Four states are enough

Most software that reads documents is built to produce a single answer. Ask it what a bank balance was in March and it returns a number. That behaviour is useful when there is one source and it is reliable. It is the wrong behaviour in financial remedy work, where the same fact is described by several parties who have different information and different interests, and where the disagreement between those descriptions is often the thing that matters.

Pivot is built the other way round. Instead of resolving conflicting evidence into one answer, it records what each source says about the same entity and assigns that entity a state. There are four states, and the design claim of this post is that four is sufficient for disclosure work: any further precision belongs to the fee earner, not the system.

What an assertion is, and why the word matters

In a financial remedy case — the part of a divorce that divides money and property — each party completes a Form E, the standard financial statement required by the Family Procedure Rules. A Form E entry is not evidence. It is an assertion: a statement by a party about what they own, owe and earn, which they are required to support with documents such as bank statements, pension statements and payslips.

That distinction is doctrinally ordinary and technically load-bearing. It means a disclosure record has two different kinds of input for every entity:

• what somebody said about it, and
• what a document shows about it.

If a system stores only one merged value per entity, it cannot express the relationship between those two inputs. Storing them separately, and typing the relationship, is what produces the four states.

The four states

Corroborated. A party asserted the entity, and a document in the record supports it. A declared current account with a balance of £4,210 at the Form E date, and a statement page for that account showing £4,210 on that date, is corroborated. The state records agreement between an assertion and a source, nothing more. It does not mean the entity is complete, correctly valued or relevant.

Uncorroborated. A party asserted the entity and no document in the record supports it yet. This is the ordinary state of a matter early on, before documents are read in, and it is also what remains when a supporting document was requested and has not arrived. Uncorroborated is not an allegation. It is a work item: either the document exists and should be obtained, or the assertion needs to be reconsidered.

Contradicted. A party asserted the entity and a document in the record says something materially different. A declared salary of £68,000 against payslips totalling £91,000 is contradicted. This is the state most often misread, so it is worth being exact: contradicted is a statement about two records disagreeing, not a finding about a person. There are innocent explanations for many contradictions — a bonus that was not treated as salary, a valuation date shifted by a month, an account closed and reopened. The system's job ends at showing which two items disagree, on what figure, and where each came from. Characterising the disagreement is a fee earner's judgement, and if it goes further than that it becomes a submission to a court, which software should not be drafting on its own authority.

Unasserted. A document in the record refers to an entity that nobody declared. A standing order to a savings account that does not appear anywhere in the Form E is unasserted. This state exists because omission is the failure mode that a per-assertion check cannot catch. If the system only walks the list of declared entities and looks for support, anything left off the list is invisible. Walking the documents and looking for entities nobody claimed is the complementary pass.

Why the states change meaning at merge

Pivot builds one disclosure record per party, then merges them once both sides have exchanged. Merge is where the states earn their design.

Before merge, "unasserted" means the client's own documents mention something the client did not declare. That is a completeness problem inside one record, and it is usually resolved by asking the client.

After merge, "unasserted" can also mean that the other side's documents refer to an entity neither party declared, or that one party declared something the other's documents contradict. The same word now covers a cross-record gap, which is a different kind of finding: it is the raw material of the questionnaire, the formal request for further information and documents that follows Form E exchange under Part 9 of the Family Procedure Rules.

Contradiction is not created by merge either. A single party's Form E can contradict their own bank statements, and Pivot flags that before the other side is involved — which is the point, because the party's solicitor would rather find it than have it found for them. What merge adds is contradiction across parties, where two people describe the same asset differently.

Why the system stops at four

An obvious next step is to grade the states: strongly corroborated, weakly contradicted, probably an innocent explanation. Pivot does not do this, for two connected reasons.

The first is that a graded score would have to be produced from evidence the system cannot see. Whether a £23,000 discrepancy in declared salary is a bonus treated inconsistently or a deliberate understatement depends on things that live outside the documents: what the client has said in conference, what the marriage looked like, what the other side has done so far. A score generated without that context would be presented with a confidence the underlying data does not support.

The second is a division of responsibility. A fee earner signs the documents that go to court, and a party signs the Form E under a statement of truth. The person carrying that liability needs to see the underlying disagreement, not a system's opinion about it. A state that says "these two figures differ, here is each one and the page it came from" is checkable in seconds. A score that says "moderate risk" is not checkable at all.

So the states are deliberately thin. They tell you which of four situations an entity is in, and every state carries the items that put it there. The decision about what any of it means stays with the person who is accountable for it.

What this buys a practitioner

Four states convert an unstructured reading task into a work queue. Uncorroborated entries are documents to chase. Contradicted entries are questions to ask, either of the client or through the questionnaire. Unasserted entries are gaps to close before a statement of truth is signed. Corroborated entries are the part of the record that can be relied on and does not need re-reading.

That is the argument for modelling disagreement rather than resolving it. The disagreement is the work. A system that averages it away has deleted the output.