How every number is computed

coverage_green / coverage_total  (rounded to nearest integer percent)

• coverage_green = obligations with all linked controls effective and no open high/critical findings
• coverage_amber = obligations with at least one watching signal (overdue test, medium finding, etc.) but no high/critical breach
• coverage_red = obligations with at least one open high/critical finding OR no effective control mapped
• coverage_total = green + amber + red (i.e. in-scope obligations only)

• Out-of-scope obligations (in_scope = false)
• Obligations in draft, under_review, superseded or withdrawn lifecycle states

Recomputed on every snapshot read; no caching.

verify_audit_chain(tenant_id) — recomputes each entry's SHA-256 hash of (previous_hash || canonical_payload) and confirms the linked chain matches the stored hashes.

• audit_trail_entries rows for this tenant in chronological order
• Per-entry canonical payload (deterministic JSON serialisation)

• Entries from other tenants
• Soft-deleted rows (audit trail is append-only — none exist)

Verified live on every governance snapshot fetch.

A 'break' means the recomputed hash for at least one row no longer matches the stored hash — i.e. someone or something has bypassed the trigger that writes new entries. Investigate immediately.

count(issues) WHERE status IN ('open','in_progress') AND target_close_at IS NOT NULL AND target_close_at < now()

• issues.status
• issues.target_close_at

• Closed or waived issues
• Issues without a target close date

Live on every read.

validations.confirmed_this_period / max(issues.resolved_this_period, 1)

• validations rows with outcome = 'confirmed' and validated_at within the last 90 days
• issues rows with status = 'closed' and closed_at within the last 90 days

• Validations for issues closed outside the 90-day window
• Self-validations (validator_id = closer_id are not counted)

Recomputed per snapshot. Rolling 90-day window.

A rate below 100% means some closures in the period have not yet been independently validated. Not necessarily wrong, but the gap should be closing, not widening.

round(100 * passed_tests / executed_tests) where executed_tests = passed + failed (waived excluded)

• monitoring_tests with status IN ('passed','failed') within the active plan window

• Tests with status = 'waived' (waivers require their own justification and are reported separately)
• Tests scheduled but not yet executed

Recomputed per snapshot. Plan window = plan.period_start → plan.period_end.

round(100 * (accepted_items - covered_items) / max(accepted_items, 1))

• monitoring_plan_items with status = 'accepted'
• covered_items = accepted items that have at least one monitoring_test scheduled or executed

• Plan items in 'proposed' or 'deferred' status

Live on every plan read.

ratifications_with_attendance / ratifications_in_window  (ratio, not %)

• ratifications in the last 90 days
• committee_attendance rows linked to those ratifications

• Ratifications older than 90 days

Recomputed per snapshot.

An approval without recorded attendees is not defensibly 'a board decision' — fix the data, not just the number.

Σ ( obligation.weight × max_open_finding_severity_weight × (1 + age_days/180) ) for in-scope obligations with at least one open finding

• obligations.weight (default 1, max 5)
• findings.severity → weight mapping: low=1, medium=2, high=4, critical=8
• max age in days of any open finding against the obligation

• Out-of-scope obligations
• Closed or waived findings
• Findings with severity = 'info'

Snapshotted daily into regulatory_exposure_snapshots; live recompute available.

The aging multiplier caps at +1.0 (i.e. 2x at 180 days and beyond) so a single very old finding cannot dominate the score indefinitely.

Deterministic mulberry32 RNG seeded with sha256(snapshot_content_hash || method || sample_size || strata_weights)

• population_snapshot.content_hash (SHA-256 of canonicalised row bytes)
• sample_selection.method ∈ { random, stratified, risk_based }
• sample_selection.sample_size
• sample_selection.strata_weights (for stratified / risk-based)

• Rows excluded by the snapshot's filter expression (recorded on the snapshot row)

One-shot. Re-running with the same inputs deterministically reproduces the exact selected IDs — verifiable from the audit trail.

Stratified / risk-based allocation uses largest-remainder to apportion the total sample size across strata.

Sum of time saved (hours), automated decisions, and avoided rework, computed from concrete counters — never modelled.

• Findings auto-classified by Compass (with human accept rate)
• Decisions ratified with full audit trail
• Issues closed inside SLA
• Tests executed via the platform vs uploaded externally

• Modelled or projected benefits
• Anything that requires a human to estimate a percentage

Live counters; rolling 90-day comparison.

Numbers come from concrete counters, not from converted estimates. The methodology section deliberately does not include a £-conversion — that is the firm's call, not the platform's.