An edge is a claim about curvature, not just direction.

The gradient of a function points downhill. It is first-order information, and it is what almost every method uses. It says nothing about how quickly the slope is changing, in which directions it changes fastest, or whether the point you have arrived at is a genuine minimum or just somewhere the ground happens to be flat.

The Hessian, the matrix of second derivatives, answers exactly those questions. It is the local shape. First-order information is what everyone uses; second-order information is what tells you how wrong you can be.

The same object turns up in four places this firm works, and they are the same mathematics wearing different clothes.

First-order questions ask which way: is this up or down, does the strategy work, what was the return. They are the questions everyone asks, and they have the useful property of always producing an answer. Second-order questions ask what shape the answer sits in: how stable is it, in which directions is it fragile, and what is it resting on that has not moved yet.

A number reported without its curvature is not a finding. It is a coordinate. Two research processes can produce the same coordinate while one of them has learned something and the other has not, and nothing in the number distinguishes them.

Three consequences follow, and they are the same three whether the object is a derivatives book, a fitted model or a business decision.

Being wrong has structure

Errors are not scattered evenly. They concentrate along the directions the analysis was flattest, because those are the directions nothing in the data pushed back on. Knowing where those directions lie is worth more than another decimal place on the estimate, and it is a different question from how large the error is.

Interaction is where things break

A book flat on every individual risk can still be badly exposed to two of them moving together. The same holds for a research process: assumptions rarely fail one at a time. The off-diagonal terms are the ones nobody reports, and they are the ones that decide what happens on the day several things move at once.

A result that could not have been wrong is not a result

Stating in advance what would count as a negative finding is the only thing that makes a positive one mean anything. A test that could only ever have confirmed the hypothesis has measured the procedure, not the market.

None of this is exotic. It is standard practice that gets skipped because first order is cheaper, arrives sooner, and is usually close enough. The firm's position is that usually is doing a great deal of work in that sentence, and that the exceptions are not randomly distributed: they cluster exactly where the second-order structure was never computed.

Where curvature is badly conditioned (a long, narrow valley, which is the normal shape of a calibration problem), first-order methods crawl along the flat direction and overshoot the steep one. The step size that is stable across the narrow axis is far too small for the long one.

The same structure appears in execution. An optimal schedule trades off market impact against timing risk, and the classical Almgren-Chriss formulation solves exactly that: a quadratic objective whose curvature sets how aggressively the order should be worked. Scheduling against VWAP, participation or implementation shortfall benchmarks is the applied version of this argument.

Newton's method uses the Hessian to choose the step instead of guessing it, which is why it does not zigzag. On a quadratic it arrives in one step. Real objectives are not quadratic, but they are locally close enough that the same information is what makes the difference between a fit that converges and one that wanders.

The Hessian of the log-likelihood at its maximum is the observed Fisher information, and its inverse is the asymptotic covariance of the parameter estimates. Stated plainly: the curvature of the objective is the uncertainty on what was fitted.

This is the overfitting argument made exactly rather than rhetorically. Two calibrations can report the same point estimate and the same in-sample fit while one of them has genuinely identified a parameter and the other has not. Nothing in the reported level distinguishes them. The curvature does.

Classifying a critical point is a statement about the eigenvalues of the Hessian there. All positive is a minimum; all negative a maximum; mixed signs a saddle: a point that is uphill in one direction and downhill in another, and therefore not a place anything stays.

A parameter set that sits on a sharp ridge in backtest space is a saddle. It looked like an optimum because the search only ever moved along directions where it was one. It will not be there next year, and no amount of in-sample performance says otherwise.

A firm without a track record is asking to be judged on something. This is the something: a stated method, precise enough to be argued with.

In practice it means the second-order structure gets computed and reported rather than assumed: the full risk matrix rather than its diagonal, the curvature of a fit alongside its point estimate, the conditioning of an optimisation rather than only its final objective value.

It also settles several things about how the firm operates, which would otherwise be matters of taste.

Method over outcome

An outcome is first-order information. A good result from an unsound process is a coordinate that happened to land well, and it carries no information about the next one. Work is therefore presented so that the method can be audited independently of whether the answer was welcome.

Negative results are deliverables

A flat likelihood is a finding: it says the data never identified the parameter. An engagement that establishes a strategy does not work has produced something of value, and it is reported as a result rather than buried as a failure to deliver.

Knowing the shape of the firm's own competence

The same argument applied inwards. Work outside where the method is strong gets declined rather than accepted and learned on a client's budget. That is why a scoping conversation can end with a recommendation to go elsewhere.

No track record

A track record is a level. It is first-order information about a path that will not repeat, and presenting one would contradict the argument the rest of this page makes. What is offered instead is a method stated precisely enough to be argued with, which is the only claim that survives the standard being applied to itself.

It is skipped right up until two things move at once.