feat(longitudinal): consolidate lagged correlation and discreteDRIFTstd - #310
feat(longitudinal): consolidate lagged correlation and discreteDRIFTstd#310seonghobae wants to merge 299 commits into
Conversation
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthrough
Changes종단 모델링 계약
Priority: ⬇️ Low Estimated code review effort: 4 (Complex) | ~60 minutes Change: Feature Sequence Diagram(s)sequenceDiagram
participant Caller
participant longitudinal_core
participant EventTimeInterval
participant NumericalRecovery
participant LongitudinalError
Caller->>longitudinal_core: 시간 관측과 수치 입력 전달
longitudinal_core->>EventTimeInterval: 이벤트 간격 검증
EventTimeInterval-->>longitudinal_core: 승인된 typed interval 반환
longitudinal_core->>NumericalRecovery: CWC, occasion-mean 또는 표준화 복구 실행
NumericalRecovery-->>longitudinal_core: 추정량 또는 검증 오류 반환
longitudinal_core-->>Caller: 결과 또는 LongitudinalError 반환
Merge Risk: 🟡 Moderate · up to The current head can return an incorrect mixed-sign mean and no longer schedules the required hourly workflow. Current-head validation evidence must also be refreshed, so these issues should be resolved before merge. 🚥 Pre-merge checks | ✅ 3 | ❌ 2❌ Failed checks (2 warnings)
✅ Passed checks (3 passed)
Full details: Linked Issues checkExplanation
Full details: Out of Scope Changes checkExplanation
✨ Finishing Touches 💡 1🛠️ Fix failing CI checks 💡
📝 Generate docstrings
🧪 Generate unit tests (beta)
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. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
This comment was marked as resolved.
This comment was marked as resolved.
Sorry, something went wrong.
|
@opencode-agent review | Independent exact-current-head formal review requested for |
|
Current-head authority update: ordinary-forward review repair is now |
|
Terminal current-head evidence for |
|
Fresh owner-path authority refresh (2026-09-12): central CodeQL bootstrap Current-head CodeQL run An explicit This central-control movement does not change #495 scientific acceptance or the TEPP numerical-owner RED. Keep |
|
Tip For best results, initiate chat on the files or code changes.
The live Draft state of The queued protected-handler run does not provide terminal handler or SARIF evidence. The current CodeQL consumer failures remain a producer-after-consumer settlement issue. The later GREEN dispatch producer does not change those failed consumer results. The unavailable CodeRabbit approval path does not satisfy the independent-approval requirement. PR ✏️ Learnings added
🧠 Learnings usedYou are interacting with an AI system. |
|
Fresh central-gate authority correction for this unchanged TEPP head
Fresh same-head Required CodeQL #2106 formal reviews remain COMMENTED only and |
|
Central-gate authority refresh for current
The central adapter currently preserves provider-body redaction but exposes only byte counts, so I filed Therefore #310's #495 estimand, Monte Carlo acceptance, and sole mixed-sign half-ULP product RED are unchanged. Do not rerun/push this TEPP head merely for review freshness, do not consume mutable fast-mlsirm, and do not mark Ready until the existing causal landing bar closes. |
|
Fresh external-owner authority for unchanged TEPP exact head
No TEPP estimand or arithmetic changes follow from these owner-state updates. |
|
Owner-path currentization (2026-09-13 KST): reusable numerical/GPU prerequisite This does not change the scientific/numerical dependency boundary. #1717 remains Draft/open/mergeable and the real GPU RED remains the per-shader-stage SwiftShader limit 10 versus the marginal E-step/score topology using 17/18 storage buffers. No predecessor GPU/CI GREEN transfers to
|
|
Owner-state handoff (supersedes only the stale fast-mlsirm head/base lines in the PR body; #495 estimand/scientific acceptance is unchanged):
Therefore |
|
Owner-path refresh (2026-09-13 KST): Longitudinal scientific authority on this PR is unchanged; do not alter the #495 estimand/denominator/Monte Carlo contract. Reusable numerical prerequisites have moved. |
|
Fresh numerical-owner handoff: |
|
Fresh exact-head gate follow-up for the numerical prerequisite: |
|
Superseding the earlier queued-handler note: exact protected handler |
|
Fresh owner-state handoff superseding the stale fast-mlsirm heads in this PR body (scientific estimand/acceptance is unchanged):
Therefore |
|
Correction to the fresh #1717 handoff above: the branch pins Accordingly, the proposed packing direction is only a candidate until the owner records the controlled adapter's combined limit: E-step target 10 storage + 1 uniform requires combined >=11; score target 8 storage + 1 uniform requires combined >=9. The owner must also validate packed arena binding size/buffer size/alignment. Owner correction: fast-mlsirm #1717 comment This does not change TEPP's scientific contract: |
|
Fresh numerical-owner handoff for this exact scientific head
The local mixed-sign half-ULP RED therefore remains the correct consumer-side blocker. Do not weaken it, skip it, or reinterpret the scientific target. |
|
Owner-state handoff, current exact heads: reusable numerical prerequisite |
|
Fresh numerical-owner handoff: |
|
Fresh numerical-owner authority supersedes the older #1717 head references in this PR body. On that exact head, controlled #1717 is still not merge-authorized: Required CodeQL PR
|
|
Numerical-owner handoff refresh: fast-mlsirm #1717 is now |
|
Owner-path refresh only; no TEPP numerical/scientific contract change. fast-mlsirm protected
Therefore |
1 participant
Longitudinal Modeling landing vehicle
This Draft is TEPP's Longitudinal Modeling landing vehicle for typed event-time lagged association, stationary/discrete drift-diffusion standardisation, within/between decomposition, CWC and occasion-mean irregular residual composition, explicit irregular-rate estimand identity, known-truth recovery, and the DDD boundary that keeps reusable domain-neutral numerical arithmetic out of TEPP ownership.
Exact authority
ba10820e0d28cc33d1b91ef37f6f6d163b3d91e9;main@a243f18da4a4ca8a8d068c39922537f1f8ed6ad0;ContextualWisdomLab/fast-mlsirm#1814, implementation PR #1816;ContextualWisdomLab/fast-mlsirm#1717;ContextualWisdomLab/.github#810;.github#2106, then producer/consumer successor.github#2040.Mergeability is not scientific, review, security, or release acceptance. Predecessor checks and reviews are diagnostic lineage only and never transfer across a head change.
#495 scientific contract
tepp.irregular_rate.lag_pair_average.v1/IrregularRateEstimand::LagPairAverageV1is the implemented pair-weighted target.tepp.irregular_rate.unit_average.v1/UnitAverageV1remains typed but intentionally fail-closed until the reusable immutable finite-mean dependency exists.IrregularRateSummarypreserves candidate/contributing units, candidate/admitted/refused pairs, and refusal populations; follow-up multiplicity is not silently reinterpreted as a design or multiple-membership weight.The 4,096-replicate informative-missingness contract records attempted/recovered/failed replicates, truth conditioned on actually admitted lag-pair counts, bias with Monte Carlo uncertainty, RMSE/stability, interval coverage with uncertainty, and exact replay. These are scientific acceptance oracles, not hidden production implementation of
UnitAverageV1.The current review-repair sequence does not change those estimands, formulas, refusal policy, or Monte Carlo thresholds. The redundant private irregular-rate facade was removed so crate-public composition points directly at canonical
irregular_residual.Current TEPP exact-head evidence
Rust Foundation
34672270380remains terminal RED onba10820...:mixed_sign_mean_rounding_contract::half_ulp_tail_changes_the_final_mixed_sign_rounding, actual bits5080060379673919488, correctly-rounded expected5080060379673919487;Documentation Quality
34672270392and SAST Semgrep34672270367are GREEN. Existing submitted reviews remain COMMENTED; there is no qualifying independent current-headAPPROVEDreview.Numerical-owner boundary and fresh owner state
The mixed-sign half-ULP contract proves TEPP's local
scaled_compensated_meanis not acceptable as reusable finite-binary64 authority. TEPP will not add another summation kernel, relax the assertion, skip/xfail it, rewrite coverage, or pin mutable fast-mlsirm source.ebffb8c5584d1ce971432e259bca083057dde351onmain@16d2f35cd5608eb6728e41128a0eb7d6b960d081, open/Ready/mergeable/unmerged. The restack is an ordinary two-parent merge; predecessor GPU/security/Python receipts do not transfer. Fresh repository CodeQL is GREEN while CI, Semgrep, fuzz, Security, and delegated CodeQL continue settling. The controlled GPU repair remains E-step 8 storage + 1 uniform and score 3 storage + 1 uniform; current-head real-pipeline parity must be reacquired before landing.1138ffdd27f8a9d9334f2be436f76592302058acon the same protectedmain@16d2f35..., open/Draft/mergeable/unmerged. The candidatefast_mlsirm.binary64_mean@1.0.0remains mutable and therefore non-consumable by TEPP. Its six-file binary64 delta is preserved while the protected-main advance was adopted by ordinary two-parent merge; alld97decc...receipts are predecessor evidence. CI/fuzz are skipped under Draft admission and the fresh security/CodeQL lanes are reacquiring.v0.9.1, immutable but predating the finite-mean contract. TEPP therefore has no released ACL to consume yet.Central gate evidence
Security
34672270381on this TEPP head remains terminal RED only in dependency-review support; exact checkout/head verification succeeds while the support check fails before the pinned Dependency Review step. Scorecard, Trivy, and OSV are GREEN but do not replace authoritative base/head dependency comparison. That owner remains.github#810.Required CodeQL
34672270345reproduced the producer-after-consumer settlement defect on this TEPP head. Canonical handler owner.github#2106is now at611ccd73460ab0188e0085956ade6180bb28a91a, open/Ready/mergeable on protected.github/main@64f483db9d052322c65bcdf1675d66138156f306. Its current-head Python Security, Security Scan, runtime-quality and Semgrep lanes are GREEN; CodeQL PR34742112070is RED because both compatibility consumers successfully read a verdict and then fail enforcement, while the dispatch producer starts later and succeeds. This remains executable evidence of lifecycle ordering/settlement, not a leaf source/SARIF finding. TEPP will not add fixed sleeps, blind/manual reruns, no-op pushes, fabricated receipts, or provider/model routing copies.Documentation ownership
docs/product-technical-gap-baseline.mdanddocs/TRACEABILITY.mdremain single-writer-owned by Draft #435. Comment handoff is not checked-in source currentization. That lane must replace predecessor #1717/#1816/.github heads with the current authorities above while preserving the #495 estimand identity and evidence trace.Landing bar
Keep this PR Draft until the causal chain closes in order:
No self-approval, bypass merge, force push, destructive rebase, gate weakening, source-rewriting coverage trick, mutable dependency, skip/xfail, CPU-as-GPU substitution, provider/model hard-code, paid fallback, or purpose-complete self-modifying workflow is authorized.