mars

Parity Audit Gap Classification

This note classifies the current repo gap matrix using the parity-audit rubric and the matrices already recorded in docs/.

In this note, nice-to-have corresponds to the rubric’s optional bucket, and upstream-only or intentionally out of scope corresponds to the rubric’s out of scope or intentional bucket.

Source Basis

• Parity Audit Rubric
• mars / earth repo gap matrix
• Consolidated Parity Table
• Parity Audit Recommendations

Rows already marked compatible in the repo gap matrix are treated as aligned baseline, not gaps, and are not repeated below.

Bucket	Matrix rows	Source-backed reading
parity-critical	Example outputs and documented claims; Serialization and multioutput edge cases	The rubric treats canonical examples, serialization semantics, and docs claims that could mislead users as parity-critical (docs/parity_audit_rubric.md:10-20, docs/parity_audit_rubric.md:77-78). The repo gap matrix still says example-output parity is not fixture-locked and that pickle / multioutput remain expected failures (docs/parity_audit_repo_gap_matrix.md:36,40). The consolidated parity table already mirrors that split (docs/parity_audit_parity_table.md:28-29).
nice-to-have	Tie handling	The repo already has deterministic regression fixtures, but the current matrix does not document a user-facing tie policy (docs/parity_audit_repo_gap_matrix.md:35,52-57). That is audit-visible evidence hardening, not a confirmed behavioral break. The parity table keeps it as a partial contract gap rather than a defect (docs/parity_audit_parity_table.md:26).
upstream-only or intentionally out of scope	Warning behavior; Formula / interface ergonomics; Packaging / versioning / release behavior	These rows are already documented as deliberate scope choices: narrower warning policy, estimator-centric API surface, and Rust-first multi-registry packaging (docs/parity_audit_repo_gap_matrix.md:33,39,41, docs/parity_audit_rubric.md:34-39, docs/parity_audit_parity_table.md:24,30).

Track Implications

• Serialization and multioutput remain separate implementation-track candidates.
• Tie handling should stay in the evidence queue until the upstream contract is pinned down more explicitly.
• The upstream-only or intentionally out of scope boundaries should remain documented so they are not reclassified as missing parity.
• The current bucket list is limited to rows already present in the repo gap matrix; it does not exhaust the R earth parity-critical surface. R-side plots, prediction intervals / variance models, GLM-style extensions, and update workflows remain upstream requirements that are not yet represented as closed repo parity items.

This site is open source. Improve this page.