D18-F04-A12 · QUALITY & DISTRESS · SYNTHETIC DATA

Dechow-Dichev Accrual Quality

Change one synthetic accounting driver, watch rolling past/current/future CFO window, fitted accrual, residual, and dispersion recompute, and explain the result without crossing the minimum sample and future-CFO availability gates.

The learning loop

1 · PredictName the direction before moving a driver.
2 · InspectFind the ratio, signal, contribution, or residual that changed.
3 · ReconcileTie the components to the raw score or model residual.
4 · BoundInterpret the screen without turning it into a finding.
Make a prediction before stepping

Base → current bridge

Scenario base
Current state
Headline delta

Keyboard: Left/Right step, Home resets, and Space plays or pauses when focus is outside a form control.

Current model output

Frozen equation

WCA_t = β0 + β1 CFO_t-1 + β2 CFO_t + β3 CFO_t+1 + ε_t; AQ = sqrt(sum(ε²)/(n-k))

Component and diagnostic ledger

Changed cards are purple. Start here before interpreting the headline.

Response across 61 states

Model output across the selected accounting scenarioA deterministic synthetic sensitivity line with the current state marked.

Interpretation gate

Boundary: the minimum sample and future-CFO availability gates

Visible intermediate: rolling past/current/future CFO window, fitted accrual, residual, and dispersion

Open exact point-in-time input and structured output

Input

Output


Synthetic teaching data only. A model score or residual is a historically specified screen—not an audit conclusion, rating, fraud finding, default forecast, or investment recommendation.