Complete this before the work starts. A measure defined after the fact is not a measure; it is a description of what happened, chosen once the outcome was known.
The sheet
Playbook ____________________
Accepted action ____________________
Accountable owner ____________________
Date accepted ____________ · Re-measurement due ____________
Decision this measure will support ____________________________
1. The signal this addresses
What condition raised this work? Name the signal, not the concern.
2. The measure
One number, computed the same way twice.
| Field | Value |
|---|---|
| Measure | |
| Unit | |
| Source system | |
| Query or method | |
| Query or method version | |
| Evidence classification and retention | |
| Known source delay |
3. The population
Over whom or what is this computed? The same population must be readable at re-measurement, if it will have changed, say how.
4. The baseline
| Field | Value |
|---|---|
| Value at acceptance | |
| Date and time captured | |
| Captured by | |
| Source record or extract ID |
Capture this now. Not at completion. If the baseline is read after the work finishes, the person who did the work has already influenced the number, and the comparison becomes a reconstruction. Reconstructions drift toward whatever makes the result readable.
5. The declared direction
Which way must this move to count as improvement?
- Down · [ ] Up
Which dimension does this address?
- Likelihood, how often the condition occurs
- Impact, how bad it is when it does
- Not declared
If not declared, residual will not move on this verification. That is deliberate: inferring the dimension from a unit of measure would be an inference presented as a measurement.
6. The noise threshold
How much movement is required before it counts as movement? A measure with no noise threshold will report improvement from ordinary variation.
6A. Expected magnitude and decision threshold
The noise threshold says what counts as real movement. The decision threshold says how much movement is material enough to change a decision. They are not the same.
| Field | Value |
|---|---|
| Expected magnitude | |
| Minimum decision-relevant magnitude | |
| Basis for expectation | Prior data · pilot · expert judgement · unknown |
An outcome can be Verified because it moved beyond noise and still be too small to change the risk decision. Report both facts.
7. Re-measurement conditions
| Field | Value |
|---|---|
| Due date or interval | |
| Required observation window | |
| Minimum sample | |
| Population-change tolerance | |
| Source-lag allowance |
8. Rival explanations and guardrails
What else could move the measure, and what must not get worse while this one improves?
| Field | Entry |
|---|---|
| Known confounders or concurrent changes | |
| Population-selection risk | |
| Guardrail measure | |
| Displacement or workaround to watch | |
| Privacy, workforce or legal review needed | Yes · No · Not known |
This does not turn an operational before-and-after measure into a causal study. It prevents the most obvious rival explanation or harmful side effect from being ignored.
If the source, population or window cannot meet these conditions, the result is inconclusive, not unverified. If the due date has not arrived, it is pending, which is a state rather than a result.
Four measures that reliably fail
Anything counting activity. Sessions delivered, modules completed, messages sent. These move whenever the work happens, which means they always verify, which means they verify nothing.
Anything whose population changes with the intervention. If the intervention alters who is in scope, the before and after are computed over different denominators and the comparison is meaningless.
Anything with a reporting lag longer than the re-measurement window. The measure will read as unchanged because the data has not arrived, and unchanged will be recorded as unverified when it should be pending.
Anything only one person can compute. If the method lives in someone's head, the second measurement is not the same measurement.
Before you file this
Read section 2 and section 4 together and ask one question: could someone else, given only this sheet, take the same measurement in ninety days and get a number comparable to mine?
If the answer is no, the sheet is not finished, and the work should not be accepted yet.
Industry basis and IO addition
NIST and ISO methods expect monitoring, measurement, assessment and continual improvement. IO's added discipline is temporal and evidentiary: define the measure before execution, capture the baseline at the accountability event, preserve the query version and denominator, and keep an unsuccessful result. The sheet supports those frameworks; completing it is not a control assessment or a conformity determination.
References: NIST CSF 2.0 (opens in a new tab) · NIST AI RMF (opens in a new tab) · ISO/IEC 27001 (opens in a new tab)