Editorial roadmap

Six evidence-rich pieces, not a content factory

A ninety-day plan organized around the professional questions this site can answer with original artifacts and defensible scope. These are briefs, not published articles.

Brief 1

Governed Data & Analytics

What Makes Institutional Data Trustworthy?

Direct answer to prove: Trust is earned when a decision-maker can inspect a measure's definition, source, owner, period, transformation, review status, and known limitations.
Original artifact: A sanitized evidence table and source-to-decision lineage diagram.
Evidence plan: Primary governance standards, public institutional guidance, and confidentiality-safe first-hand operating context.
Boundary: No employer records, screenshots, internal schemas, or unverified impact metrics.

Brief 2

Governed Data & Analytics

What Makes Data AI-Ready?

Direct answer to prove: AI-ready data is not merely clean: its definitions are agreed, its origin is traceable, access matches responsibility, it has test cases, and a person reviews what comes out.
Original artifact: A readiness rubric contrasting analytic, retrieval, and decision-use requirements.
Evidence plan: Primary risk-management standards and public technical documentation.
Boundary: The rubric will distinguish general controls from domain-specific legal or safety obligations.

Brief 3

AI Systems & Evaluation

How to Evaluate an AI Agent Before Production

Direct answer to prove: Evaluate the complete decision loop—task selection, tool use, permissions, evidence, abstention, recovery, and human escalation—not just answer quality.
Original artifact: A test matrix with normal, adversarial, ambiguous, and recovery scenarios.
Evidence plan: Official evaluation guidance, system documentation, and reproducible test fixtures.
Boundary: No claim of production model development; the piece focuses on governance and evaluation design.

Brief 4

AI Systems & Evaluation

What an AI Agent Audit Trail Should Contain

Direct answer to prove: A useful trail reconstructs intent, identity, inputs, retrieved evidence, tool calls, permissions, outputs, overrides, and the final accountable decision.
Original artifact: A vendor-neutral event schema and a worked fictional trace.
Evidence plan: Primary audit-control standards and official platform documentation.
Boundary: The example will use fictional data and will separate observability from legal compliance advice.

Brief 5

Quantitative Finance, Risk & Decision Science

Why Portfolio Optimization Fails Out of Sample

Direct answer to prove: Optimization amplifies estimation error: small changes in expected returns and covariance can produce large, unstable allocations that disappear outside the sample.
Original artifact: A reproducible notebook comparing naive, constrained, resampled, and robust allocations.
Evidence plan: Original research papers, documented assumptions, public market data, and tested code.
Boundary: Educational research only; results will not be presented as a strategy or investment track record.

Brief 6

Quantitative Finance, Risk & Decision Science

How to Backtest Without Lying to Yourself

Direct answer to prove: A credible backtest fixes the hypothesis first, prevents leakage, accounts for costs and repeated trials, tests regime sensitivity, and reports failure modes beside returns.
Original artifact: A backtest review checklist plus a notebook that demonstrates common leakage and selection errors.
Evidence plan: Primary research on backtest overfitting, reproducible code, and explicit data lineage.
Boundary: No cherry-picked live-performance implication and no investment advice.

Sequencing should be revised after Search Console and Bing Webmaster Tools expose real query data. A future piece on Goodhart’s Law in operating dashboards belongs in the next queue once a concrete decision-table artifact is ready.

The publication standard every piece must meet →