JD Responsibility to Proof Scoring Table
Source tier: Core Pack
Small sample of what to expect. The full pack includes more sections.
| JD Responsibility | JD Signal Being Tested | Strongest Evidence | Evidence Detail | Confidence | Interview Phrase | Preparation Focus | Likely Challenge Probe | Weak-Answer Trap | Full Response Excerpt |
|---|---|---|---|---|---|---|---|---|---|
| Build and deploy ML risk models | Model design, validation discipline, and production ownership | [Previous Firm] slippage/IS prediction model - XGBoost, walk-forward, Optuna, production-deployed | Feature stack covered liquidity, order-urgency, venue interaction, and market-impact proxies. Deployment used staged confidence weighting before full decision influence. | High | I built and deployed the slippage prediction model at [Previous Firm] - XGBoost, walk-forward validation, Optuna tuning, live decision integration. | Name feature families, validation design, and one concrete failure mode with corrective action. | Explain why walk-forward validation was chosen over random splits for this use case. | Backtest looked strong, so deployment was straightforward. | The model objective was pre-trade impact prediction. Inputs included order size vs ADV, urgency, spread state, realised volatility, and routing context. Validation used walk-forward windows to avoid temporal leakage. Deployment was production-grade, with model output feeding live execution decision support. |
| Order-flow analytics and client classification | Pattern recognition quality and risk-segmentation judgment | 7 years TCA at [Previous Firm] - systematic fill-quality analysis, adverse selection detection, execution pattern measurement | Frameworks separated toxic-flow indicators from benign execution variance and included account-level persistence checks to reduce false alerts. | High | My TCA work at [Previous Firm] is order-flow analysis - classifying execution patterns and identifying systematic fill anomalies. That is the analytical core of client risk classification. | Be explicit on signal logic: timing asymmetry, fill drift, and cross-account behaviour clustering. | How did you reduce false positives without missing high-risk accounts? | We used several indicators together and flagged unusual flow. | The analytical structure is the same as toxic-flow detection: identify persistent asymmetry, isolate behaviour that cannot be explained by normal market variance, and convert that signal into review-ready risk classification outputs. |
| Fraud and abuse detection | Method transfer versus direct investigative ownership | Adverse selection analysis in TCA - same analytical problem as latency-arbitrage detection | Detection methodology overlaps strongly with abuse-screening logic; direct investigation authority and legal process ownership are clearly separated. | Medium | My direct experience is through TCA. Adverse selection analysis is structurally identical to latency-arbitrage detection. The operational investigation workflow is where I would rely on the team initially. | Use bounded language: methodological overlap is strong; direct investigation ownership is not overstated. | Where does your responsibility stop and compliance-led investigation begin? | TCA and fraud detection are basically the same work. | Detection and investigation are separated explicitly: the model identifies systematic anomalies, human governance determines defensible action. That boundary is acknowledged directly rather than overstated. |