docs: isolate LaRT owner repair and verification handoff

[seonghobae]

Stacked on #1107. Current head 536dc4f; base autoresearch/20260909-kpi-loop at de21ffd. Normal merge preserves the parent citation reconciliation and both LaRT handoff paragraphs. No history rewrite.

Effective successor delta remains four documentation files: AGENTS +4, CLAUDE +4, doctoring +162, Gap +9, zero deleted lines. All added lines match the previous 14a6a94-to-8f71ff08 delta exactly; the LaRT runbook is byte-identical. Patch IDs differ because parent context changed, not because successor additions were discarded. Runtime, tests and Rust are identical to the parent.

At current head all six citation contracts passed in 1.29s. This is focused source evidence, not full-suite or installed-package acceptance. Previous hosted results do not validate either new head; independent current-head approval remains required.

Upstream repair: Toby-X/Latency-Response-Theory-Model#2 at e5c82a91918a26fab469efd9f04d5152088b5c84. Historical four local loading/contract tests passed; full module, estimator, statistical effects and release remain unverified. Last observed upstream run 34692344683 required maintainer approval; its current state is not reverified here.

Visual evidence: parent citation-change sections in all five documents were directly inspected in a real local browser at 1265 x 712, English/default, de21ffd. No clipping or overlap observed. Historical GitHub diff inspection at 8f71ff0 remains historical; the newly restacked GitHub body/diff inspection is pending. This is not whole-document, mobile, multi-locale, product UI, estimator or KPI acceptance.

Draft. No PR was closed, no protected merge, release or deployment occurred, and no KPI gain is claimed.

[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.}

[seonghobae]

Visual verification at exact head 0024522: directly inspected GitHub rendered AGENTS.md Autonomous research handoff, CLAUDE.md Read AGENTS.md first, and the complete Gap retrospective-calibration section through the final preprocessing paragraph. Chromium 1265 x 712, English/default, file tree collapsed. Changed guidance, headings, links and paragraphs were legible without observed clipping, overlap or horizontal overflow. The Gap initially displayed a loading spinner; inspection was performed only after content rendered. Doctoring receipt text at this same head was directly inspected in the preceding turn. These bounded views are not whole-document, mobile, other-locale or product-UI acceptance. No predecessor evidence was removed and no estimator/KPI/hosted-approval claim is made.

[seonghobae]

Pinned data-permission audit

At upstream 8cb9639eb162ff3732df82d4e190e7f902bde19d, the full recursive
Git tree returned truncated=false. License/notice/terms-name inspection found
only root LICENSE, data/README.md and data_generation/README.md.

The root README
explicitly describes the code as MIT-licensed. The
LICENSE
uses standard software/documentation wording. The fully read
data README
describes post-processed matrices and absent raw generations, but supplies no
separate data-use grant. This bounded inspection does not establish that no
permission exists elsewhere, or a legal prohibition on use. It does establish
that code MIT status alone is insufficient evidence for our data acceptance gate.

Keep data permission/provenance unresolved in this successor. Do not label the
matrices forbidden, licensed customer outcomes, or a complete failed-request
cohort. A dataset-specific grant and upstream benchmark provenance still need
verification before redistribution/adoption. No upstream code was executed,
no matrices were copied, and no estimator or production defaults changed.