feat(analysis): bind Rubin loading uncertainty to an analysis-run profile

[seonghobae]

Consolidation status

rubin_loading_uncertainty_v1 / tepp.rubin_loading_uncertainty.v1 is a Draft Analysis Run composition over protected-main psychometric_core::{recover_loading_point_estimate_mean, combine_draw_level_ols_loadings}. It is not a second estimator, Mislevy person-level plausible-value implementation, ESEM/DSEM sampler, CWC profile, or independent protected-main landing authority.

Current #374 head is 7fb76d1d5b338ecfb652a11a136c984c092ada15 on protected base main@a243f18da4a4ca8a8d068c39922537f1f8ed6ad0. The surviving Analysis Run vehicle is #416@03f8de2ed0a0fb842d2022d411814e440df7cfb4. Preserve every valid source/test/ADR/doctoring/scientific-evidence delta through ordinary non-force conflict resolution. Simple Close is not consolidation.

Scientific / temporal repair

The branch repairs real predecessor defects: per-row immutable snapshot_id + AvailableTime; instant-based KnowledgeCutoff binding; cross-snapshot fail-closed and same-snapshot future censoring; 256-draw and 1,000,000 admitted-cell application resource envelopes; reachable artifact count validation; exact binary64 Rubin-total integrity; symmetric 256 KiB artifact wire admission; protected-main robust point-estimate invocation; fallible digest propagation; and terminal validation_status = "validated" separated from artifact inference.

Repair lineage remains 152cb391... RED → dd46f8eb... causal repair → 8f0e7433... contract migration → ADR 0034 Proposed at a4a0ab01... → doctoring b3094612... → dependency cleanup 7fb76d1d.... Historical replay requires same-snapshot rows with AvailableTime > KnowledgeCutoff to leave the earlier scientific matrix/result unchanged; cross-snapshot rows are provenance violations, not censored history. The 256-draw and one-million-cell ceilings are resource limits, not psychometric recommendations.

Scientific acceptance and activation gaps

Issue #503 owns recovery/coverage acceptance. Draft child #504 is now exact 3d153b382e2382efb532f49bb1e2a8b93be0b73f. It provides eight 512-replication known-truth scenarios and a separate 512-replication leakage-safe rolling-origin study, with explicit attempted/recovered/failed accounting, robust point-estimate bias/RMSE and Monte Carlo uncertainty, Rubin Ubar/B/T, and a scoped large-sample normal coverage diagnostic centered on Qbar.

#504 also repaired its own provenance semantics. RED 16e3c151d6cadd79c1ec8e32d74a8401333be42c demonstrates that different generated Monte Carlo populations were initially aliased to one immutable snapshot/run identity. Repair ed2ac416ac42cb15909cefde0581af408de652e7 gives every generated population a distinct snapshot and every executor call a distinct run/idempotency identity. Rolling-origin views share one replication-specific snapshot because they are historical views of one population, but use distinct run receipts. CodeRabbit's valid scenario-description finding was repaired at ba90324ccb072ae24dcc1f7f54cbc579d219fe4e: the 8 scenarios are n × m × paired(lambda,sigma), not a 16-cell four-factor cross.

Issue #505 owns a separate Validation Evidence / Projection-policy gap. The product can currently combine arbitrary finite complete-data draw matrices after temporal/resource admission but does not bind a versioned draw-generator/imputer contract or validation-evidence identity for the generator/analysis pairing. Correct Qbar/Ubar/B/T arithmetic therefore must not be promoted into generally validated inferential semantics for arbitrary supplied draws. #504's ADR 0034 at 3d153b... records this fail-closed activation boundary under ADR 0014 and scopes the checked Monte Carlo evidence to its declared generator/design.

Do not describe #374/#504 as scientifically accepted or release-ready while #503 lacks exact-head completion, and do not treat #503/#504 as closing #505. LLM judgment and synthetic unit fixtures cannot substitute for claim-specific validation evidence.

Documentation / overlap authority

ADR 0034 remains Proposed; the newer scientific-evidence child carries current #503/#505 and replicate-identity wording. Shared docs/adr/README.md, docs/TRACEABILITY.md, docs/product-technical-gap-baseline.md, and CHANGELOG current-state reconciliation remain #435 / surviving-lane single-writer work. This branch also contains unrelated psychometric_core/src/error.rs MANIFESTVARstd Display-coverage overlap independently owned by #353; exclude that overlap from this scientific fold rather than copying it.

Review / fold gate

Live ruleset 18156473 requires one qualifying current-head approval, stale-review dismissal after pushes, all review threads resolved, central required workflows, and non-fast-forward protection. Checks/reviews from #374 do not transfer to #504 or the final fold head.

Before successor-based Close, verify on #416/successor exact head that the temporal provenance, resource bounds, Rubin-total integrity, robust point-estimate owner call, provider/domain status separation, bounded wire/digest path, #503 scientific evidence including per-replication snapshot/run identity, #505 activation boundary, ADR/doctoring, TRACEABILITY, and every other valid delta survived ordinary conflict resolution. Then reacquire exact-head format/lint/test/rustdoc, authored line+branch coverage, security/SAST/CodeQL/documentation gates, review resolution, and qualifying independent approval.

No self-approval, force push, destructive rebase, skip/xfail, coverage-denominator manipulation, scanner suppression, no-op rerun, mutable dependency consumption, provider-routing copy, or predecessor evidence transfer is authorized.

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

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.}

[devin-ai-integration]

Note

This report is out of date. Scroll down for Devin Review's latest report on this PR.

Devin Review found 3 potential issues.

[devin-ai-integration]

Devin Review found 5 new potential issues.

[seonghobae]

Restack on protected main (a243f18)

Non-force merge of origin/main (merge commit ec2bf9c4); the only conflict was the CHANGELOG.md append, both entries kept. ADR 0034 already carries an admitted maturity value (active-PR).

Local evidence on the pushed head (toolchain 1.98.0): cargo test -p analysis_engine -p psychometric_core 403 passed / 0 failed, clippy -D warnings clean on both crates, cargo fmt --all --check clean, documentation/workspace/docstring contracts PASS, git diff --check clean.

🤖 Generated with Claude Code

[seonghobae]

Scientific acceptance child #504 now carries issue #503 candidate evidence on exact head ac324d74911f15c840a5c1d1c419ef192ea7d0fa, stacked directly on this PR's 7fb76d1d5b338ecfb652a11a136c984c092ada15 head.

The child adds 8 × 512 repeated-sampling known-truth scenarios plus a 512-replication rolling-origin replay contract. It consumes validation_core for bias/RMSE/coverage/Monte-Carlo summaries, keeps robust point_estimate_mean as the recovery target, and centers the diagnostic interval on Rubin Qbar with T. #503 remains open until exact-head checks and review are GREEN.

Do not close #374 as superseded unless the surviving #416/fold head is verified to inherit both this PR's temporal/resource/artifact repairs and #504's scientific evidence/test/docs. Predecessor check receipts do not transfer.