Add the operator x complexity composition table

README.md

@@ -49,6 +49,25 @@ answers are years and a same-`(s,r,o)`-other-time variant is ill-defined.
 **Per-question flags ship in `benchmark/functional_negatives.jsonl`** — restrict
 negative-dependent evaluation to the functional subset.
 
+## Composition
+
+Counts by temporal operator and hop complexity, over the full 8,710 questions
+(the 70/10/20 train/dev/test split is stratified by complexity):
+
+| Operator | 1-hop | 2-hop | 3+-hop | Total |
+| --- | ---: | ---: | ---: | ---: |
+| Point-in-time | 1,349 | 1,259 | 274 | 2,882 |
+| Before/after | 1,135 | 1,065 | 131 | 2,331 |
+| Interval | 403 | 435 | 26 | 864 |
+| Sequence | 1,113 | 1,241 | 279 | 2,633 |
+| **Total** | **4,000** | **4,000** | **710** | **8,710** |
+
+The 3+-hop column is structurally capped, not undersampled. `tkgl-smallpedia`
+is point-in-time, so a *k*-hop chain needs every hop valid in the same year, and
+Wikidata's year-density around an anchor entity is 0–3 facts/year — long chains
+that also satisfy answer-uniqueness are simply rare. Validity-window TKGs
+(YAGO3, ICEWS) would lift this.
+
 ## Quickstart
 
 Three steps, standard library only, no install. Scoring your own retriever