refactor(linking): replace stale fixed-anchor Rust ownership branch

[seonghobae]

Controlled clean-lineage replacement for #663 / issue #662

Protected main is 7b57f77747b9a9bd911660c76b1be0ca0dc2d145; this Draft descends directly from that exact protected head. Draft #663 is 197 protected-main commits stale and its only unique path is the fail-first ownership regression carried here. No checks/reviews from #663 transfer.

Intentional fail-first boundary

Commit 3cefab9decf304a93bb2286969079925208fa004 preserves the exact public-boundary RED: link_fixed_item_parameters() must obtain linked theta/alpha/b and scale/shift evidence from a package-owned _core.link_fixed_item_parameters entrypoint. On current protected-main production behavior the sentinel is expected to receive zero calls, proving fixed-anchor transformation arithmetic remains Python-owned. Setup/import/fixture/changelog failure is not valid RED.

Minimum GREEN after current-lineage RED proof

1. Revalidate the current protected-main fixed-anchor estimand and affine sign/scale convention.
2. Move the current scale/shift plus linked theta/alpha/b transformation arithmetic into mlsirm-core, reusing an existing Rust primitive only if exact equivalence is proven.
3. Expose the thinnest _core.link_fixed_item_parameters PyO3 boundary. Python keeps bounded validation, immutable marshalling, MLSIRMParams construction and evidence transport only; missing/incompatible compiled capability fails closed.
4. Preserve anchor/factor validation, dimensions with no anchors, finite positive slope semantics, xi/zeta/tau, return types and caller immutability.
5. Add direct Rust↔public parity/property evidence plus deterministic known-affine scale/shift and linked-parameter recovery using bias/MAE/RMSE. Correlation alone is insufficient.
6. Keep CPU f64 as reference; benchmark before any GPU work.
7. Add complete rustdoc/docstrings, APA 7 equation/source traceability grounded in Stocking & Lord (1983) and Kolen & Brennan (2014), authoritative changelog material only after GREEN, exact changed-production statement/branch coverage, then one unchanged final head with Python 3.12/3.14, Rust/PyO3, package/reinstall, existing GPU/fuzz, Security Scan, SAST and current-head review evidence.
8. Canonical Architecture/TRD/ADR/UML/ERD/traceability remains solely docs: establish canonical product and architecture baseline #604/docs: close canonical documentation fitness gaps without parallel authority #621. No dependencies/workflows/version/release/hosted-state or unrelated linking methods belong here.

Once this clean-lineage authority is established, #663 should close as superseded with lineage preserved. Advances #662.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: eeebb426-f713-42d5-a67c-5b2ed55c31e6

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[seonghobae]

@opencode-agent address

Exact bounded GREEN implementation handoff for clean-lineage Draft #737 / issue #662. Final pre-write guard is mandatory: refetch source head 3cefab9decf304a93bb2286969079925208fa004, protected main 7b57f77747b9a9bd911660c76b1be0ca0dc2d145, comments/reviews/threads, and exact python/fast_mlsirm/linking.py / Rust linking / PyO3 target blobs; abort without writing if either ref moved or another writer targets this branch/paths. Preserve the fail-first commit/history; no force push, destructive rebase, temporary writer workflow, or competing canonical documentation branch.

Exact current RED is now proven on synthetic integration 85157da86f993e1064ebbc6244ec7de41556eff1. The CPython 3.12 job built the Rust/PyO3 extension, verified Rust as the resolved production backend, then completed the full suite with exactly 1 failed, 3282 passed, 2 skipped. The only failure is tests/test_linking_fixed_anchor_rust_ownership.py::test_fixed_anchor_linking_delegates_transformation_to_rust because the proposed compiled-core sentinel receives zero calls (len(calls)==0). Rust workspace/PyO3, package/wheel/reinstall/release acceptance, enterprise sales-readiness smoke, GPU-no-skip and fuzz are GREEN. The separately tracked marginal _log_sigmoid RuntimeWarning remains #583/#564 and is not owned here. This is valid runtime RED, not setup/import/fixture/changelog failure.

Implement only the minimum root-cause GREEN:

1. Re-read current link_fixed_item_parameters() and preserve its exact fixed-anchor affine convention, anchor/factor validation, dimensions-with-no-anchors behavior, positive finite slope semantics, xi/zeta/tau preservation, return types and caller immutability.
2. Move result-affecting fixed-anchor scale/shift plus linked theta/alpha/b transformation arithmetic into mlsirm-core. Reuse an existing Rust linking primitive only if exact estimand/sign/scale equivalence is demonstrated; do not silently substitute Stocking-Lord/Haebara or another characteristic-curve criterion.
3. Expose the thinnest _core.link_fixed_item_parameters PyO3 contract. Python may validate bounded shapes/types/IDs, marshal immutable contiguous inputs and construct MLSIRMParams/evidence only. Missing/incompatible compiled capability must fail closed rather than retain independent Python production arithmetic.
4. Add Rust unit/property tests and direct public Python↔Rust field parity over ordinary one- and multidimensional cases, multiple anchors, dimensions with no anchors, duplicate/out-of-range anchors, malformed factor IDs, extreme-but-finite slopes, non-finite inputs, immutability and stable non-reflective errors.
5. Add deterministic known-affine linking recovery: recover transformation coefficients and linked item/person parameters with aligned bias/MAE/RMSE. Correlation alone is insufficient.
6. Keep CPU f64 as reference and benchmark before any GPU proposal; this linking kernel is small.
7. Ground doctoring/equation traceability in Stocking & Lord (1983), Kolen & Brennan (2014), and any primary source actually governing this fixed-anchor transform. Distinguish simple fixed-anchor affine linking from characteristic-curve methods.
8. After focused GREEN, add authoritative changelog material and complete rustdoc/docstrings, then run focused linking ownership/robustness/recovery tests, full Python 3.12/3.14, Rust/PyO3, package/reinstall, existing GPU/fuzz, Security Scan, SAST, changelog check and git diff --check on one unchanged head. Keep Draft for fresh current-head review/governance.
9. Canonical Architecture/TRD/ADR/UML/ERD/traceability belongs only in docs: establish canonical product and architecture baseline #604/docs: close canonical documentation fitness gaps without parallel authority #621 after implementation stabilizes. Do not touch dependencies/workflows/version/release/hosted state or unrelated linking methods.

Stop writing immediately if source/base/writer identity changes.

[seonghobae]

Superseded by clean current-main Draft #752. Fresh compare proved #737 head 3cefab9decf304a93bb2286969079925208fa004 is 30 protected-main commits behind and its only unique path is tests/test_linking_fixed_anchor_rust_ownership.py. The prior OpenCode handoff was explicitly conditional on protected main remaining 7b57f777...; that condition is false, so the write lease is stale. #752 recreates the exact ownership RED from protected main 8a56d7b1b024e18a5a96a94db4ede1bf6648ead1 on head ac6c200fea7e0cb47a870e020faf753a61744b33. No predecessor checks/reviews/approval transfer.