Knowles Knowledge Centre•Diagnosis

How to Identify the Gap Between Your Stated Rule and Your Last Twenty Decisions

A diagnostic method for working out whether an uneven pattern is a discretion problem, a rule problem, or neither — before you commission any training.

A gap is the distance between the rule somebody believes they apply and what their own record shows they did. It is narrower than it sounds, and it is not a verdict on anybody: a discrepancy is the normal condition rather than a discovery, and the useful question is its size and direction rather than whether one exists.

This guide sets out a three-step diagnostic method, the evidence sources that support it, sample questions for separating a discretion problem from a rule problem, and thresholds for deciding when development is the right response.

Practical guide•For HR and L&D teams•Updated October 2026

About this guide

Published byKnowles Training Institute
AudienceHR, L&D and line managers scoping a bias-related need
Use beforeSelecting or commissioning any unconscious bias programme

Why uneven patterns are misdiagnosed

The common failure is starting from the solution. Somebody observes that the same few names get the interesting work, and the request enters the L&D pipeline as awareness training to be sourced rather than twenty rows to be counted.

Three things go wrong when a request is accepted at face value.

The observation names a pattern, not a decision“It is always the same people” describes an outcome. Until somebody names the decision that produced it and the rule that was supposed to govern it, there is nothing to diagnose and nothing to teach.
The pattern is treated as evidence of attitudeAn uneven distribution is an outcome. It can be produced by discretion exercised inconsistently, but equally by a rule that encodes the unevenness and is being followed faithfully, or by a pool that nobody widened.
No alternative explanation is testedIf training is the only option considered, it will be the option chosen — regardless of whether it addresses the actual constraint.

A manager departing from their own rule and a manager following a rule that was never fair produce the same uneven list. Only one of them is a development question.

The rule–record–cause method

The method works backwards from a pattern to a specific rule, then tests whether discretion was the binding constraint.

1

Get the rule in writing

Ask whoever makes the decision to write down, in one scoreable sentence, the rule they believe they apply. Do this before showing them any records. “Whoever is next in the rotation” is usable. “I try to be fair” is not, and should be sent back.

If no rule can be written, that is the finding. A decision made repeatedly with no articulable rule is pure discretion, and the response is to write one rather than to raise anybody’s awareness.

2

Count the last twenty

Retrieve the last twenty decisions of that kind, in sequence, from a system rather than from anybody’s recollection. Have somebody other than the decision-maker mark each as following the rule, departing from it with a recorded reason, or departing with nothing recorded.

For a rotation rule, the three counts might come out at eleven, six and three. The third number is the one that matters, and until the rows were marked there was no number at all.

This is the step most organisations skip. Without it, everything downstream is guesswork.

3

Test whether discretion was the constraint

Ask the question that separates a discretion problem from a rule problem: if the rule had been followed exactly, every time, would the pattern have gone?

If no — because applying the stated rule faithfully still produces the same distribution — the constraint is the rule, not the discretion. Awareness training is then aimed at the wrong thing entirely, and the work is to examine what the rule encodes.

If yes, establish whether the departures were unrecorded, recorded but unjustified, or justified individually and uneven in aggregate. Those three states require different responses.

Where the evidence comes from

A diagnosis is only as good as what it is built on. Most organisations already hold more usable evidence than they realise, and reach first for the two sources least able to show a gap.

Evidence source

What it shows well

What it cannot show

A written rule, dated before any records were opened

The standard the record can actually be scored against

Whether the rule itself is defensible

The last twenty decisions from a system, in sequence

What was decided, without reconstruction

Why each one was decided that way

The same rows scored by somebody other than the decider

A count that justifications have not been attached to

The circumstances that made a departure reasonable

The pool each decision was drawn from

Whether the unevenness precedes the decision entirely

Anything about how the decision was made

A conversation with the decision-maker about their reasoning

What rule they believe they apply

What they actually did — the account is reconstructed

An implicit association test or an attitude survey

That associations and attitudes exist in a population

Any individual’s decisions — it measures neither

Note the pattern: the two sources most organisations reach for first — a conversation about reasoning and a test of associations — are the two least able to show a gap, because one returns a constructed account and the other was never designed to describe an individual’s behaviour.

Use at least two independent sources before concluding a discretion gap exists. A dated written rule plus twenty rows scored by a third party is evidence. A test result or an impression of somebody’s attitudes is neither evidence nor, in the first case, a use its developers endorse.

Diagnostic questions that separate cause from symptom

Ask these in sequence. The first “no” usually identifies the constraint.

Discretion availableDoes the decision-maker genuinely have room to decide at the moment of deciding? If not, there is no gap to find and no training changes anything.
Rule articulableCan a scoreable rule be written at all? If not, the finding is that an undisciplined discretion exists, which is a design question rather than an awareness one.
Record availableDo the last twenty decisions exist somewhere retrievable? Without them the audit becomes a conversation, and a conversation returns the rule rather than the practice.
Rule sufficiencyWould faithful application of the rule remove the pattern? If not, the rule is the exposure and the discretion is irrelevant.
PoolWas the unevenness already present in what the decision chose between? A perfectly applied rule cannot widen a pool nobody widened.
CapabilityWith real discretion, a writable rule, a retrievable record, a rule that would remove the pattern and a pool that was not already uneven — does the record still show unrecorded departures?

Only the last question identifies a development need. The first five identify design, recording, rule and pipeline problems that a course will not resolve — and the fourth of them can make a course actively misdirected.

Decision thresholds

A diagnosis should end in a decision, not a report. These thresholds keep that decision defensible.

Finding

Threshold

Response

Faithful application would not remove the pattern

Rule followed exactly in every row and the distribution unchanged

Not a development need. The rule is the exposure, and examining what it encodes belongs to whoever owns it.

No rule can be written at all

Decision made repeatedly with nothing articulable governing it

Write one. The cheapest fix the method produces, and much the most common starting point.

Unrecorded departures, running one way

Several of twenty departing with nothing recorded, mostly in one direction

Development is appropriate. Scope it to the scoring and repair stages, not to bias awareness in general.

Every manager shows the same gap on the same rule

Identical discrepancy across unrelated decision-makers

Almost always a rule that is unworkable in practice. Redesign it once, centrally, rather than training everybody who departs from it.

The pool was already uneven

Decisions drawn from a set that was narrow before anybody decided anything

Nothing to fix in the deciding. The exposure is upstream, and the audit will keep returning a clean sheet.

The threshold that matters most

If faithful application of the rule would not remove the pattern, the constraint is the rule rather than the person applying it. That single check prevents more wasted training spend, and more misdirected blame, than any other step in this method.

A worked example

A Singapore technology firm notices that the same four engineers have taken every client-facing assignment for two years. All eleven team leads are booked onto an unconscious bias course.

1

Get the rule in writing

Asked before seeing any records, the lead responsible writes: “Client-facing work goes to whoever has completed a client assignment least recently, provided they have passed the internal review.”

2

Count the last twenty

A colleague scores the last twenty allocations from the project system. Eighteen followed the rule exactly. Two departed, both with a recorded client request.

3

Test the constraint

Asked whether faithful application would have removed the pattern, the answer is no: only five engineers have passed the internal review, and the rule makes that a precondition. The rotation was working perfectly within a pool of five.

Diagnosis: this is not a discretion gap. The lead’s record is almost immaculate against her own rule. The pattern comes from the internal review, which has admitted five people in three years, and which nobody had looked at because it was not thought of as an allocation decision.

Response: a bias awareness course for eleven leads would have addressed a gap that does not exist, and left the review untouched. What is needed is an audit of the review itself — its criteria, its last twenty outcomes, and who decides them — which is the same method applied one step upstream.

Frequently asked questions

?
What counts as a gap?The number of the last twenty decisions that departed from the stated rule with nothing recorded, together with the direction those departures ran in. It is distinct from a bad outcome: a rule can be followed exactly and still produce an uneven distribution, which the method is specifically designed to distinguish.
?
How do you find one objectively?By writing the rule down first, retrieving the last twenty in sequence from a system, and having somebody other than the decision-maker score them. The test is reproducible in a way that a conversation is not, because an account of one’s own reasoning is constructed to fit the rule one has just stated.
?
Should we audit everybody who makes these decisions?Start with one decision rather than one population. Auditing eleven managers on a rule nobody has examined produces eleven numbers and no finding; auditing one rule properly usually locates the exposure, which is frequently not where the pattern appears.
?
What if somebody refuses to write a rule?Then record that, because it is a finding rather than an obstruction. A decision made repeatedly with no articulable rule is unconstrained discretion, and the useful response is to design a rule with them rather than to send them on a course about their mind.
?
How often should this be repeated?Tie it to events rather than to the calendar. Re-run it when a pattern is noticed, when a rule changes, or when a decision acquires a new record. Annual bias audits tend to produce ritual rather than insight, and twenty fresh rows are worth more than a yearly cycle of the same ones.

Conclusion

Identifying a discretion gap is mostly an exercise in disproving easier explanations. The method is not complicated:

1Get the rule written down, before any records

2Have somebody else score the last twenty against it

3Ask whether faithful application would remove the pattern

The discipline it requires is the willingness to conclude that the problem is a rule everybody is following properly.

An organisation that reaches that conclusion sometimes is diagnosing properly. One that always reaches for an awareness session is not diagnosing at all.

Where the diagnosis does point to a discretion gap, the next question is which stage of the method it sits in — covered in you cannot inspect your own bias, but you can audit your record. Organisations that decide facilitated development is appropriate can review the Understanding Unconscious Bias Training Course in Singapore, evaluating scope and expected workplace application against the diagnosis rather than against the course description.