Dated properties

Every property on every record is a series of dated values rather than a single value, whether it's a rate, a limit, a fee, a tax residency, a bank account or a classification. What was it on 3 March? What if it steps up next month? And what if you find out on 10 March that it should have changed on 1 March?

Where a greenfield design oversimplifies

When you design a ledger from scratch, it's tempting to keep it simple and give each value a single field: one rate, one limit, one address, one classification. It's usually the first thing to catch you out.

A single field holds one value with no date. The moment it changes the history is gone, and the moment a change is back-dated everything calculated from it since then is wrong, and nobody knows by how much. So teams bolt on a history table, a "current" flag and a batch job, and then spend years keeping the three in step. We know, because we built it that way ourselves and spent years keeping the three in step by hand. That's a big part of why LedgerTM is event-sourced, and why every value is held as a series of dated values rather than a single field.

A rate as a date seriesA timeline from January to April. The rate is 8% from 1 January and 9% from 1 March; the 9% entry was recorded on 10 March. A read as of 5 March answers 9%; a read of what was known on 8 March answers 8%.Rate, as a date series1 Jan1 Feb1 Mar1 Apr1 May8% from 1 January9% from 1 Marchentered on 10 March, when the omission was foundRead as of 5 March: 9%what was true thenKnown on 8 March: 8%what the event stream held thenTwo time axes: when it was true, and when we knew. Event sourcing gives the second; dated properties give the first.
One property, a rate, with two entries. Two different questions get two different right answers.
Not just rates

The rate is only the example, and the rule covers every property of every kind.

Numbers, text, flags, classifications and references to other records are all dated, on accounts, counterparties, operating entities, structures and relationships alike. There's no special list of "financial" properties that get a history while the rest don't, because any calculation might depend on any property.

On an account

Rate, margin, day count, limit, instalment, fee, interest method, collateral cover, the account that collects the instalment.

On a counterparty

Tax residency, KYC status, hold, address, roles, the bank account payments go to.

On an operating entity or structure

Accounting currency, posting group, cost centre, the legal entity a structure reports into.

On a relationship

Who guarantees whom, from when, until when.

The model

Every write carries a value date

On every record and for every kind of value, from rates and limits to references and flags. If no date is given the trading date is used, and it's always captured at the time of writing, because a date you didn't capture can never be recovered.

Every read is as of a date

There's no way to ask for "the current value". Current means "as of today's trading date", so an undated read can never slip into a calculation by accident.

Forward-dated changes wait

An entry dated next month is invisible to everything until the day walk reaches it. There's no scheduler and no batch job, because the series answers the question and the date does the rest.

Back-dated changes revalue

Enter a change with a value date in the past and the account replays from that date, works out the difference and posts it as new value-dated events, without ever editing the original entries.

Worked example

On 10 March you find that a rate change agreed for 1 March, 8% to 9%, was never entered. You enter it now, with a value date of 1 March. The same mechanics serve a stepped rate plan agreed up front, a pricing discount or markup granted from a date, or a mistake found later.

What you seeWhat happens
The change appears immediately in the change listRecorded with value date 1 March, on the 10 March trading date
The account revaluesNine days of accrual, 1 to 9 March, are recomputed at 9% and the difference is posted
A statement asked for 5 March shows 9%The as-of read answers on the value date
An entry keyed to 10 March is untouchedOnly days on or after the value date move
The same change dated 1 April insteadNothing happens today. It sits on Staged Changes and takes effect when the day walk reaches 1 April
With event sourcing, two clocks

Event sourcing records when we knew, and dated properties record when it was true.

Together they give a snapshot of any account, counterparty, entity or structure at any point on either clock: what the book looked like on 5 March as we understood it on 8 March, and as we understand it now. That is what makes value dating and transaction adjustments safe, and what an auditor asks for months later.

A late correction, a stepped rate or a dated discount is just a dated entry rather than a data fix, and the statement, the accrual and the audit trail all agree, because they all read the same series.