DecisionDrift 0.2.0

New features

• dd_decompose() – flow decomposition of each change in the decision rate into inflow (uptake) and outflow (reversal) using the marginal-transition identity. Reports turnover, net endpoint change (NEC), cumulative marginal burden (CMB), cumulative transition burden (CTB), and, for unbalanced panels, the gap between paired and cross-sectional change.
• dd_group_decompose() – symmetric midpoint decomposition of aggregate change into within-group and composition components, composition drift D_C, and group-specific uptake and reversal. Supports time-varying group membership.
• dd_cluster_bootstrap() – unit-level cluster bootstrap that keeps each unit’s full history. One set of draws can be shared by all inference functions.
• dd_global_test() – global test that uptake, reversal, and/or prevalence are constant across waves, overall or by group. Max-t and Wald statistics with a centred bootstrap null; familywise-adjusted p-values and simultaneous confidence bands. Contrasts compare each step with the previous one (contrast = "consecutive", default) or with the first one (contrast = "baseline", more power against gradual drift).
• dd_simulate() – simulate panels under stable, marginal, masked, abrupt, group-masked, and composition-change scenarios, or custom uptake and reversal sequences.
• inst/simulation/mc_study.R – script for the Monte Carlo study comparing the prevalence trend test with the global stability test.
• dd_decompose, dd_group_decompose, and dd_global_test have print() and plot() methods.

Changes to existing functions

• dd_audit() no longer returns "no drift detected", "marginal drift", "moderate drift", or "strong drift". Those labels came from fixed, uncalibrated cut-offs. The audit now runs one cluster bootstrap, the flow and group decompositions, and dd_global_test(), and returns a conclusion stating the test result at level alpha. verdict is kept as an alias of conclusion. New arguments: global_test, components, alpha, thresholds (user-supplied practical cut-offs), and seed.
• dd_changepoint(): the event-alignment result no longer has a "strong"/"weak" evidence label; it returns alpha and a logical significant. Documentation now notes that the wave-level methods have little power with few waves.

Bug fixes

• dd_transition(): reversal_drift held the mean absolute change in uptake P(1|0), not reversal. It now holds the mean absolute change in reversal P(0|1). The uptake value is now returned as uptake_drift. TDI values are unchanged. Code that used reversal_drift from 0.1.0 was reading uptake drift and should switch to uptake_drift to keep the old numbers.
• dd_transition() and dd_indices() documentation now gives the TDI and CDB formulas in terms of uptake, reversal, and the decision rate.
• plot.dd_transition() now shows uptake and reversal.
• Repository URLs updated to github.com/subirhait.

DecisionDrift 0.1.0

Initial release

Core modules

• dd_build() — construct a drift_panel object from long-format panel data.
• dd_prevalence() — detect drift in decision rate; computes Decision Drift Index (DDI).
• dd_transition() — detect drift in transition structure; computes Transition Drift Index (TDI).
• dd_entropy_trend() — track rolling Shannon entropy and switching rate over time.
• dd_group_drift() — detect group-differential drift; computes Group Differential Drift (GDD).
• dd_changepoint() — change-point and regime-shift detection via CUSUM, segmented regression, and event alignment.

Summary indices

• dd_indices() — compute all four original drift indices in one call: DDI, TDI, GDD (Group Differential Drift), CDB (Cumulative Drift Burden).

Robustness & sensitivity

• dd_robustness() — test stability of drift conclusions across analytic choices (balanced panel, leave-one-period-out, leave-one-group-out, min_waves grid, bootstrap CI).
• dd_sensitivity() — probe vulnerability to data problems: decision miscoding, missing-wave attrition, threshold shifts, and subgroup composition shifts.

Flagship audit

• dd_audit() — integrated three-layer audit (detection → decomposition → stress-testing) with a single-call verdict.

Plotting

• S3 plot() methods for all module outputs.
• Multi-panel composite via plot.dd_audit() (requires patchwork).

Relationship to decisionpaths

• decisionpaths asks: What happened? (path construction, DRI, entropy, equity)
• DecisionDrift asks: Did the system change? (drift detection, decomposition, stress-testing)
• These packages form complementary layers of a longitudinal decision audit framework.