Benchmark

Summarisation on real financial data stays 96.2% accurate. Traditional masking drops to 41%.

RSP is benchmarked against placeholder masking, tokenisation, and raw-data baselines across 500 financial services test cases. Summary results are shown below. Full methodology and per-model breakdowns are available during a technical briefing.

Summarisation Accuracy

Raw data (baseline)100%

Presential RSP96.2%

Placeholder masking41.3%

Tokenisation12.8%

BLEU / ROUGE-L composite · 500 financial services test cases · methodology on request

System Performance

Transformation latency45-90ms

Entity classification50+ types

Reversal accuracy100%

Confidence gate≥ 95%

Fail-closed: entities below confidence threshold are redacted, never passed through

Full Benchmark Report Available

We share benchmarking methodology and full results with design partners and during technical evaluation. No gatekeeping. We know you need to verify independently before moving forward.

Request the Benchmark Report