Destiny v2.2.0

[2.2.0] - 2026-08-26

This section describes the 2.2.0 source candidate. It becomes a published
GitHub Release only after the matching v2.2.0 tag workflow succeeds.

Added

• Added an opt-in incremental research path with strict long-form point-in-time
  controls and hashed controls catalogs.
• Added paired ridge models on identical rows and folds: M0 uses all declared
  controls, while M2 adds every preregistered astronomy feature under the same
  alpha grid and fit budget.
• Added Newey-West HAC inference, circular moving-block bootstrap, and a full
  M2-pipeline joint circular-shift placebo over the declared cyclic group.
• Added conservative union-null max-p aggregation and Bonferroni family
  adjustment, with explicit limits on what the local runner can certify.
• Added validator-clean controls/hypothesis templates and a complete synthetic
  end-to-end incremental demo generator.

Dependencies

• NumPy is now a direct, exact-pinned research/all dependency because the
  incremental contract directly relies on float64 arrays and numpy.linalg.
  Python-specific pins keep resolution explicit; the core install is unchanged.

Compatibility and evidence boundary

• With incremental_model absent or exactly { "enabled": false }, the legacy
  univariate path and result schema remain 1.0.0.
• Enabled incremental hypotheses require hypothesis schema 2.0.0, both
  controls inputs, and emit research result schema 2.0.0. Core result schema
  remains 2.0.0; astronomy schema remains 1.0.0.
• A positive loss differential or adjusted threshold diagnostic does not change
  the evidence label. This release never emits CONFIRMATORY_PASS and does not
  claim causality, a complete financial baseline, stable alpha, or live-trading
  readiness.

Destiny v2.1.0

2.1.0 - 2026-08-26

Changed

• Split the command-line, provider, output, ephemeris, feature, contract,
  statistics, registry, and research-runner boundaries into focused modules.
  Existing commands, exit codes, and JSON behavior remain compatible.
• Added health profiles: doctor --core, doctor --research, and
  doctor --all; plain doctor remains equivalent to --all.
• Added bootstrap.py --installer auto|uv|pip. Auto mode prefers a valid frozen
  uv lock and otherwise explains why it falls back to pip.
• Added explicit --allow-private-index-env support. Installer children use an
  exact environment allowlist by default; the opt-in forwards only enumerated
  pip/uv private-index environment variables and named username/password
  credentials. User-level pip/uv config-file discovery is unchanged.
• Narrow doctor profiles now leave Node/iztro not_checked; default/all performs
  strict pinned-version metadata validation and a bounded static require smoke.
• Added a tag-driven release path with an exact-pinned build backend, two
  isolated byte-identical wheel/sdist builds, SHA256SUMS, and GitHub/Sigstore
  build provenance.
• Expanded the regression baseline to 122+ collected tests across the
  deterministic engines, agent-safety boundaries, and historical research.

Compatibility and evidence boundary

• The distribution and Skill version is now 2.1.0. The main result JSON
  schema remains 2.0.0; astronomy and research schemas remain 1.0.0.
• Historical research remains a preregistered univariate association MVP.
  This release does not claim financial-control-adjusted incremental evidence,
  a paired M0-vs-M2 result, causal efficacy, or production-trading readiness.