Entropic.Scree 1.0.2

* Update the paper reference to the all-versions DOI, 10.5281/zenodo.22028086.
* Reframe the package description and help around measurement mapping, Primary
  Informational Rank, the Extended Signal Tail boundary, and spectral-volume and
  informational-gravity accounting, following the October 28, 2026 preprint.
* Return a named measurement_loadings matrix by default: every retained measurement
  on every selected primary informational axis (through K_roots), including zeros.
  Keep the full named eigenvectors separately for inspection and rank updates.
* Enable bipolar loading lists by default. bipolar_top_n = NULL returns all
  positive and negative loadings per primary axis; an explicit fraction/count
  remains available for abbreviated pole summaries only.
* Return raw_eigenvalues, centered_trace, NSMR (ratio), and NSMR_percent
  (percentage units) for quantities already computed. Distinguish raw eigenvalues from floored
  working weights and describe idiosyncratic allocation without attributing it to
  identified sources of error.
* Preserve public function names. Output-name changes are listed below. Core MI, detector, working-weight and primary allocation
  formulas are unchanged on previously supported complete, nondegenerate data.
* Update.Entropic.Scree refreshes the complete primary loading matrix from saved
  eigenvectors. Correct its NULL-argument guard so omitted boundaries are accepted.
* Returning eigenvectors by default increases memory and eigendecomposition work
  compared with the former values-only default. Set both extraction flags to FALSE
  to opt out. Saved version-1.0.1 objects without vectors do not gain loadings via
  Update.Entropic.Scree; it does not rerun their eigendecomposition.

Submission-candidate audit corrections (October 28, 2026)

* Reject categorical state counts exceeding num_bins before native counting;
  native code also rejects invalid encoded states instead of skipping them.
* Missing inputs are rejected by default. categorical_missing = "category"
  explicitly represents categorical missing values as a collision-free state.
  numeric_missing = "median_indicator" fills numeric NA/NaN with its observed
  sample median and adds a 0/1 companion. Infinite values and wholly missing
  numeric columns are rejected. Original input objects are never modified.
* Consolidate identical indicator patterns and retain all associated parent
  names. Filter original measurements first. Indicators bypass numeric QR
  and are mandatory while any parent remains, even below the entropy
  threshold or duplicating another retained column. Required constant
  indicators are invalid. With no retained parent, apply ordinary binary
  constant, duplicate, and entropy filters. Return preparation and
  protection receipts.
* return_processed_data now returns retained ORIGINAL-UNIT data only when
  filtering or opted-in missing handling changed columns/values. Internal
  MI discretization/coding alone does not trigger return. An unchanged input
  gives NULL plus an explicit processed_data_status and message. This changes
  the former internally coded data export; no files are written automatically.
* Recompute both auxiliary FSIG vectors on updates; expose explicit names and
  explicit cutoff/source metadata. Correct the no-macro fallback endpoint to
  the mean-reference count b, rather than b-1. Old aliases are removed.
* Enforce integer ordered ranks in initial confirmation and post-hoc updates.
  Updates no longer silently truncate/clamp invalid selections.
* Reject numerically degenerate centered spectra before reporting resolved
  results; disclose numerical positive axes and floor contribution. Algebraic
  loading exports that include nonpositive directions carry explicit warnings.
* Correct ratio, automatic/confirmed/fallback labels and nominal t-threshold
  claims. Document every returned metric, its units, and preparation behavior.
* Preserve the Appendix A reference Tr(M_c)/m, using the UNFLOORED trace. The
  threshold was not changed to the mean of the floored working weights.

Notation-consistency pass (October 28, 2026)

* Use S_epsilon for the paper's sum of floored working weights and its matching
  mathematical symbol in the help. Reserve m_plus for raw positive spectral mass;
  the two quantities
  are no longer represented by the same internal identifier.
* Align the help equations with the paper's matrix notation, spectral floor,
  P_sig, AIG, FSIG_i, T_i, and raw positive/negative spectral masses. Explain
  the mapping between paper symbols and public R result identifiers.
* Name internal selected/trailing eigenvalue sums as spectral weights rather
  than variances. Use P_sig for the selected spectral-weight fraction.
* Keep numerical results, defaults, and preparation behavior unchanged while
  adopting the current output names listed below.

Breaking output-name changes in 1.0.2

  Earlier name                      Current access
  bipolar_modules$Factor_1 (etc.)    bipolar_modules$Axis_1 (etc.)
  SCDR                              NSMR_percent
  FSIG_extended_bulk                FSIG_auxiliary_bulk
  FSIG_extended_kaiser              FSIG_auxiliary_mean_reference
  top_of_bulk                       auxiliary_diagnostics (see below)

* The old output names are removed, not returned as duplicate aliases. Update
  scripts to use the current names. The auxiliary FSIG mappings also replace
  identifiers used in earlier local 1.0.2 drafts.
* NSMR remains a ratio. NSMR_percent = 100 * NSMR preserves the units formerly
  used by SCDR. The paper's mention of SCDR describes the earlier R output name.
* Axis_1, Axis_2, etc. match measurement_loadings columns. Positive_Anchor and
  Negative_Anchor remain the within-axis pole-list names.
* top_of_bulk formerly mixed two index conventions. Use
  auxiliary_diagnostics$macro_boundary_index for the first index after a
  detected gap (NA if none); macro_endpoint for the inclusive endpoint;
  macro_endpoint_source for its provenance; and bulk_allocation_cutoff for
  the actual auxiliary allocation endpoint. These are not interchangeable.
* Update.Entropic.Scree returns only current output names, including when it
  receives a supported older result. Saved inputs are not modified in place.
* This naming migration does not change the numerical estimation or allocations.

Documentation presentation

* Keep version comparisons and output-name migration guidance in NEWS.
  Function help and source comments describe the current interface, required
  inputs, numerical definitions, and behavior without development history.
