SCALAR / METHODOLOGY · 02
One method, kept on the record.
Scalar publishes a single quantitative methodology and holds to it. Every output traces to its inputs, every model change is versioned and logged, and the latency the instrument is sold against is audited quarterly. This page is that record.
THE PIPELINE
Ingest, model, signal.
Ingest
The instrument reads primary-exchange feeds directly: no consolidated tape, no resampled vendor data. Every message carries the venue's own timestamp, and clocks are reconciled against the exchange at packet level. A feed that gaps more than 500 ms is flagged and excluded until it re-syncs.
PRIMARY VENUES ONLY · NO CONSOLIDATED TAPE
Model
Raw series are cleaned, aligned to a common calendar, and passed through the estimator set: realised-volatility windows, curve fits, dispersion measures. Parameters are fixed by the current model version and are never tuned per market. The same code path runs every asset class.
PARAMETERS FIXED BY VERSION · ONE CODE PATH
Signal
Estimates are normalised to the instrument's standard outputs: level, change, and percentile against the full record. Signals publish to the API and the terminal simultaneously; wire-to-signal latency is measured on every release and bounded at 80 ms.
LEVEL · CHANGE · PERCENTILE OF FULL HISTORY
THE MODEL REGISTRY
Every change, numbered.
Each change to the estimator set ships as a numbered version, reviewed and signed by all three desk leads before release. Prior versions remain reproducible from archived inputs.
| Version | Released | Change | Status |
|---|---|---|---|
| v4.2 | 2024-11-18 | Dispersion estimator moved to median-of-windows; outlier weight capped at 2.5σ. | CURRENT |
| v4.1 | 2024-05-06 | Roll-yield convention unified across commodity classes. | SUPERSEDED |
| v4.0 | 2023-09-25 | Estimator set rebuilt on venue-native calendars; v3 parameter file retired. | SUPERSEDED |
| v3.3 | 2022-03-14 | Inflation-surprise measure added for euro-area prints. | SUPERSEDED |
| v3.2 | 2021-06-21 | Latency bound tightened from 120 ms to 80 ms. | SUPERSEDED |
| ≤ v3.1 | 2014–2021 | Parameter history archived; every release reproducible from stored inputs. | ARCHIVED |
THE AUDIT TRAIL
Audited since 2014.
The methodology has been under external audit since 2014. Each quarter, wire-to-signal latency is measured against primary exchange timestamps in all 42 markets; each year, a sample of published outputs is re-derived from archived inputs. No audit has returned a material exception.
Q1 2014 → Q4 2024 · ONE TICK PER QUARTER
THE RESEARCH DESK
The desk behind the method.
Three people own the method, and their names are on the registry. The desk is small by design: every change passes the same three pairs of eyes.
HEAD OF METHODOLOGY
Dr. Ingrid Voss
Built curve and dispersion estimators at two European clearing houses; has owned the model registry since 2016.
REGISTRY SIGNATORY · JOINED 2016
HEAD OF MARKET DATA
Marcus Oyelaran
Fifteen years on exchange connectivity; runs the ingest layer and the venue-timestamp reconciliation the latency audits lean on.
INGEST OWNER · JOINED 2015
HEAD OF DELIVERY
Emilio Marchetti
Former proprietary-desk platform lead; accountable for the API, the terminal, and the 80 ms bound being a fact rather than a claim.
RELEASE AUTHORITY · JOINED 2018
NEXT · 03 / ACCESS
When the record is enough.
If the method answers the question, the next step is access. Requests are reviewed by a person on the desk, within two business days.