Research
The sufficiency problem,
measured.
Context can fail an agent in five ways. GammaLex publishes the research behind detecting each one, so the outcome can be checked, not trusted.
The five questions behind Tensor.
Each is a way context fails, and the reason we are building Tensor. We publish the research so the answers can be checked.
Not found
How do you know something was never there?
Contradictory
When two records disagree, which wins?
Stale
How do you know the ground has moved?
Ambiguous
When a request could mean two things, which is it?
Restricted
What if the best record is off limits?
Published openly.
It started with checking the sources behind AI answers, moved to durable decision records, and now asks whether an agent has enough context.
- 1
Dataset
Do the sources hold up?
Open Evaluation Corpora began it, as the DALI Citation Benchmark.
- 2
Benchmark
Can it be reproduced?
Reproducibility Benchmarks tested whether conclusions can be reproduced.
- 3
Framework
How do we describe verification?
Verification Taxonomy gave it a shared language.
- 4
Specification
What should a decision record hold?
Evidence Package Specification made it a durable record.
- 5
Repository
Can anyone check it?
Open Repository puts it all in public.
- ●
Today
Is the context enough to execute?
Context sufficiency is where it leads: one answer from Tensor, with the reasons.