What we do
Turning fragmented, multi-source price data into a single, coherent picture.
Most markets don't have one price — they have many, one per venue, each updating on its own clock. That fragmentation hides useful structure: which venue is setting the price, which is lagging, and where the gap between them is real rather than an artifact. We make that structure measurable.
Multi-source ingestion & clock normalization
We pull time-stamped prices for the same asset from many venues and put them on one canonical clock — so feeds that stamp time differently become genuinely comparable, with each source's real range checked so a misalignment can't hide.
Fair-value reconstruction & de-margining
We rebuild a robust consensus price and strip out structural distortions — such as a stablecoin trading off its peg — so what remains is a genuinely like-for-like gap rather than an artifact.
Divergence, lead-lag & phantom filtering
We measure which venue leads and which lags on asynchronous tick data, quantify divergence in basis points, and discard frozen or uncorroborated quotes with a robust outlier guard before anything is called a signal.
Pre-registered validation & reproducibility
Every claimed structure is tested against a surrogate null with parameters fixed in advance, each component ships an offline self-test, and every result regenerates from free public data — the discipline is the deliverable.
Where it applies
One method, many markets.
Crypto — cross-exchange & basis monitoring
Two completed case-studies on one unchanged engine: BTC cross-exchange price discovery (efficient, yet provably cointegrated, replicated across six independent windows); the perpetual–spot basis across BTC, ETH and SOL (majors efficient, funding-tethering replicated across six windows, SOL's apparent difference window-specific).
Read the analysis →
Case studyEnergy — day-ahead vs real-time
The same engine on wholesale power: the day-ahead premium, and whether the real-time-minus-day-ahead basis is predictable once the daily rhythm is stripped out — a genuine, replicated structure, and the first non-crypto market on the unchanged core.
Read the analysis →
Case studySovereign rates — cross-market coupling
The same cross-source method applied to government bonds: how strongly US Treasury and euro-area AAA sovereign 10-year yields move together, measured across three independent seven-year blocks (2007–2026). Robust to removing the most volatile days — and honest that the dominant signal is same-day co-movement, with the one-day-lag piece reported as coupling, not leadership.
Read the analysis →
MethodologyRetail — fuel price dynamics
The engine pointed at ~9,600 French fuel stations, asking whether prices converge toward local competitors. A layered null stack — mechanical and staleness controls, a synchronized-feed de-duplication, and a geography placebo — found the apparent convergence is not specifically local, and surfaced a real synchronized identical-price sub-finding. A methodology demonstration of the engine rejecting a prior-confirming signal, not a delivered capability.
Read the analysis →
MethodologyCloud infrastructure — spot pricing
Pointed at AWS spot prices across availability zones: does one zone revert because it diverged from the others, or because all zones track a shared regional factor? The engine established — before collecting the test window — that this design cannot separate the two (a reflection problem), and closed it there rather than ship an answer it could not stand behind. Measurement only; identification honesty is the point.
Read the analysis →
Execution-quality & multi-venue monitoring
For anyone trading or settling across venues: which venue leads price, where fills leak basis points, and when a basis widens. Measurement and monitoring only — methodology, never advice.
Case study = built and shown. Research direction = an honest extension, not yet delivered work. Methodology = capability we apply, kept at research level.