Tools
The research generates instruments. The instruments get built.
Every workbench here came out of a specific piece of research and answers one question that could not be answered by reading it. Each runs on real data and each is open — no sign-in, no gate.
They are working tools rather than finished products. Several are actively being developed, some are narrower than the research behind them, and a few would need real engineering to run against a live book. That is the honest state of them, and the direction of travel is toward more of this and less commentary: research that cannot be run against data is just a view.
1 · Energy and compute
A data centre is a load shape before it is anything else, and a GPU-hour is electricity plus a margin. These two decompose the physical side of the compute complex.
What am I actually exposed to across an energy book?
430+ atomic risk factors across 470+ contracts on ICE EU/US, NYMEX-CME and Nodal, decomposed into dated factor legs with the full ISO → hub → zone hierarchy, cross-exchange offset detection and a worked two-portfolio case study.
Runs on: Contract specifications and settlement hierarchies across four exchanges
What does a GPU-hour cost once its electricity is priced properly?
The compute supply curve through to PJM, ERCOT, WECC and CAISO basis: spark spread, heat-rate mapping, take-or-pay versus spot optimisation, index decomposition and trade synthesis across fourteen modules.
Runs on: Regional power curves, GPU rental indices and measured device power
2 · Margin and position sizing
Margin is where unlike risks meet. These decide what a book costs to carry, and how large a position should be once that cost is known.
What does one portfolio-margin engine do with two unlike risks?
BTC and SPX perpetuals against event-contract binaries in a single framework — an eight-cluster Monte Carlo over 5,000 paths, an implied-volatility solver and surface, cluster expected shortfall, concentration floors and a venue-aware margin comparison.
Runs on: Live Kalshi and Polymarket books, with a simulated path engine
How large should this position be, given what it consumes in margin?
Kelly and fractional-Kelly sizing solved against the margin framework rather than after it, so position size and margin consumption are decided together instead of in sequence.
Runs on: The margin engine above, plus venue fee schedules
3 · Cross-venue and structure arbitrage
The partition rule is the sharpest tool in the event-contract section: a mutually exclusive, collectively exhaustive set must price to one. These three look for where it does not.
Where does the same question print at two different prices?
Kalshi against Polymarket against Manifold on the same contract, with the fee and slippage model that decides which of those gaps actually survives execution.
Runs on: Live venue quotes across three exchanges
Do the outright and the match-path prices agree with each other?
Tournament-structure arbitrage where the partition rule has real teeth — outright versus match-path dispersion, built on live venue quotes.
Runs on: Live venue quotes, tournament bracket structure
What per-game probabilities does a series price imply?
Best-of-seven series maths into implied per-game probabilities, with a dutching detector and LP-optimal portfolios across Polymarket and Kalshi.
Runs on: Live Polymarket and Kalshi series markets
4 · Things you can take away
Not everything should live on this site. This one runs on your machine, against products you are actually being sold.
Built for a desk
Most of these started as someone's problem.
If one of these is close to what you need but pointed at the wrong market, or you have a book that wants decomposing and no tool that does it, that is a normal way to start a conversation. The same applies in reverse: if a tool here is wrong, the argument behind it is on this site with its method stated, and corrections are welcome.