Experiment

Decision Memory

Every important decision remembers what was decided, why, using what evidence, under which rules and for how long it matters.

FoundationalOperationsNext
KNOWREMEMBERUSEEXCHANGELEARNANTICIPATE

01

Problem to address

Approval or rejection is often treated as the end state, while the reasoning, evidence and conditions that created that state are not retained in a durable, reusable form.

02

What the evidence tells us

Long-lived decisions need evidence that still makes sense years later. Approving bad data can lock in errors and create later surprises. Officers also need a record they can defend.

03

The proposition to explore

The idea

Treat the rationale, evidence, rule version, conditions and lifecycle of a consequential decision as part of the decision itself.

04

Hypothesis

If every consequential decision stores its rationale and evidence as part of the decision itself, later reviews, inspections and disputes will be faster, fairer and less dependent on private archives.

05

How it could work

Create a standard decision object linking outcome, reason, rule/policy version, evidence, actor, date, conditions and review/lifecycle events. Retention follows the liability and regulatory context.

06

Example: bring it to life

Ten years after a capital intervention, the current caseworker can see the original approval conditions, supporting photos, later site observations and any agreed changes, rather than asking the farmer to reconstruct the story.

Ten years later an officer can see:

Grant approved: 2027

Rule version: X

Evidence: photographs / invoices

Intended outcome: wildlife corridor

Conditions: Y

Inspection: 2028

Condition recorded: satisfactory

Subsequent deterioration: reported 2034

Decision: removal permitted

Reason: structural failure

Now the institution can remember the decision for as long as the person remains liable.

07

AI opportunity

AI can summarise the decision package, compare versions and surface missing evidence. It should never manufacture a rationale that was not recorded.

08

Potential value

  • Fairer retrospective casework
  • Clear audit trail
  • Reduced re-investigation
  • Stronger organisational accountability

09

Reduce uncertainty without overcommitting

Smallest meaningful experiment

Apply a prototype decision-memory template to a set of new real decisions for 4–6 weeks. Ask a different officer to review them later and assess whether they can understand the decision without speaking to the original maker.

10

Open questions and risks

What we still need to learn

  • Adds capture burden if badly designed
  • Different decisions need different retention periods
  • Reasoning quality matters as much as technical storage