The Physical Evidence Gap
Why digitally valid evidence is not enough for high-consequence systems.
Read the perspective ↗
4SI Research
Evidence before
assertion.
Independent research for advanced AI and high-consequence systems—beyond model-only analysis.
Research agenda
Four questions
define the work.
The program examines where physical evidence can change institutional control—and where it cannot.
- 01
Expected human
How can a system establish that the expected human is physically present for a specific action?
- 02
Expected object
How can a physical object remain verifiably bound to its claimed origin, custody and state?
- 03
Renewed authority
When should an intelligent system be required to obtain renewed, physically grounded authority?
- 04
Material impact
Which loss pathways can physical verification materially change—and at what cost?
Published work
One research program.
Multiple records.
Visual editions improve accessibility. DOI-backed and versioned records support durable citation.
Verification as a Control Primitive
A physically grounded authority condition for systems approaching consequential execution.
Read the visual edition ↗Physical Trust at the Boundary
A systems view of people, objects and action where digital certainty is insufficient.
Read the visual edition ↗The Economics of Physical Verification
Loss pathways, intervention points and the cost conditions under which an added boundary may be justified.
Open the research note ↗Scientific papers and persistent records
The full publication library, source index and citable records.
Open the library ↗Live source-backed signals
Connect the research thesis to current incidents, positions and scenario analysis.
Explore Intelligence →Research standard
Make the boundary
of every claim visible.
Research quality depends on provenance, separation of fact and inference, explicit limitations and durable records.
Source
Prefer direct, official and persistent records. Identify provenance and date.
Separate
Distinguish documented fact, inference, proposed control logic and future capability.
Bound
State assumptions, boundary conditions, limitations and unresolved failure modes.
Preserve
Publish versioned records and DOI-backed papers where durable citation matters.
Correct
Keep material corrections visible and connected to the original record.
Test
Translate the claim into a question that a controlled workflow can answer.
Research collaboration
Bring one boundary worth testing.
We collaborate where a defined institutional problem can produce evidence useful beyond a single implementation.