ExecGraph Research
The graph is compounding faster than the database.
Since August, recorded relationships have grown far faster than the number of people represented. Under the fixed reference, GraphSAGE has moved from trailing profile similarity to leading it on both primary measures. The next question is whether those different recommendations are actually more useful to sellers.
Recorded snapshots: August 21 to October 16, 2026Methods, sources and limits included
+47.5%People in the graph47,326 to 69,810
10.13xRecorded relationships21,348 to 216,169 (+912.6%)
4.30xPeople with a connection through October 49,289 to 39,944 (+330.0%). October 16 not reported.
The opportunity is to build relationship context around people, not merely add names. More structure gives the network model more to evaluate. Better sales recommendations remain the goal, not a conclusion.
Research notes
Latest recorded test: October 16, 2026EXG-RN-001 · Graph learning / Link prediction / Energy
Recorded relationships reached 10.13 times the August count, while the graph population grew 47.5%. On October 16, GraphSAGE scored 0.9205 ROC AUC against 0.8890 for the semantic reference under the unchanged historical test. Its alternate Warm Path V2 shortlists remain unjudged for commercial usefulness.
Fixed-reference series: August 21, 2026 to October 16, 2026Supplementary comparisons: September 20, 2026 to October 16, 2026Version 1.4Not peer reviewed
GraphSAGE AUC across the checkpoints
*September 20 used the changed setup and remains supplementary. October 16 is the fourth fixed-reference point. On October 16 the semantic reference scored 0.8890. Native V2 ranking is a separate comparison.
Three questions the testing must answer
Can the model recognize documented connections?Test how well it distinguishes a recorded professional connection from a pair with no recorded connection.Measured in the fixed-reference relationship tests.
Does it recommend different people?Compare its contact shortlist with the existing Warm Path recommendations, using the same eligible people.Compared for 632 targets on September 20, October 4 and October 16.
Are those recommendations useful?Ask independent reviewers whether the contacts fit the sales opportunity and the proposed connections have supporting evidence.Not yet independently judged.
A higher test score or a different contact list does not, by itself, mean a better sales recommendation.
Research and data partnerships
Explore a scoped research or data partnership.
ExecGraph welcomes inquiries from AI labs and research teams studying graph representation learning, relationship prediction, temporal robustness, or graph-assisted retrieval in industrial domains.
Proposals can cover controlled evaluation or a potential licensed training study. Describe the research question, intended data use, required fields, evaluation design, and security arrangements. Availability depends on source rights, third-party terms, privacy review, and a separate written agreement. No underlying dataset or model-training license is granted by these pages or by ordinary platform access.
Inquiries go to support@execgraphenergy.com. No raw records, proprietary feature definitions, or model artifacts are published here.