Crucible cleanses, models and traces your data into a reconciled substrate, with sealed evidence you can reproduce on demand.
Most risk numbers fail not in the modelling but in the data beneath them: scattered across systems, reconciled by hand, impossible to defend when a supervisor asks how a figure was reached.
Crucible is the layer that fixes this. It takes raw, scattered source data and turns it into a single trusted substrate, recording how every figure was reached. It is the foundation Launch POC and RIOS are built on, and it supports the decisions your people make rather than making them.
Crucible is RiskMetrica's deterministic, offline-capable data engine. It cleanses, models and reconciles scattered source data into a single trusted substrate, and records evidence-grade lineage for every figure it produces.
Scope: The foundation that Launch POC and RIOS build on. It runs inside your environment, supports the decisions your people make, and never makes them.
Connectors read your source systems, the engine reconciles and models the data, and every step is recorded as lineage. The result is a number you can prove, traced back to where it came from.
The figure exactly as it lives in your own systems, read only, on your hardware.
Disagreement between systems resolved, identifiers matched, lineage attached.
The number with its interval, calculated from the trusted substrate.
Sealed and lineage bound as a CI-1 evidence record you can reproduce.
Read only, on your hardware, nothing leaves, independently verifiable.
Crucible examines each source through four lenses to judge whether it can be trusted. Each is scored from one to five, and the single highest lens, the binding constraint, governs the plan.
How the data physically comes out: on-premise against cloud, and how stable the schema is.
Residency, GDPR and special-category data, supervisory expectations and outsourcing.
Ownership, egress fees, redistribution limits and the lead time on change control.
Schema sprawl, identifiers, free text, duplicates and the canonicalisation burden.
The lens scores feed a data-quality roll-up that sets the width of the Risk Appetite Position interval: wider when the evidence is thin, tighter as it strengthens. We do not hide uncertainty, we expose it and carry it into the number.
Every Crucible run is captured as a hash-chained evidence record. The chain is append-only and tamper-evident, so the history of a figure cannot be quietly rewritten.
A replay re-executes from that record alone and returns a byte-identical result. A supervisor or your own team can re-prove any number, months later, from the evidence pack on its own. This is the difference between a report you are asked to trust and one you can independently verify.
Each run is sealed as a record holding the input and output hashes, the pinned code version and seed, the timestamp, the named signer, the recorded policy decision, the engagement metadata and the trace id. The chain is append-only and tamper-evident, and a replay reproduces a byte-identical result from the record alone. Talk to us for the methodology and a replay walkthrough.
Every figure traces back to its source, transformation by transformation. This is what makes the output audit-ready and defensible in a supervisory walkthrough.
Each number links back to the exact source records behind it.
Every transformation is recorded and can be explained on demand.
Your own team can reconstruct any output from the raw input.
Crucible runs inside your walls. It makes no runtime network calls, nothing leaves the environment, and the same inputs always produce the same outputs. Where it matters, it runs fully air-gapped.
Not a cloud tenancy and not a managed service. Crucible deploys within your own environment, on your terms.
Any ingress or egress is through a single controlled, signed staging step. There is no external connectivity at runtime.
The trusted substrate flows into Launch POC. There it becomes a defensible Risk Appetite Position, the likely position with its range and the probability of reaching a limit, and a control-effectiveness proof that shows which controls actually work.
Both arrive sealed as CI-1 evidence, reproducible by your own team. The board sets appetite; the platform exposes the position and the control picture, it does not decide them.
See how Launch POC proves it in 21 daysCrucible is the engine beneath the pathway. Launch POC proves it on your data in 21 days. RIOS rolls out the full operating model, module by module, at your pace.
Crucible is RiskMetrica's data engine. It cleanses, models and reconciles your scattered source data into a single trusted substrate, and records evidence-grade lineage for every figure. It is the foundation that Launch POC and RIOS are built on.
Every run is captured as a hash-chained CI-1 record. A replay re-executes from that record alone and returns a byte-identical result, so a reviewer can re-prove any figure from the evidence pack on its own, months later.
Yes. Crucible is deterministic and offline-capable. It makes no runtime network calls, nothing leaves your environment, and it can be deployed air-gapped where that matters.
Crucible produces the trusted substrate and the sealed lineage. Launch POC turns that substrate into a Risk Appetite Position and a control-effectiveness proof in 21 days. RIOS rolls the full operating model out on the same foundation, module by module.
Talk to us about putting your source systems through Crucible and proving the result in 21 days with Launch POC.