Closing statements, title documents and lender schedules are filed alongside project invoices, and carry figures far larger than anything they sit next to.
Property work generates a document set that mixes two very different scales of figure in one folder: project costs, which are many and individually modest, and transaction documents, which are few and very large.
The process
A project folder accumulates supplier invoices, subcontractor claims, permits and receipts as work proceeds. Alongside them, the transaction itself produces closing statements, adjustment schedules, title documents, lender payout figures and tax apportionments. Everything is filed together, because everything is relevant to the same property.
At some point somebody totals the folder.
Where it breaks
A closing statement carries the transaction value. Filed among invoices, it presents as another document with a supplier, a date and an amount. Its amount can exceed the entire remainder of the folder combined, so a single such document dominates a total on its own.
Adjustment schedules restate figures that exist elsewhere. Tax and utility apportionments at closing correspond to amounts also charged directly. Counting both duplicates them.
Lender documents carry balances, not costs. A payout figure or a draw schedule is a financing position. It is not project expenditure, and it is the largest number in most folders after the transaction itself.
Payment confirmations restate invoices already present. The invoice is filed and the confirmation of its payment is filed beside it, both carrying the same amount and both looking like supplier documents.
Filing is not classification. Every one of these documents is in the folder for a good reason — somebody judged it relevant, which is a much weaker claim than it being a project cost. The folder’s name records where documents were put, not what they are.
How we would build it
Classify before totalling, always. Every document is assigned a type before any figure it contains is used for anything. Types that are not costs are routed out of the totalling path entirely rather than being included and later adjusted for.
Classify from the content, not the file name. A name is often uninformative and sometimes actively misleading — a forwarded email attachment carries the mail system’s naming, not the document’s nature. The body text is what distinguishes a closing statement from an invoice, and it does so unambiguously.
Handle exempt and non-standard tax treatment as part of classification. Several document types in this set carry no recoverable tax, or a different treatment entirely. Deriving tax by applying a standard rate to them is wrong by construction, not by accident, and classification is where that is known.
Sort by amount descending and examine the largest documents first. This is the cheapest diagnostic available and it finds this class of problem faster than any other, because the documents that do most damage are precisely those that least resemble their neighbours. It works before any automation is built, and it is worth doing once by hand on any folder to see what is in there.
Retain the classification and its basis on every record, so that a figure excluded from a total is visibly excluded and explicable, rather than absent.
What the system proves about itself
The output distinguishes three states rather than two: documents that are costs and carry a figure; documents that are not costs, examined and correctly carrying no figure; and documents not yet resolved.
That third state is the one that matters operationally. A document with no amount is only unfinished if nobody has looked at it — once it has been established as a payment confirmation or a closing statement, having no amount is the correct final answer, and treating it as outstanding sends finished work back onto somebody’s list.
The reconciliation that proves the whole thing is simple: the total of cost documents, plus the value explicitly excluded by classification, must account for every document in the folder. Nothing is silently dropped, and the excluded value is stated rather than assumed.