Changes in version 0.1.0 (2026-11-13)                  

Initial Release

docketscape provides a statistical modeling framework for decomposing
court delay into substantive legal complexity ("what a case is") and
institutional congestion ("where the case sits") using network
representations and survival analysis.

Core Features

  - Case Standardization & Censoring (clean_cases):
    
      - Validates case-level filing and decision dates.
      - Handles right-censored pending matters relative to observation
        cutoffs.
      - Explicitly flags invalid records (e.g. decisions preceding
        filing, filings after observation end) without silent deletion.
      - Masks durations to NA for invalid cases to prevent non-positive
        durations in downstream models.

  - Statutory Network Representation (build_section_graph):
    
      - Constructs undirected section co-occurrence networks from case
        citations.
      - Employs sparse matrix crossproducts (Matrix::crossprod) for
        scalable graph construction.
      - Preserves edge weight invariants (strictly positive integer case
        co-occurrence counts).

  - Profile Diffusion (diffuse_profiles):
    
      - Smooths sparse statutory citation profiles across co-cited
        sections via one-step convex combination diffusion.
      - Enforces row-stochasticity (unit row sums, $L_1 = 1.0$) across
        all tuning values $\alpha \in [0, 1]$.

  - Case Similarity Graph (build_case_graph):
    
      - Constructs $k$-nearest-neighbour case graphs using inverse
        document frequency (IDF) weighted cosine similarities.
      - Uses chunked sparse evaluation and integer-ranked tie-breaking,
        scaling to large dockets within minimal memory.
      - Enforces graph invariants: no self-loops, no multigraph edges,
        maximum similarity weights on mutual edges.

  - Neighbourhood Survival & Relative Lateness (neighbourhood_survival,
    relative_lateness):
    
      - Computes weighted Kaplan-Meier survival curves from legally
        similar peers.
      - Evaluates relative lateness percentiles ($1 - S(T)$) anchored in
        empirical peer timelines.
      - Guards against extrapolation: flags unestimable cases when
        durations exceed peer follow-up.

  - Institutional Congestion Indexing (court_load):
    
      - Measures baseline court docket load at the monthly filing
        interval.
      - Implements fast $O(N \log N)$ findInterval() event tracking.

  - Semi-Parametric Delay Modeling (fit_delay_model, decompose_delay):
    
      - Fits Cox proportional hazards regression on legal scores and
        baseline court load.
      - Handles small sample sizes ($N < 3$ or zero events) gracefully
        without numerical crashes.
      - Decomposes linear predictors into legal and court load absolute
        contribution shares.

  - Leakage-Aware Validation & Auditing (validate_models,
    project_validation_legal_score, check_*):
    
      - Enforces strict chronological temporal holdouts preventing
        future-neighbour leakage.
      - Projects validation cases into training representations without
        validation-to-validation contamination.
      - Provides diagnostic audits: check_future_neighbours(),
        check_court_overlap(), and check_near_duplicates().

Methodological Safeguards

  - Observational and non-causal framework.
  - Does not rank judges, evaluate judicial performance, or assess
    judicial diligence.
  - Contains zero bundled empirical DDL records; all package examples
    and tests use synthetic data structures.