Provenance coverage report
Four typed assessments of one balance in one currency, never one score. Split the assessed balance into non-overlapping attributed amounts a_i, each with its evidence set. Provenance coverage Q = 100 × Σ a_i·v_i / Σ a_i, where v_i is 1 when amount i meets the published evidence policy P at time t, else 0; an empty balance has no score; £900 evidenced plus £100 unresolved is 90% coverage, never '90% lawful'. Evidence reliability is the calibrated error rate per evidence class measured against later outcomes, not weights chosen by hand. Financial-crime findings are a list with case identifiers and are never averaged into coverage. Investment risk is assessed separately on what the money funds. Attribution is pro rata and exact: a transfer of x from balance B carries x·a_i/B from each source, computed in exact rationals and rounded once at display by one published rule; splitting, recombining or circulating money never duplicates attributed amounts and copies of one document never become independent evidence; only new evidence changes an assessment. Credit origination, repayment, foreign exchange and flows beyond the observed network are explicit source kinds. The assessment is F(ledger snapshot, evidence, policy version, time): identical inputs give identical results and identical explanations; the verifier is total and terminating. The customer sees the four assessments with plain explanations and a correction route; nobody is asked to investigate their family; an incomplete record is unknown, never guilt. Rules that apply: MLRs 2017; FCA Consumer Duty; UK GDPR art 22 and DUAA 2025; BIS: payments at par; Bank of England: money creation by lending.