Skip to content

Rules

Cross-period analysis and context

A compliance rule asks whether a posting breaks a stated requirement. A forensic rule asks whether it is unusual against this client's own history. Both are deterministic, both explain themselves, and neither produces a score.

Context is not a finding

A transaction can be unusual without being wrong, and most unusual transactions are neither. VLDTION therefore keeps the two apart: the materiality engine measures, and rules decide. Measurements are attached to findings that already exist to make them readable; they do not generate findings of their own.

No risk score

There is no composite number and no weighting. “₪286,400, 12.4% of the period’s expense postings, 3.8× the median manual journal, no comparable posting last quarter” is four facts a reader can check. “Risk 87” is one they cannot.

What the materiality engine measures

MeasureHow it is computed
AmountThe net amount of the posting, before VAT.
Share of the accountThe posting ÷ everything posted to that account in the period.
Share of the periodThe posting ÷ everything posted on the same side in the period.
Against the medianThe posting ÷ the median of every other posting of the same document type.
Largest comparable last periodThe largest posting of the same type in the previous period, or a statement that there were none.

Every measure carries its own denominator and the population it was taken over, so a reader who disagrees can recompute it. A measure that cannot be computed honestly. A percentage of nothing, a ratio to a zero median. Is omitted rather than reported as zero.

Comparing two periods

The comparison screen takes two stored runs of the same client and reports what changed, keyed by the content-addressed identity of each finding:

  • Appeared, raised in the later period only.
  • Resolved, raised in the earlier period only.
  • Recurring, the same exception in both.
  • Changed, recurring, but the amount moved by more than 1%.
  • Counterparties. Appeared, stopped trading, or continuing, matched on business number where the export carries one and otherwise on the exact normalised name.
  • Measured differences, postings, documents, journals, accounts, counterparties, totals, median and mean posting, VAT.

The forensic rules

RuleWhat it looks forNeeds the previous period
RULE-F-001A manual journal at period end that is large against the client's own journalsNo
RULE-F-002A counterparty that did not trade last period and is immediately materialYes
RULE-F-003An account dormant last period carrying material activity nowYes
RULE-F-004The same period-end adjustment, to the same account, within 1% of the same amountYes

Each is specified in full. What, why, when, what it requires, how it detects, what evidence it produces and what it does not prove, in the repository under docs/rules/forensic/. All four are development rules: they are VLDTION’s own checks, not statements of Israeli law, and they are labelled as such wherever they appear.

Without an earlier period they say nothing

A cross-period rule with no period to compare against reports insufficient data, not a pass. A workspace holding one quarter will see three of these four rules report exactly that, and the rules screen says so.

Identity stays conservative

Two counterparties are the same when the business number matches, or when the normalised name matches exactly. Nothing is fuzzy-matched. A counterparty that changed its name will therefore appear as new, which is a fact worth seeing rather than a defect to paper over.