Institution route · Technology and platform owners
Add evidence infrastructure without ripping out the stack.
Enterprise platforms already run the institution. The missing layer is a side-car that recomputes, governs and proves the result without forcing a migration.
Founder-direct intake · NDA before data · review-grade evidence
Three problems every platform owner recognises.
The issue is rarely the absence of platforms. It is whether their most important numbers can be recomputed, bounded and replayed when the institution is asked to defend them.
Reports become audit debt.
Pain. Your stack reports numbers it cannot reproduce on demand. Every unreplayable report is audit debt waiting for a review cycle.
SAA. Hash-pinned results with a replay path an external reviewer can run.
Verification should not become replacement.
Pain. Most risk vendors answer a verification question with a replacement project.
SAA. A side-car layer beside Aladdin, Bloomberg, Murex, MSCI or internal engines. The system of record stays yours.
AI certainty is not evidence.
Pain. AI vendors sell certainty. Model-risk teams need dated, scoped, falsifiable claims.
SAA. A typed claim matrix – can claim / cannot claim – attached to every output.
Why it matters now
Supervisory expectations on model risk now reach AI-assisted analytics, and agent-driven workflows consume risk numbers faster than quarterly validation cycles can police them. The cheapest control that survives both is an evidence layer: recompute, compare, hash, replay. Verification is a smaller project than migration – and it is the one project that makes every other number on the platform defensible.
Four technology questions before a risk layer can be trusted.
Platform owners need something deployable, inspectable and bounded – not a science project hidden behind screenshots.
Where does the data go?
Pain. Risk pilots die when data movement is unclear.
SAA. API hand-off, tenant-side or air-gapped patterns, with data-residency boundaries.
Can the answer be replayed without the vendor?
Pain. A platform that cannot replay its own output becomes audit debt.
SAA. Reproducibility manifest and SHA-256 evidence chain attached to outputs.
Will it replace the incumbent?
Pain. Rip-and-replace is the wrong starting posture for banks and large asset owners.
SAA. Independent challenger beside Aladdin, Bloomberg, Murex, MSCI or internal engines.
What is attested today?
Pain. Security and compliance claims must be dated, scoped and honest.
SAA. Status register, no SOC 2 Type II claim before report, no STAC claim before external review.
What the layer is measured to do.
Each capability is stated as a bounded lane, not as a universal promise.
The same input returns the same answer, provably.
1 hash · 1,000 fresh processes · 0.0% driftDeviation from the Monte Carlo reference is published per lane.
CVaR 99.9 max 0.035% · canonical laneEnvelopes are stated per workload, with lineage.
tick p99 0.44-0.46 ms · pre-trade p99 0.85-0.93 msA verdict is deterministic, challenged, or blocked – never silent.
hash-logged routing · no silent degradationClaims carry status and dates. Nothing is promoted past its evidence.
status register · no SOC 2 Type II / STAC claim before evidenceAll figures are public-tier lane numbers; precise distributions and run SHA are released under NDA.
Four deployment routes for platform teams.
Start with the least invasive route that can evidence the answer.
API hand-off
REST call from your pipeline; target: first signal ≈ 30 days after data access · estimate, pilot-scoped.
Side-car deployment
Runs beside your stack in tenant, air-gapped or sovereign-cloud pattern. No rip-and-replace.
Governance export
Evidence, hashes, release versions and boundary matrix for model-risk and audit functions.
Security posture review
Status register, security.txt, control domains and NDA trust pack for diligence.
Evidence, not status theatre.
Each institution route returns the same object: numbers, sources, owner, replay path and cannot-claim boundary.
DRCET and deterministic compute
An open method for checking whether a fast risk engine returns the Monte Carlo answer, with deterministic numbers and replayable evidence.
See methodology →Routes to the right surface
Risk Analyzer, Global Risk, ARIN and KOKON are used only where their claim boundary fits the institution workflow.
Open solutions →Independence Charter
No replacement claim, no hidden endorsement claim, no stronger claim than the evidence can replay.
Read charter →Founder-direct intake
Bring one integration constraint.
We map the least invasive deployment route, evidence trail and boundary before touching production data.