Most firms tell you their work is rigorous. We would rather show you the controls. The seven below run on every engagement, calibrated to scope — and each one is something you can ask to see in action before you sign.
1. Assumptions register
Every number can be traced back to the assumption behind it. Every model output we deliver carries a written register of its assumptions — economic, demographic, behavioural, methodological. Each assumption has a named owner, a date, a source, and an effective basis. Changes are versioned. Reviews are scheduled. The register answers the question, "where did this number actually come from?"
The columns we maintain in every register: owner, date, source, basis, version, last-reviewed, next-due, status. A sample redacted template is available on request.
2. Model validation protocol
Validation runs alongside the build, not as a box ticked at the end. It is methodology review, code review, reperformance, and challenger calculation, applied as the work progresses. We document each layer. The validation report is a deliverable in its own right — written to a standard that holds up under Prudential Authority, PRA and SR 11-7-style review.
Our validation findings register tracks: finding ID, severity (1–4), category (methodology / data / code / governance), owner, status, target close date, resolution evidence. A sample redacted register is available on request.
3. Peer review
Nothing leaves the team on one person’s say-so. No senior signs off their own work. Every model, every output, every report goes through peer review by a second senior. Findings are written, not verbal. Disagreements are recorded with the resolution. Two pairs of eyes, always.
4. Version control and reproducibility
Six months on, we can reproduce any number we delivered — and so can you. Calculation engines, whether commercial, in-house code, or a Symphony accelerator component, are version controlled. Every run is reproducible from a tagged commit, a set of inputs and a configuration. At handover you receive the source, not just the output — so the work can be re-run and changed without us. That is the difference between auditable work and aspirational work, and between a partner and a lock-in.
5. Scenario testing
We find where a model breaks before you do. Models are tested against a defined set of scenarios before they leave us. Base case is one. Stress is another. Edge cases — the ones the model is least likely to handle well — are deliberately constructed. We document where the model breaks down and what to do when it does.
6. Issue classification and tracking
Every problem is logged in the open, never buried. When something goes wrong — a number that does not reconcile, a method that breaks, an assumption that fails review — it is logged, classified, owned and tracked to resolution. The register is visible to you. There are no hidden issues.
7. Structured handover
The work is done when you can run it, not when we send the final email. Handover is a documented deliverable — named owners on your side, written procedures, signed-off test results, and a defined contact path for questions afterwards. We scope and resource it, rather than treating it as an afterthought.
These controls are how we hold ourselves to the seven principles. To see them in action on a specific scope, the place to start is a Diagnostic Assessment.