Functional decisions made at the program level —
about data, architecture, access controls, model design —
compound quietly over time.
They do not show up in delivery reports.
They show up later, in security failures, bad targeting, corrupted records,
and outcomes that affect the lives and livelihoods of the people the system was built to serve.
The US Government — and the DoD in particular — deserves better than that.
So do the people depending on it.
The problem
The DoD and its people are among
the most mission-driven in the world.
The AI systems built to support them
are not always worthy of that.
the most mission-driven in the world.
The AI systems built to support them
are not always worthy of that.
Why it matters
The people affected should be
named first —
before the technical argument.
named first —
before the technical argument.
Service members. Veterans. Civilian personnel.
People whose benefits, records, and safety
run through systems nobody fully audited.
That is the ethical concern. It is also the technical one.
These are not separate.
This is not a political position.
It is not about being anti-AI or anti-contractor.
It is about the measurable gap between
what was purchased and what was actually delivered —
and what that gap costs the people inside the system.
People whose benefits, records, and safety
run through systems nobody fully audited.
That is the ethical concern. It is also the technical one.
These are not separate.
This is not a political position.
It is not about being anti-AI or anti-contractor.
It is about the measurable gap between
what was purchased and what was actually delivered —
and what that gap costs the people inside the system.
What we do
Independent analysis of where —
and by how much — the governance
structure of AI systems employed
in process decision-making
risks mission-critical failures.
and by how much — the governance
structure of AI systems employed
in process decision-making
risks mission-critical failures.
Every major federal AI program has contractors who built it
and contractors who maintain it.
None of them are paid to tell you it is not working.
Margin of Error has no stake in what was purchased.
We annotate the impact of systems already in use —
modeling failure modes, auditing real-world behavior,
and quantifying what the original delivery never measured.
That independent position does not exist anywhere else
in federal subprime contracting.
We are building it.
and contractors who maintain it.
None of them are paid to tell you it is not working.
Margin of Error has no stake in what was purchased.
We annotate the impact of systems already in use —
modeling failure modes, auditing real-world behavior,
and quantifying what the original delivery never measured.
That independent position does not exist anywhere else
in federal subprime contracting.
We are building it.