Anand's Four-Moves Capstone — Chapter 14 Worked Example ============================================================= The Copilot-first-output through four responsible-AI moves trace behind Anand’s headcount-allocation dashboard review in Chapter 14 of The Defensible Decision. Provenance, Fairness, Decision-Boundary, Disclosure — each move’s actions, purpose, and surfaced issue locked in one dataset. Wholly synthetic. Any resemblance to a specific real headcount-allocation dashboard is coincidental; the four-move pattern is drawn from analytics-governance patterns Chapter 14 catalogues. Files ----- - anand-headcount-fourmoves.csv — flat rows: one per move-action pair - anand-headcount-fourmoves.json — canonical scenario contract with source hash What it represents ------------------ Anand’s analyst produced a plausible one-page FTE-by-function dashboard in ~15 seconds from an eight-line Copilot brief. Aggregate attrition sits at 8% with a green treatment; the top-line looks calm. Anand runs the four responsible-AI moves in order before the CFO’s meeting: Provenance assembles the defensibility bundle; Fairness decomposes the aggregate attrition and reveals the under-30 cohort at 22% (2.75x aggregate); Decision-Boundary matches verification to role (recommendation tier + escalation clause for the archival risk); Disclosure amends the standard AI-directed footer to name the fairness callout as a deliberate post-generation addition. The pass surfaces two substantive issues that would have shipped without it. Invariants (locked) ------------------- The scenario carries executable invariants that Chapter 14’s release contract enforces: - Exactly 4 moves in canonical order: Provenance (1), Fairness (2), Decision-Boundary (3), Disclosure (4). - Every move has a purpose and at least one action; no action is empty. - Fairness move surfaces an issue about the under-30 attrition cohort. - Decision-Boundary move surfaces an issue about archival or escalation risk. - Copilot first output has NO provenance footer. - Aggregate attrition is 8% with a green treatment (the misleading calm top-line). - Ship verdict is ship after the pass. - At least 2 substantive issues surfaced by the pass. - Both the fairness callout and the escalation clause would NOT have shipped without the pass. - Defensibility bundle carries at least 6 items (matching Chapter 9’s bundle standard). Why the invariants matter ------------------------- Chapter 14’s pedagogical point is that responsible AI in analytics is a four-move discipline, not a single «check for bias» step. If the moves collapsed into fewer than four, one class of failure (provenance / fairness / decision-boundary / disclosure) would go unaddressed. If the pass surfaced zero substantive issues, the moves would look like ceremony. If the fairness cohort were even mildly worse than aggregate (say 10% vs 8%), the callout would be a stretch; the under-30 22% figure is deliberately large enough that the callout is unambiguously load-bearing. What it does NOT represent -------------------------- - Actual headcount data from any real financial-services firm. - Specific demographic categories beyond the age-band decomposition (the fairness move can decompose along tenure, function, gender, ethnicity, etc.; the dataset models only the axis that surfaces the issue in this scenario). - The CFO’s ultimate allocation decision (the chapter frames the artifact as the input to that decision, not the decision itself). Reproducibility --------------- The dataset ships from the book source at AlexBooks/applied-ai/ai-directed-viz/data/anand-headcount-fourmoves.json and is re-emitted here by AlexBooks/applied-ai/ai-directed-viz/scripts/generate-anand-fourmoves.mjs --site-root . License ------- Released for teaching use alongside The Defensible Decision. Use freely.