Changes in version 0.2.0 (2026-10-15)                  

First CRAN release. The method is unchanged from 0.1.0 and follows the
published paper (Jamme, 2027, doi:10.1007/978-3-032-37883-5_9). This
release focuses on robustness and documentation. The API may still
evolve until 1.0.0.

Changes

  - pm_plot_risk_max() gains x_axis = "n" and overlays the frontier
    found by pm_suggest_n() (mark_frontier).

Bug fixes

  - pm_calib_dominance() now passes level to the CI loss metric (it was
    always computed at 95%).
  - Filtering calibration tables on sigma_nu, n, beta or rho is no
    longer affected by floating-point representation of seq() grids.

Internal

  - purrr and dplyr are no longer a dependency.
  - Unit tests added.

                        Changes in version 0.1.0                        

First working version of the package. The API is still unstable:
function names and arguments may change without deprecation until
version 1.0.0.

Parameters

  - pm_params() gathers every choice in a single object: the policy,
    organised by disclosure scenario (dominance, prule, diff), and the
    three mechanism parameters (sigma_nu, sigma_eps, n), filled in as
    calibration proceeds.

Calibration — step 1, the differencing noise

  - pm_calib_diff() returns the theoretical grid of sigma_eps values
    meeting a given (beta, tau) couple, with the loss floor they imply.
  - plot() on that grid draws the risk-utility frontier.
  - pm_commit_diff() records the decision and sets sigma_eps.

Calibration — step 2, the dominance noise

  - pm_calib_dominance() returns, over a grid of dominance levels rho,
    the scenario-I risk and the information loss for each candidate
    (sigma_nu, n).
  - summary() reduces that grid to the worst-case risk and the loss
    range of every parameter combination.
  - pm_suggest_n() solves for the largest n meeting the risk ceiling —
    the calibration rule of step 2 — with an optional safety margin.
  - pm_commit_dominance() records sigma_nu and n, and reports the
    resulting trade-off.

Figures

  - pm_plot_risk_profile(), pm_plot_risk_max() and pm_plot_tradeoff()
    reproduce the calibration figures of the paper. The first two accept
    any scenario; the third is specific to the risk metric selected.
  - pm_plot_impact() superposes the empirical and theoretical losses
    depending on rho.

Applying the mechanism

  - pm_perturb() applies Y' = Y (1 + rho^n nu + eps) to an aggregated
    table and returns it with rho, the derived keys and the perturbed
    total, plus a pm_meta attribute recording the full configuration.
  - pm_draws() derives the two Gaussian draws from a cell key by SHA-512
    hashing and quantile inversion, so that perturbations are
    reproducible over time.

Metrics

  - assess_loss_expectation() and assess_loss_ci() give the a-priori
    information loss conditional on dominance.
  - assess_risk_I() and assess_risk_II() give the a-priori risk measures
    for the external and internal inference scenarios.
  - assess_risk_diff() gives an a-priori upper bound of the risk measure
    for the differencing scenario.
  - assess_utility_empirical() and assess_risk_empirical() measure what
    was actually achieved on the perturbed table, alongside the
    theoretical values predicted for the same cells.