Metadata is enough, until it isn't.
The less you collect, the less you can know. That constraint can be a feature rather than a defect.
There is an obvious temptation when building a data product: collect more. More fields, more context, more text, more identifiers. More information appears to promise a better answer.
But additional information changes the product you are building. If the question is whether work is occurring between organisations, the contents of the work can be unnecessary to the measurement.
That constraint is useful. It forces the signal to earn its value from structure: which organisations are interacting, whether verified activity is increasing or decreasing, how unusual the change is relative to a baseline, and whether the pattern persists.
There are things this approach cannot tell you. It cannot tell you what somebody promised in a message. It cannot tell you sentiment. It cannot reconstruct a negotiation. Those are not accidental omissions. They are boundaries.
Boundaries matter because a signal should not need to become surveillance in order to become useful.
The harder methodological question is therefore not 'how much more can we collect?' It is 'what is the minimum observation that still carries economic information?'
We expect the answer to evolve. The principle should not.