Beginner question: What counts as a 'touch' in your model?

I'm evaluating several attribution platforms for a new project, and I keep hitting a foundational question: what exactly constitutes a 'touch' in your model? The marketing materials are full of the term, but the technical implementation details are often buried.

For example, I've seen platforms handle this in wildly different ways:
* **Platform A** logs a touch for any tracked event (page view, button click) with a new session ID, counting multiple events in one user visit as a single touch.
* **Platform B** uses a configurable rule engine where a touch is only counted on specific, goal-oriented events (e.g., `add_to_cart`, `start_checkout`).
* **Platform C** seems to deduplicate touches based on a combination of `user_id`, `channel`, and a time window (e.g., only one touch per channel per user per 24h).

This has major implications for the resulting attribution weights. If I'm instrumenting my own events, I need to know what the platform will actually *do* with them. My specific questions are:

1. Is a 'touch' strictly a discrete event your SDK captures, or can it be a derived state from a sequence of events?
2. How do you handle deduplication logic? Is it based on session, user, or campaign identity, and is that window configurable?
3. In a cookieless or blended environment, how does this definition change? If you're using probabilistic stitching, does a 'touch' then represent a confidence score rather than a binary event?

I'm looking for concrete, technical answers, not high-level philosophy. The granularity of the touch definition directly impacts the model's output, and I need to align that with our internal tracking semantics.

benchmark or bust