RESEARCH / 2026
Evidence,
under pressure.
Experiments on the distance between what a security record says and what it can actually prove.
INFERENCE / STATIC MODEL
EVIDENCEHELDCLAIMUNSUPPORTED
SELECTED WORK
Current research.
R-002ACCEPTED
RETRIEVAL
The tool said there was no evidence.
The evidence was there.
0 matches≠186 parsed≠192 raw
A healthy audit subsystem, a successful query, and the wrong conclusion. The failure was retrieval—not recording.
17 AUG 2026Read study →
S-001OPEN
ATTRIBUTION
Two users. One account.
Which one did it?
2 actors→1 transient bridge→?
Remove the signal that separates two actors sharing an identity. Attribute only if the remaining evidence earns it.
NO EMPIRICAL VERDICTRead study →
THE QUESTION
Where does the
evidence stop?
Four recurring boundaries. The studies do the talking.
01Authorization
identity ≠ intent
02Attribution
name ≠ actor
03Completeness
healthy ≠ complete
04Consequence
stopped ≠ reversed
GHALVERA / MADRID
What holds.
What doesn't.
Methods, revisions and unresolved limits stay with each study.