Methodologies / Advanced Labs Methodology 16
Predictive Accountability Architecture
The Forecast Vintage is the organizing concept. The methodology is used in Advisory Practices.
KATHLEEN REYNOLDSEXECUTIVE GOVERNANCE | INCIDENT COMMANDWritten By Kathleen Reynolds Advisory
Read This Methodology As Text
Forecasts are versioned assertions about future states, not verified outcomes.
The Stages
- Define. Specify the decision, affected parties, time horizon, material downside and falsifiable claims. Here, a fictional regional service provider forecasts an increase in demand after expanding paid acquisition. The target and horizon are defined first.
- Authenticate. Separate observed evidence, unverified inputs, estimates and validated results; retain source and cutoff. The forecast vintage is frozen at its source cutoff, and each input keeps its source and provenance class.
- Challenge. Construct the strongest rival hypothesis and identify discriminating evidence. Marketing treats the forecast as a growth opportunity. Finance identifies working-capital strain, Operations identifies fulfillment constraints and Compliance identifies geographic eligibility questions.
- Compare. Evaluate reversible, conditional, deferred and full interventions against a documented baseline. A no-change baseline, phased expansion and full expansion are compared, with a stated range rather than a single number.
- Authorize. Assign a human or governing body with actual decision rights; record conditions and exception triggers. The record names who can halt the experiment and what constitutes unacceptable downside.
- Observe. Measure outcomes, error and unintended consequences without changing the original forecast vintage. Calibration, drift and asymmetric loss are monitored as observations arrive.
- Reconcile. Preserve amendments, contradictory events and subsequent decisions as linked but distinct records. A later forecast becomes a new vintage, linked to the first. The governance disposition is preserved.
What It Does Not Claim
- A wide range can produce nominal accuracy without decision usefulness; small samples and changed conditions invalidate simplistic grades.
- A well-calibrated forecast may still be economically useless if it arrives too late or targets the wrong variable.