After, the <content> tag has been examined, the builder moves on to the next condition, the <triple> which looks like this:

<triple subject="?start"
        predicate="http://www.xulplanet.com/rdf/relatedItem"
        object="?relateditem"/>

The <triple> condition is used to follow arcs or arrows in the RDF graph. The predicate attribute specifies the labels of the arrows to follow. In the triple used here, the predicate is "http://www.xulplanet.com/rdf/relatedItem", so we know that the builder will look for arrows with that label. The builder can either follow the arrows in a forward or backward direction, but only one direction per condition. It determines which direction to use by examining which data is known and which data is not known yet. Recall the data in the potential results set:

(?start = http://www.xulplanet.com/rdf/A)

This was the seed data created by the <content> condition. You might be able to guess that the builder fills in the value of the ?start variable as the triple's subject, giving something like this:

<triple subject="http://www.xulplanet.com/rdf/A"
        predicate="http://www.xulplanet.com/rdf/relatedItem"
        object="?relateditem"/>

The builder doesn't actually change the triple, but this might clarify how the builder is working. The builder looks at both the subject and object of the triple and tries to resolve any of the variables using the known data. The ?start variable has a value "http://www.xulplanet.com/rdf/A", so that is filled in. The ?relateditem variable doesn't have any known value, so it will be left as is. Once the variables are filled in, the builder can query the RDF graph.

The query will look for all arrows that start at the node "http://www.xulplanet.com/rdf/A" with the predicate, or arrow label "http://www.xulplanet.com/rdf/relatedItem". Since ?relateditem isn't known, the builder will allow any value for the node the arrow points to, and will look in the datasource for all possible values. In a <triple> condition, the subject is always the start of an arc, while the object is what it points to. For this triple, the builder will follow the arrows in the forward direction. Here is the RDF graph again:

Starting at node A and following the relatedItem arcs, we can see that there are three possible values for the ?relateditem variable, B, C and D. This is new data, so the builder adds it to the graph. Since three values have been found, the network will now have three potential results:

(?start = http://www.xulplanet.com/rdf/A, ?relateditem = http://www.xulplanet.com/rdf/B)
(?start = http://www.xulplanet.com/rdf/A, ?relateditem = http://www.xulplanet.com/rdf/C)
(?start = http://www.xulplanet.com/rdf/A, ?relateditem = http://www.xulplanet.com/rdf/D)

You may note that the ?start variable is repeated for each result. This is because the builder copies the existing data for each new potential result and adds the new data. Internally, this isn't quite true; the builder actually maintains only one copy of the similar data but uses data structures in such a way which make it appear as if it is duplicated.

You may find this a bit confusing, but this should become clearer later with more specific and practical examples.

Since the <triple> was the last condition, the builder moves on to the content generation phase, creating matches out of the three potential results.