[OVVO-Financial]

Purpose — R 14.0 parity integration on the complete Python port stack

This is an audit/staging PR, not yet merge-ready. It tests the cumulative implementation branches from #148 through #189 against the fresh R 14.0 reference cache produced by #190. It intentionally adds no Python behavior changes.

Provenance

• Statistical oracle: OVVO-Financial/NNS@7b07a837c130ba7220dbe16c584221508757d5ae (NNS 14.0).
• Generated fixture source: PR Regenerate R parity cache for NNS 14.0 #190, workflow run 42248672336, installed exact upstream R 14.0 source archive.
• Applied all 59 generated tests/_r_cache/*.json shard blobs unchanged from Regenerate R parity cache for NNS 14.0 #190, including missing entries caused by unsuccessful live-R calls (do not silently carry over 13.3 or invent values).
• Kept the current cumulative Python tests/_r.py helper, changing only its NNS version marker to 14.0; copied Regenerate R parity cache for NNS 14.0 #190's older helper would discard subsequent tests and wrappers.
• Preserved accepted vendored native snapshot NNS-core@a0d3b93b6ac451bd38e5e9375ef06e6d0d314b9d rather than regressing to Regenerate R parity cache for NNS 14.0 #190/main's older core commit; updated only the R oracle metadata in sync/nns_source.json.
• Legacy tools/NNS_13.0.tar.gz and r_src_tree_hash remain separate and must not be represented as R 14.0 source.

Existing pre-integration signal

• Tip of partial moments: co-moments reject empty input and recycled targets #189: all four Python 3.11–3.14 Native backend CI checks green on R 13.3.
• Regenerate R parity cache for NNS 14.0 #190 (cache against old main): 461 failed / 1,279 passed / 12 skipped in cached parity, with failed live-R regeneration and missing reference keys. Its generated reference blobs are useful; the failures do not establish R 14.0 parity.

Required gates before any merge

1. Measure NNS_R_CACHE_ONLY=1 python -m pytest -q tests/parity on this combined tree across Python versions. Publish counts by function and distinguish genuine numeric/schema mismatches, missing cache entries, R exceptions, and stochastic differences.
2. Re-run fresh live R 14.0 regeneration using this combined tree (not main). Track any R exceptions and fill every reachable deterministic cache entry before considering a green parity gate.
3. For remaining discrepancies, fix the Python implementation or accepted NNS-core snapshot with explicit R examples/tests; never weaken tolerances, fabricate cached outputs, or change R 14.0 truth to force green.
4. Reconcile this with Regenerate R parity cache for NNS 14.0 #190 before merging Fix the Boston stack parity tests, the curve-fitting gist and CI tool pins #148–partial moments: co-moments reject empty input and recycled targets #189; avoid two competing cache baselines. Keep this PR draft until all parity, invariants, native-build, and formatting/test gates pass.

This staging branch isolates the baseline-fidelity question from the 41 unmerged implementation slices.

[OVVO-Financial]

Initial full-stack/R14 baseline verdict (2026-11-06)

The first combined Native backend CI on the complete 41-PR Python stack + #190's R14 fixtures (run 42254499440) failed on all four Python versions at the parity step (invariants passed). Python 3.11's summary: 1,934 failed / 175 passed / 12 skipped. Importantly, all 1,934 failures are RuntimeError: R cache miss across 35 test modules; this is an oracle-coverage mismatch, not evidence of 1,934 wrong Python numeric outputs. The cache in #190 was regenerated on the older Python main test matrix, whereas #189 adds many parity calls and a newer helper.

Corrective action in progress

The existing Regenerate R parity cache workflow has been updated on this draft audit branch only to trigger exactly once on its workflow YAML push, using pinned R NNS 14.0 commit 7b07a837c130ba7220dbe16c584221508757d5ae. Its triggered run is 42255158508. It installs the exact R 14.0 tarball, regenerates the entire full-stack parity cache from empty, runs offline verification, and proposes the resulting cache on a separate PR targeting this branch, not main; it emits an explicit failed workflow result if regeneration or verification fails (while preserving artifacts and PR for analysis).

Native-source dependency

The Python stack currently vendors NNS-core@a0d3b93b6ac451bd38e5e9375ef06e6d0d314b9d (core PR #34), but NNS-core also has an unmerged follow-up stack through #40. Do not present the core snapshot as verified against R 14.0 until that upstream native layer has been reconciled independently.

Keep #190 and this audit draft unmerged until missing references are resolved and live/cache R14 parity genuinely passes. Do not replace numeric discrepancies with relaxed tolerances or fabricated fixtures.

[OVVO-Financial]

Integration notice: This is the R14 audit layer atop the 41 stacked implementation PRs. Its six commits contain no Python source changes. Current implementation fixes and the fully regenerated R14 parity cache reside on draft #192, based on this PR. Do not merge this cache-only audit or retarget it to main ahead of strict R14 parity; do not reset #192 from this branch because that would discard code fixes. Full dependency and function ownership ledger: docs/R14_PARITY_INTEGRATION_GATE.md on #192.