World Model Readiness
Engraved foundations instrument

For your company

Foundations

Module · Does anyone check if it worked?

The Outcome Loop Check

Most companies produce recommendations and never learn whether they worked. A closed loop runs from recommendation to action to a measured outcome and back again, and it is the single mechanism that lets AI improve instead of just repeating itself. This module checks whether even one such loop exists, and where your chains break.

Question 1 of 5 · A closed loop exists

Can you name one recommendation whose real-world outcome you actually measured and fed back?

Not a dashboard, not a forecast: a full loop. Someone recommended, someone acted, the result was measured, and that result changed the next recommendation. If you cannot name one, you do not yet have a loop.

Question 2 of 5 · Outcomes, not outputs

Do you measure the outcomes of your decisions, or only that they happened?

Shipping a recommendation is an output. Whether it moved the number it promised to move is the outcome. Most reporting stops at the output because the outcome is harder and sometimes embarrassing.

Question 3 of 5 · Recommendations reach action

Do your recommendations turn into action, or die in a deck?

A recommendation nobody acts on cannot have an outcome to measure. The most common break in the chain is the earliest one: the insight is produced, admired, and never actioned.

Question 4 of 5 · You know where it breaks

Do you know exactly where your recommendation-to-outcome chains break?

Every open loop breaks at a specific link: the recommendation is ignored, the action is unrecorded, or the outcome is never measured. Knowing which link is the difference between fixing it and blaming the model.

Question 5 of 5 · Someone closes the loop

Is there a named person accountable for closing the loop, not just running the model?

Loops stay open because closing them is nobody's job. The model has a team, the dashboard has a team, but the full chain from recommendation to measured outcome usually has no owner at all.

For the statistics · one click each

Three questions for the public picture

These do not affect your score. They feed the anonymised, aggregated statistics; groups under 8 respondents are never shown.

Can you point to one recommendation whose real-world outcome you measured?

Yes, several
Yes, one
Not sure
No

What usually happens to a recommendation in your company?

Actioned and reviewed
Actioned, never reviewed
Sometimes actioned
Dies in a deck
We do not track

How do you usually learn whether a past decision worked?

Measured against a target
Anecdote and memory
Gut feel
We rarely check

Your context

Used to calibrate the report. Company size and sector remain in the anonymized dataset; your email does not.

What the five levels look like

Every dimension in this assessment is scored 1 to 5. This is what the levels mean, dimension by dimension. The graded report diagnoses where your own answers land and what to do about it.

A closed loop exists

  1. 1None exist
  2. 2Cannot name one
  3. 3One, informally
  4. 4A few loops
  5. 5Loops everywhere

At the low end: Without a single closed loop, your AI cannot learn from consequences, only from inputs. Build one end to end on a low-stakes decision before scaling anything. What good looks like: Multiple closed loops mean your system learns from reality, not just data. Protect the measurement step; it is the first thing that quietly gets dropped.

Outcomes, not outputs

  1. 1Outputs only
  2. 2Activity metrics
  3. 3Some outcomes
  4. 4Outcomes tracked
  5. 5Outcomes drive next step

At the low end: Measuring activity instead of results tells you effort, not effect. Pick one recurring decision and define the single outcome that would prove it worked. What good looks like: Outcomes that drive the next decision are the definition of a learning organisation. Keep the outcome definitions honest; vanity metrics reopen the loop silently.

Recommendations reach action

  1. 1Die in decks
  2. 2Rarely actioned
  3. 3Sometimes actioned
  4. 4Usually actioned
  5. 5Action is default

At the low end: Recommendations that never become action make the whole loop impossible downstream. Track the action rate first; you may find the problem is not your models at all. What good looks like: When action is the default, the rest of the loop becomes worth building. Watch for silent vetoes where action is logged but nothing actually changes.

You know where it breaks

  1. 1No idea
  2. 2Vague sense
  3. 3One chain mapped
  4. 4Chains mapped
  5. 5Breaks monitored

At the low end: If you cannot see where the chain breaks, every fix is a guess. Map one chain link by link and mark the exact step where the trail goes cold. What good looks like: Knowing your break points means you fix loops instead of blaming algorithms. Monitor them, because break points move as processes change.

Someone closes the loop

  1. 1Nobody owns it
  2. 2Model team only
  3. 3Unclear ownership
  4. 4Named owner
  5. 5Owner with authority

At the low end: An unowned loop will not close itself; the measurement step has no natural champion. Name one person accountable for the whole chain, not just the model. What good looks like: A loop owner with authority is what keeps the chain closed under pressure. Make sure they can change the recommendation, not just report on it.