The why engine,
embedded in every surface.
Compass is RegAlign's suggest-only assistant. It drafts findings, recommends mitigations, explains every rating with a why-trail — and never executes. Every action stays human-approved, every suggestion carries its source, every dismissal is recorded against the outcomes register so noise gets tuned out.
Mode badges on every surface
Mode is global, per-surface, and per-action. The badge is visible to every operator at all times.
Why this matters: Compass cannot quietly slide into autonomy. The mode badge is the contract — change it and the audit trail records who, when and why.
"Explain this" drawer
- · Inherent: High (TCB licence + non-resident clients ≥ 30%)
- · Mitigation: Strong (controls C-014, C-022 both effective)
- · Last test: 2026-Q2 · Pass · 30 sample / 0 exceptions
- · Residual: Medium (within appetite)
CRA framework v3.2 · RCSA cycle 2026-Q2Why this matters: No black-box scores. Every Compass output names its inputs and links to the controls, tests and policies that fed it.
Outcomes register — Compass tunes itself
Every suggestion's fate is recorded with a rationale. "Noisy" dismissals reweight the prompt; "useful but dismissed" suggestions stay live so future operators see them again.
Why this matters: The supervisor's question — "how do you know it's any good?" — has a numeric answer. Not a vibe, not a vendor white-paper, your outcomes against your data.
Auto-execute is disabled in pilot. Mode changes are audit-trailed with a justification field.
Every rating, draft and recommendation names the framework, the controls and the records that fed it.
Acted / dismissed-useful / dismissed-noisy / auto-tuned, with rationale. Reportable to the board.
Compass is firewalled from the MLRO substrate. Confidentiality before convenience.
See it on your own data
The sandbox runs Compass against the seeded Saltire tenant with a populated outcomes register and a live Explain-this drawer on every CRA domain.