OK, I'll try. First, it is important to note that we do not look up any compounds in JChem tables; we use MolSearch to compare targets derived from a student response to queries that a question author specifies.
The student sketches the mechanism of a given reaction. The mechanism consists of a series of steps, each in a box, connected by graphical arrows. In each step are some compounds and electron-flow arrows.
We evaluate the mechanism that the student drew in many different ways. One way is to make sure that the student used all of the starting materials that they were supposed to use. We classify the compounds in the student's mechanism as response products (have no electron-flow arrows touching them), response starting materials (have electron-flow arrows touching them and not produced by electron-flow arrows in a previous step), and response intermediates (neither response starting materials not response products). We then compare the response starting materials to the starting materials that the question author requires them to use. (In some cases, a starting material may have different acceptable forms; e.g., HCl or H+ may both be acceptable.)
Now, consider the question, "Draw the mechanism of the Heck reaction between PhBr and CH2=CHCO2Et catalyzed by (Ph3P)4Pd." An author writing the question would require the student's mechanism to contain the starting materials PhBr, CH2=CHCO2Et , and (Ph3P)4Pd. But if the student writes a mechanism in which CH2=CHCO2Et coordinates to (Ph3P)2Pd(Ph)Br, he would place a multicenter attachment point on the π bond so that he could show the formation of a dative bond. And then the search for CH2=CHCO2Et in the mechanism would fail.
Now, the author could add the multicenter attachment point to the CH2=CHCO2Et that JChem is supposed to look for in the mechanism. But, aside from this additional step being counterintuitive for the author, suppose the student draws a different mechanism, one in which the insertion of the π bond into the Pd-Ph bond occurs without prior coordination of the π bond to Pd. Here the student would not add a multicenter attachment point to the π bond. Now, you might not want to mark such a mechanism correct, but regardless, you wouldn't want the program to say that the student failed to use CH2=CHCO2Et as a starting material.
Finally, it would be possible to require the author to write every starting material twice: once with the multicenter attachment point, and once without, and then compare the response starting materials to each one separately. But this would be very awkward, and besides it would require the author to anticipate any of the multicenter attachment points that a student might draw, which could be legion.
So, what I am now doing is looking at every atom in a target and removing it (and its associated Sgroup) if it is a bondless multicenter attachment point. But again, the sensible approach is to modify the search algorithm to say that a bondless multicenter attachment point is a ghost and is ignored, at least by default.