A Name and a Timestamp Are Not an Approval
An approval is usually recorded as a name and a timestamp. Two of the three things that matter are missing.
Neither absence was an oversight. Both were reasonable omissions right up until the year they stopped being reasonable, and nothing in any approval system marked the moment.
The first missing field: which version
A document approved on Tuesday and edited on Wednesday carries Tuesday's approval into Thursday. Most document systems let that happen quietly.
The mechanism is the same everywhere. The approval is attached to the document rather than to the state the document was in. The document is a container whose contents change; the approval points at the container. So the record says this was approved, and the thing it points at is whatever that thing is today.
Nobody designed it that way on purpose. It is what you get when the object model has a document with a current revision and an approvals table beside it, which is how almost every document system is built, because for most purposes that is the right shape.
It only misleads under one condition: when somebody asks what was approved rather than whether it was. Which is to say, during an investigation.
The second: what the approver was shown
Not what was available to them. What was actually in front of them when they agreed.
For most of corporate history this was safe to skip, because an approver had to read the thing to form a view about it. Reading it was the cheapest route to an opinion. The absence of the field was not a gap, it was a redundancy: the approval itself implied the reading.
A generated summary is cheaper than reading, and the summary is now what a great many approvers read. The assumption has quietly gone false, and it was load-bearing under every approval record in the business.
What makes this hard to see is that nothing changed in the approval system. The fields are the same, the workflow is the same, the reports are the same. One team adopted a drafting tool, and the meaning of a record elsewhere in the organisation changed on that day, with no edit to anything anybody would think to audit.
Banking worked this out and had it legislated
There is a precedent, and it is worth knowing because it is already law rather than a proposal.
Under PSD2, an authentication code must be dynamically linked to the specific amount and the specific payee. An approval cannot be lifted off one transaction and applied to another, because the thing that authorises is bound to the thing being authorised.
The reason the rule exists is instructive. Regulators did not introduce it because bank customers were careless. They introduced it because an authorisation that floats free of its object is an authorisation that can be reused, and reuse is indistinguishable from the real thing after the fact.
The reasoning holds anywhere. Bind the authorisation to the record, not to the verb. One confirmation, one act, one record of what was in front of the person who gave it.
What it looks like when it goes wrong
A deviation is approved in March. In August a customer complaint turns into an investigation, and the question is whether the deviation should have been approved at all.
The record holds. A quality lead, a Tuesday, a document reference. The document reference resolves, and the document it resolves to contains a mitigation step that was added in April, after the approval, by someone tidying up an unrelated finding. Nothing was concealed and the April edit has its own clean trail.
But the organisation can no longer answer the only question that matters, which is whether the approver saw a document that justified the approval or a document that did not. Both stories are consistent with every record held. The investigator has to assume the worse one, because the burden sits with the organisation and it has nothing to discharge it with.
That is the whole failure mode. Not a bad approval. An approval that cannot be defended, given by someone who in all likelihood did nothing wrong.
Why this cannot be added later
This is cheap to build at the start and close to impossible to retrofit, and the asymmetry is worth being blunt about because it decides when the work has to happen.
The approvals already held cannot be enriched after the fact. Nobody can go back and establish what somebody saw in 2024. The version they looked at may still exist, or it may have been superseded four times; the summary they read was generated on request and was never stored anywhere; and the person, if they are still with the organisation, will tell you honestly that they do not remember.
So the set of approvals that can ever carry their basis is the set created after the layer exists. Every day without it adds permanently to the set that cannot.
That is the layer Task Accountability Infrastructure holds, from the first approval onward, which is the only point at which it can be held.
Questions
What is wrong with recording an approval as a name and a timestamp?
It omits which version was approved and what the approver was shown. The first lets a document edited after approval carry that approval forward. The second used to be implied, because reading the document was the only way to form a view, and generated summaries have ended that.
What is dynamic linking under PSD2?
A requirement that an authentication code be bound to the specific amount and the specific payee of a transaction, so an authorisation cannot be lifted off one payment and applied to another. It is the same principle an approval record needs: bind the authorisation to the record, not to the verb.
Can approval basis be added to historical records?
No. Nobody can establish after the fact what a person saw in 2024. The version may have been superseded, the generated summary was never stored, and memory is not evidence. Only approvals created after the layer exists can ever carry their basis.
Why does an approval point at a document rather than a version?
Because almost every document system models a document with a current revision and an approvals table beside it, which is the right shape for most purposes. It misleads only when somebody asks what was approved rather than whether it was, which is to say during an investigation.