Research · Nodes
Research
Explore how connected records can inform useful action, and how to evaluate what changed afterward. The library below includes the historical insurance study, its supporting methods, and legal and governance explainers. Each finding retains its original application, population, and limits. For today's product, see the Nodes overview.
the paper · arxiv · econ.em · april 2026N=10,765 hires · one F500 carrier
Decision Traces: What Multi-System Data Fusion Reveals About Institutional Knowledge in Enterprise Hiring
- In the one-carrier study, none of 3,597 testable keywords from 8,181 parsed skills positively predicted the first production milestone after Bonferroni correction; thirty were anti-predictive.
- The reference analysis estimates $54.35 in additional annual production per producing person for each day faster to the first production milestone. This fitted relationship is a planning estimate, not a guaranteed return from accelerating a hire.
- In an exact n=50 top-performer subset, the industry-experience rule would have excluded 80%. A separate retrospective counterfactual put $17.7M in annual production at risk among producing agents the rule would have rejected. Neither finding shows actual exclusions or incremental revenue caused by Nodes.
Suggested citationShafiq, Saad Bin. "Decision Traces: What Multi-System Data Fusion Reveals About Institutional Knowledge in Enterprise Hiring." arXiv:2604.19819, 2026.
- The Digital Omnibus and AI hiring: what changed, what did notThe EU Digital Omnibus moved high-risk obligations for employment AI to December 2, 2027. Article 50 transparency duties still start August 2, 2026. What changed, and what to do with the runway.
- Mobley and the agent theory: why the human gate is the architecture, not a featureMobley v. Workday is active litigation. The court let claims proceed on an agency theory: delegate rejecting and advancing candidates to a tool, and the vendor can stand in the employer's shoes.
- Decision Traces in Hiring: Multi-System Data FusionDecision traces connect ATS, HRIS, and behavioral data. At one carrier, none of 3,597 testable resume keywords predicted production after correction.
- Do Resume Keywords Predict Job Performance? 10,765 Agents HiredIn a study of 10,765 agents hired, zero of 3,597 resume keywords predicted production after correction, and 30 were anti-predictive.
- Why ATS Keywords Fail to Predict Job PerformanceATS keywords describe past experience but do not forecast performance. A 10,765-agent study tests resume signals against production outcomes.
- Does a College Degree Predict Job Performance?In a study of 10,765 agents hired, a college degree had a weak association with production, with overlapping confidence intervals across education levels.
- The $17.7M Cost of a Single Hiring FilterOne screening rule would have rejected 2,863 producing agents worth $17.7M in annual premium credit. How connecting ATS and HRIS data exposes the cost of a filter.
- AUC 0.647 to 0.735: Data Fusion in Hiring PredictionIn an evaluable n=229 subset at one carrier, personality reached AUC 0.647 and fusion reached 0.735. What those research results mean.
- Which Personality Types Predict Sales Performance?In an evaluable n=229 subset at one carrier, personality reached AUC 0.647. The result does not support screening on a type alone.
- Leadership Drive and Sales Performance: An Observed CurveIn a 10,765-agent study, leadership drive followed an inverted-U pattern. The modest observational result is a question to test, not a screening rule.
- The Economics of Speed to Production in Enterprise HiringAt one insurance carrier, each day faster to the first production milestone was associated with $54.35 more annual production per producing hire.
- The Speed-Volume Trade-Off in High-Volume HiringAt one carrier, hiring volume, source mix, production rate, and ramp speed moved together. Connected outcomes exposed the trade-off without claiming causality.
- Prediction and Moderation: Two Tests for Hiring AIPrediction ranks who may reach an outcome. Moderation tests whether a defined condition changes that outcome differently across groups.
- When a Manager Retires, Their Judgment LeavesDecades of hiring judgment disappears when a manager retires, because it lives in no system. How decision traces capture institutional knowledge as a queryable record.