A laboratory, not a consultancy.
The Garel Lab is a private strategic analysis laboratory with an international reach. Neither a consultancy nor a software vendor: a house of applied research that designs structures, develops its own technological assets, and tests them on its own cases before handing them over.
What is at stake deserves a structure.
The laboratory's founding text comes down to three refusals.
- IThe single point of view
- IIThe untested structure
- IIIThe untraceable
Four habits, kept on every case.
We take the case, not the brief.
A case enters the laboratory whole: the documents, the constraints, the history, the room where the choice will be defended. We do not work from someone else's summary of a problem.
We build the tool, then we run it.
Every model we hand over runs first in our own production, on real cases. Nothing reaches a client that the house has not first put under load.
We attack our own structure.
Before anything leaves, someone in the house is tasked with breaking it: simulating the actors who will work around it, shaking the weightings, arguing the opposing side. What survives is what we hand over.
We deliver the argument, not the answer.
The deliverable is a document that can be reopened years later and still defends itself: every criterion sourced, every method's view recorded, every sensitivity declared, sealed.
Four disciplines, pursued as applied research.
Computation
Optimisation, sensitivity analysis, measured robustness for every conclusion. With each result we publish the smallest push that would overturn it.
Multi-agent simulation
Modelled actors, their interests, their constraints, their reactions, simulated at scale. A structure is judged by what people will actually do with it.
Game theory
Strategies, equilibria, coalitions: what each party has an interest in doing, and the rule that makes cooperation pay better than working around it.
Decision science
Every school confronted on the same case, their disagreements measured and shown rather than hidden behind a single number.
All four run in a single engine: Kernel.
Kernel is the laboratory's central asset. It models the actors, simulates their trajectories, looks for equilibria and confronts the schools of decision on the same case. Every structure we hand over has been tested in it.
ε* = min ‖Δw‖
A laboratory is also defined by the cases it turns down.
- Decisions already made
- Studies to justify a choice
- Conclusions ordered in advance
- Work that cannot be sourced
- Cases that cannot be modelled
- Volume before rigour
A small team, instrumented agents.
A small team of practitioners, supported by instrumented agents: humans decide what matters, machines carry out what must be done, and every action is traced and signed. We take on a limited number of cases at a time, the only way to work as described here.
Others advise. We structure.