https://github.com/JMdictProject/JMdictIssues/issues/122

Setting this up as an example of the approach developed in issue 122.

@@ -5 +5 @@
-<keb>前門の虎後門の狼</keb>
+<keb>前門の虎、後門の狼</keb>
@@ -8 +8,2 @@
-<keb>前門の虎、後門の狼</keb>
+<keb>前門の虎後門の狼</keb>
+<ke_inf>&sK;</ke_inf>
@@ -11 +12 @@
-<reb>ぜんもんのとらこうもんのおおかみ</reb>
+<reb>ぜんもんのとら、こうもんのおおかみ</reb>

Apologies - I didn't look hard enough at Stephen's proposal. I agree we shouldn't be using such fragments as "rK" forms, at least at this stage.

Whether or not we want to use search-only forms in this way, I don't think it's safe to do so as long as they're included with the other forms. Many sites/apps don't support the new tags yet. Some probably never will.
For the next generation of JMDict, I'd like to see search-only forms handled separately (i.e. in a different part of the XML). Then we can at least consider ideas like this.
And I agree with Marcus regarding the n-grams. The fact that both 前門の虎 and 後門の狼 have nearly identical counts suggests that they're almost always used together. Online search results appear to confirm this.

@@ -9,16 +8,0 @@
-</k_ele>
-<k_ele>
-<keb>前門の虎</keb>
-<ke_inf>&sK;</ke_inf>
-</k_ele>
-<k_ele>
-<keb>後門の狼</keb>
-<ke_inf>&sK;</ke_inf>
-</k_ele>
-<k_ele>
-<keb>前門のとら</keb>
-<ke_inf>&sK;</ke_inf>
-</k_ele>
-<k_ele>
-<keb>後門のおおかみ</keb>
-<ke_inf>&sK;</ke_inf>

I'm not sure we should be using sK for things like this? Feels like we could be opening up a can of worms here. The reason the individual parts get more hits in the ngrams is frequently a problem with the searchability of the ngrams rather than real evidence the individual parts are frequent on their own.