StatGazerDecision context
Name the decision first. Allocation, sizing, risk, vendor diligence, internal governance, and research triage require different evidence thresholds.
Review standard
Model review before the decision hardens.
A public checklist for investment teams reviewing forecasting, backtesting, risk, and ML models before capital, governance, or diligence depends on them.
The checklist
Six questions before anyone trusts the model.
The point is not to make review ceremonial. It is to create a decision-useful record: what was reproduced, what was challenged, what failed, what remains unresolved, and what should happen next.
Data and timing
Check as-of dates, report dates, joins, revisions, universe construction, survivorship, and whether every feature existed at the claimed decision time.
Reproduction path
Regenerate key results from raw or minimally processed data where access allows. Document code paths, parameters, environments, and manual steps.
Validation design
Review train/test structure, walk-forward tests, regime sensitivity, calibration, costs, liquidity, capacity, and benchmark fit.
Findings taxonomy
Separate confirmed defects, material assumptions, unresolved questions, limitations, and remediation items. Not every concern has the same decision impact.
Handoff and boundary
Deliver an issue register, remediation path, and handoff notes. Technical review is not financial-statement audit, regulatory assurance, investment, legal, or tax advice.
Regulatory vocabulary, practical use
The useful lesson from model-risk frameworks.
SR 11-7 is a banking framework, not a magic label for every investment process. The useful parallel for non-bank teams is simpler: independent challenge, conceptual soundness, outcomes evidence, documented limitations, monitoring, and ownership. That vocabulary is valuable when it keeps a review disciplined without pretending to be regulatory assurance.
Use internal MRM or assurance when
Your organization needs policy ownership, recurring controls, attestation, regulated assurance, or an enterprise model-risk program.
Use focused technical review when
The gap is concrete evidence: a reproduced result, leakage checks, validation design, a finding taxonomy, and a memo the team can inspect.
Related notes
Use the standard to read the notes.
The first notes apply the checklist to reported versus reproduced Sharpe, finding classification, and SR 11-7-style discipline outside banking.