
Computation is separated from interpretation. The result is evidence-bound.
- Every figure carries date, source, method version and limits; the engine withholds any component with an incomplete input basis instead of estimating it; a language model is never the numerical authority
- ARIN22 core: deterministic computation and replay. The claim is reproducibility, not predictive validity
- Known limits: data-licence scope, short history for low-frequency series, no intraday data; addressed by licensed sources with provenance, declared coverage and research-only labelling
How a figure is produced
- Sources: Licensed market data (CoinGecko, attributed); Kraken derived quantities only; the user's own statements for the derivatives check
- Admission: Only assets that pass the admission criterion are calculated; the list is versioned and dated
- Computation: ARIN22 core: deterministic computation and replay
- Withheld gates: A component with an incomplete input basis is withheld, never estimated
- Result: Figure with its basis, coverage, method version and limits
Not claimed
- There is no intraday data.
- No trading signals or recommendations are issued.
- Daily tail figures: calibration not established, not approved for limit setting (see the Validation record).
Asset universe
Asset Brief and full asset analysis: 17 SAA Qualified assets (universe v10, 18 Oct 2026); a further 81 assets are analysable with partial coverage.
Portfolio base measurement: 98 assets with a verified 400-day daily price history (release list v1, 5 Nov 2026), offered within each plan's own universe.
Tokenized stocks and ETFs: the instruments admitted by the risk engine (14 as of 17 Nov 2026: 7 xStocks and 7 Ondo tokens) can be added to a Portfolio check composition next to crypto, in plans that offer them. The admission list is versioned and dated.
Validation record: a separate research set of 61 assets, pooled for the prospective record; it is not the client universe.
Data sources
Market data: CoinGecko, with attribution. Kraken: derived quantities only (percentages and indices), never raw quotes. Binance's own API and exports are not used for display; venue-level depth figures come from CoinGecko under its licence.
Tokenized stocks and ETFs: each token's own CoinGecko daily closes; no other asset's history stands in for a token. A token's price return is not the holder's total return: depending on the issuer's documents, dividends are reinvested in the price or paid through the token balance.
Each module states its own history window. Example: the monthly 95% estimate uses nonoverlapping monthly observations (109 as of 4 Nov 2026, in the sample result).
Statement files for the derivatives exposure check are read in your browser for one calculation and are not uploaded; only the confirmed figures are sent. File name, SHA-256 and the lines used are kept with each figure as provenance.
What each figure carries
Every figure carries its basis (source and date), its coverage (which part of the position or history is covered), the method version and its limits.
A component with incomplete input is not estimated: it is marked withheld, with the name of the missing input.
When a figure is withheld
A calculation is not issued when its inputs or checks are insufficient. In plain words:
- Derivatives exposure: part of the positions is not recognised (partial exposure); open orders in the statement; the maintenance requirement is missing from the export (no liquidation estimate, only the cushion exhaustion point); spot and derivatives statements are from different times; basis mismatch; the scenario margin model is unavailable.
- Monthly 95%: not enough months; numerical bounds violated; tests not passed.
- Asset Brief: insufficient history; source data incomplete.
Validation status
Numerical verification: the calculation is deterministic and reproducible. A public results-only evidence repository shows 160 of 160 rows inside the 5% oracle envelope.
Market validation: a prospective record of research models (see the Validation record); no model has graduated. The monthly 95% monitor started on 6 Nov 2026; a verdict comes only after months accumulate (see Monitor status).
Determinism and replay
Each result is reproducible from its manifest and receipt; a result package carries its as-of date, method version, inputs and statuses.
Change log
2026-11-17: tokenized stocks and ETFs in Portfolio check added; section numbering corrected; percentages written without a space.
2026-11-14: annual figures for crypto use 365 calendar days.
2026-11-06: approved by the owner; asset universe roles and statement handling added.
2026-11-05: first version.
Hypothetical and saved research results only · Not investment advice
As of 17 Nov 2026.
Research information only. Not investment advice, not a recommendation, not a solicitation to buy or sell any asset. Figures are computed from the stated sources as of the stated date and carry the limits shown with them. Where the evidence is insufficient, the figure is withheld. Market data: CoinGecko. Scientific Analytics Alliance Inc., Delaware.
MethodologyPricingMarket data: CoinGecko
Math first · Agents second · Governor always · Audit forever