Back to Benchmark Library
BENCHMARK CASE
Silicon Valley Bank 2021 Market Risk
Could the Board identify rate and model-risk vulnerabilities?
- Document
Silicon Valley Bank 2021 Form 10-K — Item 7A Market Risk
- Objective
Could the Board identify rate and model-risk vulnerabilities?
- Methodology
Identical prompt, identical document, identical context window across NDOR, ChatGPT Plus, and Claude Pro. Scored on executive decision-relevance: did the output produce a usable basis for committee-level action under the question asked?
FINDINGS PER SYSTEM
What each system surfaced
NDOR
- 27.7% EVE decline under +200bp shock — material capital-base impairment.
- $5.7B absolute EVE loss disclosed without compensating mitigation strategy.
- HTM concentration amplifies rate-shock convexity beyond static disclosures.
- Deposit behavioural-model assumptions identified as the dominant model-risk vector.
- Hedge programme disclosures lack transparency on basis and effectiveness.
ChatGPT Plus
- Rising-rate sensitivity surfaced.
- Static modelling characterised as a weakness.
Claude Pro
- Missing severe-rate scenarios identified.
- Capital sensitivity concerns flagged.
VERDICT
NDOR translated the disclosed Item 7A metrics into concrete governance actions — naming the behavioural assumption to stress, the hedge disclosure to seek, and the capital threshold at which escalation becomes mandatory. The comparators identified that the disclosures were thin; only NDOR specified what to do about it.
Run the same kind of analysis on a document of your choice.
NDOR applies the same structured validation workflow to contracts, models, proposals, and reports.