A public-safe benchmark summary that keeps methods and caveats visible
The benchmark summary now behaves as the validation layer of the resource system, emphasizing which numbers are public-safe, how they were measured, and what limitations apply to each metric.
controlled enterprise workload
benchmark factual QA
hot-zone cache hits
pass / fail / error
The primary metrics without hiding the method behind them
Use this summary to view the key metrics together with routes into the layers that explain the method or supporting architecture.
0.3% hallucination rate
Measured on controlled enterprise workloads and cross-checked against SignedAI disagreement logs plus a manual validation sample.
0.92 FDIA accuracy
A public-safe summary of how the FDIA equation tracks human-evaluated ground truth in the factual QA benchmark.
Warm recall under 50ms
Summarizes hot-zone semantic cache latency while clearly separating it from the 3-5 second cold-start path.
Caveats are disclosed explicitly
Every number on this page is paired with its method and caveat so the benchmark summary does not collapse into a marketing claim.
The test environment that can be disclosed publicly
This provides the minimum context needed to interpret the numbers without pretending to be a full dossier for every workload.
The next pages to use when interpreting the benchmark
Benchmark reading is incomplete without methodology and evaluation, otherwise the numbers lose their decision context.
Continue into methodology
Use this when the team needs the fuller model for measurement, disclosure, and the public claim boundary.
Go to evaluation
Use this when the benchmark context needs to be translated into a buyer or architecture decision path.
The benchmark summary should always be read with methodology and evaluation
The numbers on this page are for framing and validation, not as a substitute for the full decision process. Pair them with methodology, whitepapers, and the evaluation hub before making business or procurement conclusions.