Ask an organisation how its last big collective session went and you will be told the satisfaction scores. They were good. They are almost always good. Six months later nothing has changed, and nobody treats the two facts as related.
They are related. A satisfaction score measures how a day felt to the people in the room, and a well-run day feels good more or less regardless of whether it produced anything durable. It is the one measure guaranteed not to distinguish a session that built understanding from a session that was merely pleasant. Every organisation buying this kind of work eventually asks the harder question, how would we know it worked, and the honest answer has to be available before the work starts, not assembled afterwards from whatever happened to be measurable.
Here is what to look for. None of these can be collected on the day, which is the point.
Whether the reasoning survives the conclusion
The first test is the simplest and the most brutal. Take someone who was in the room, six months on, and ask them not what was decided but why. Then ask someone who was not in the room and joined afterwards.
In most organisations the first person can reconstruct fragments and the second has nothing but the decision itself. That gap is the whole problem. A decision travelling without its reasoning is a rule, and rules get followed until they stop fitting, at which point nobody can tell whether the situation has changed or the rule was always wrong. When the reasoning is preserved and navigable, the second person inherits the understanding rather than the instruction, and can tell the difference.
Ask people what was decided but why to see if the reasoning survives the conclusion.
Whether the positions that lost are still legible
Every consequential decision has a road not taken, and someone argued for it. In a process that worked, that argument is still findable, still attributed, and still makes sense on its own terms.
This sounds like sentiment and is not. The minority position is the organisation's early warning system. When the chosen path starts producing surprises, the fastest route to understanding what is happening is the person who predicted it, and the case they made, in the form they made it, before anyone knew who was right. An organisation that keeps only its conclusions has thrown away its own best diagnostic. One that can retrieve the dissent, with the knowtype attached so it is clear whether the objection came from research, from experience, or from a values position, can act on it in days rather than quarters.
Whether anything appeared that nobody brought
This is the signal that separates deliberation from every adjacent activity, and it is the one worth being strictest about.
A workshop can produce a good summary of what the participants already thought. A survey can produce a precise distribution of what they already thought. Neither is a failure; both are what those instruments do. Deliberation is supposed to produce something else: a cluster of understanding that no participant walked in holding, formed in the contact between perspectives that had not previously met.
When Hunome ran a global deliberation on demographic change with thirty-six participants across the world, seven distinct clusters of understanding emerged. None had been specified in advance. No research brief could have specified them, because specifying them would have required already knowing what the deliberation was for finding out. That is the measurable form of the claim: count the findings that were not in anyone's opening position. If the number is zero, the process was a very good consultation.
Whether the understanding is still being used
Most collective work produces an artefact with a short half-life. The report is read in the fortnight after it lands and then cited from memory, increasingly inaccurately, until it is superseded by the next one. The test of a durable process is whether its output is still being added to.
A living structure gets returned to when a new question arrives, extended when someone joins with a perspective the original group lacked, and consulted when a decision downstream touches the same territory. A document is read or not read. The distinction is not about format. It is about whether the thing preserves enough structure to be continued. If the output of a process cannot be extended six months later without redoing the process, it was an event, not a capability.
Whether the alignment is real where it looks real
The most expensive failure in this category is not disagreement. It is the appearance of agreement over an unexamined difference: everyone nodding at a word that means something different to each of them, discovered eighteen months later at the point of execution.
This is what the Shared Understanding Index is built to expose: where alignment across a SparkMap is genuine and where it is nominal. A number that shows understanding is thinner in one area than the confident language around it suggests is more valuable than any satisfaction score, because it points at the specific place where more work would pay. Agreement that has never been tested is not a result. It is a risk that has not surfaced yet.
What to ask for
Before commissioning any process that claims to build shared understanding, ask what it will leave behind that can be interrogated in six months, and by someone who was not there. Ask whether dissent will be retrievable with its grounds intact. Ask how you will identify what emerged that nobody brought.
A process that can answer those questions is building a capability. A process that offers satisfaction scores is selling an event. Both have their place, but only one of them is still there when circumstances change, which is the only moment any of it was ever for.