Prepare delivery: map repair, clear media and branch navigation

[p20030920p]

The submission explains SLAM and the two selected directions before deriving the candidate map-repair question. Nine previews now show the existing measured outputs clearly, and branch navigation supports the requested delivery cleanup.

• Keep the hero, original MP4s, historical GIFs, scores, adverse results and candidate H1 unchanged. Nine new GIFs use hash-verified MP4s, native widths and 256 colors; the ConceptGraphs RViz preview includes chronological excerpts from all nine stages.
• Preserve the dependency audits, source records, bilingual scopes and the distinction between supplied-pose mapping, saved-map viewing and complete SLAM/navigation. No new inference or research result is introduced.
• Keep three public branch roles: main delivery, original-code reproductions, and combined personal learning. Hardware guides now live under physical/ in the learning branch; room2 evidence uses its immutable commit link.

Current-head self-review: 0b775d6, based on main dfd4a05. Relative to the previously reviewed f42bfc0, only six navigation documents change. Runtime code, tests, dependencies, annotations, results, media and the hero are byte-identical. This supersedes earlier-head reviews for merge purposes and is an AI self-review, not independent scientific review.

Validation: bilingual documents, local links, media provenance, changed-document links and whitespace checks pass on the prepared trees. Personal consolidation preserves both source histories; all 158 hardware files remain, with 156 byte-identical and only two README navigation updates. The room2 source commit remains publicly reachable through the learning merge. Current-head Windows/Ubuntu CI passed for push and PR. Personal-learning CI also passed. Author-method heavy jobs were skipped.

中文：保留现有研究边界、反证与候选 H1，发布清晰预览，并将分支入口统一到交付、作者复现、个人学习三类。此次新增修改仅为六份导航文档，个人学习已合并实物目录；旧实验使用固定提交追溯。没有运行新实验。

Keep draft until explicit owner approval. No auto-merge. Deleting the old branches must follow this navigation update so the delivery homepage keeps working.

[p20030920p]

Self-review / 自审

Reviewed head: 87603db88004f53b3a6b557aaf8519f1dcb28a17
Compared base: dfd4a05f02352d32b4f0a2bf522e189708c9e2b2

Item / 项目	Result / 结论
Dependency audit / 依赖	Baseline: 3 transitive-Pillow reports, 4 subprocess/viewer dependencies, 2 development-tool reports. Pillow declared; dev groups classified; four narrowly documented DEP002 exceptions. Final deptry: zero reports.
Boundaries / 边界	All three contracts pass. Disposable source copies containing a cycle, reversed layer and Torch import were each rejected. No runtime source changes or module moves.
Dead code / 冗余	Vulture at 100% confidence passes. Lower-confidence framework/config/interface candidates retained; no evidence-supported deletion found.
Locks / 锁定版本	No existing locked package version changed. Nine new packages belong to the opt-in audit tools and their dependencies. requirements.lock changes only Pillow's direct-dependency attribution.
Tests / 测试	44 passed; Ruff passed. Both synthetic CI smoke runs and report generation passed.
Evidence / 证据	36 portable records, 97 paired cells, 35 delayed cells and six post-hoc controls verified. Research protocols, results, media, annotations and source snapshots have no diff.
Docs / 双语	README changes are one navigation link per language. Existing reproduction guides each gain 26 lines; commands and scope match. Bilingual/local-link/media checks passed.
CI	Windows and Ubuntu core jobs pass on push and PR. Author-method jobs are intentionally skipped; no CUDA rerun claimed. Push · PR.
Limits / 限制	deptry scans only packaged src; separate CUDA dependency snapshots remain outside scope. Static tools do not establish runtime correctness or completeness of dead-code detection. IWYU was not run; C++ build targets are upstream-owned.

Merge status / 合并状态: Owner approval required. No auto-merge. Main remains clean and unchanged. This self-review covers only the head above; later changes require another review.

自审未发现阻塞问题；这不是独立人工审核，也不替代所有者的明确合并批准。

[p20030920p]

Updated self-review / 本轮自审

Head: c43b3b8156bac4f2bac66b18f6b64b054624851b
Main base: dfd4a05f02352d32b4f0a2bf522e189708c9e2b2

This replaces the scope conclusion of the earlier self-review for 87603db; it covers the final cumulative diff, including the dependency checks.

Review item / 项目	Conclusion / 结论
Submission narrative / 主线	One question: late pose correction, association recovery and equal candidate caps. Four related reproductions lead to counterevidence, candidate H1 and a frozen rejection rule. Why ConceptGraphs is first is explicit; no common dominant defect is asserted.
Scientific scope / 范围	Supplied-pose mapping is separate from complete SLAM/navigation. Equal caps are not memory matching. Oracle does not test bounded replay. AI-only annotations cannot confirm H1; room2 analysis remains pending.
Metrics / 指标	Original author records at 535a278 support 141-scan DUFOMap Table IV, 91-scan historical BeautyMap Table III, CG room0 scores and completed HOV-SG 20-frame scores. Old GIF subsets, native semantic protocols and incomplete full benchmarks are distinguished.
Concision / 精简	Both homepages: 147→123 lines; visible prose characters EN 6848→5961, ZH 3315→2854. This round modifies 26 existing Markdown files, net -302 lines. Earlier reasoning is a short history index; older reports/PDFs remain traceable.
Media / 媒体	All ten GIF references remain in the same order, including the first hero. Only the older baseline chart is removed from homepage display; no media/evidence/PDF bytes changed. Recorded viewing remains labelled as saved-map replay.
Bilingual and links / 双语链接	Paired prose manually compared for claims, numbers, gates and status; configured bilingual/local-link/media checks pass. All 27 added external URLs return HTTP 200.
Evidence / 证据	36 portable records, 97 paired cells, 35 delayed cells and six post-hoc controls verify. Runtime source, tests, scripts, configs, environments, original results, annotations, archive and media have no diff against main.
Dependency work / 依赖	Direct Pillow declaration preserves the existing version. Optional audit tools and three architecture contracts remain; no prior locked version changed. High-confidence Vulture is advisory evidence, not proof of no dead code. IWYU was not run.
CI	Push and PR Windows/Ubuntu CPU jobs pass on this head, including 44 tests, audits, evidence/doc checks and synthetic smoke runs. Push · PR. Author-method jobs were intentionally skipped.
Isolation / 隔离	Only this maintenance worktree was edited. Main remains clean at the base above; author, hardware, study and local-backup worktrees were not changed by this task.

No blocking issue found in self-review. This is not independent human scientific review. Keep draft; no auto-merge; explicit owner approval is required. Later changes invalidate this head-bound review.

当前仍缺 room2 的正式分析、独立身份标注和有界原型的实际成本证据。这些缺口被如实展示，未用文字精简替代实验或改写不利结果。

[p20030920p]

Current-head self-review / 当前提交自审

Head: 8a886a5e2099f7c477fb03afcea89cfcb63296d1
Main base: dfd4a05f02352d32b4f0a2bf522e189708c9e2b2

This supersedes the c43b3b8 review for the current head. The latest commit changes only the two READMEs; the earlier dependency work and evidence/history edits are unchanged.

Item / 项目	Finding / 结论
Narrative / 写作	SLAM definition → two directions → five elements and improvement effects → four paper focuses → common dependency and open question. All README section numbers removed; both languages have 133 lines versus main's 147.
Research scope / 研究边界	Pose-dependent map decisions motivate the question. A common defect is not established; ConceptGraphs is the first test. Supplied-pose mapping is distinct from full SLAM/navigation. HOV-SG's paper hierarchy is distinct from the GIF's segment core.
Preservation / 保留内容	One-off comparison against c43b3b8 verifies the hero, all ten GIF references and their order. Results, scores, evidence, H1, decision gates and branch content are byte-identical apart from removing heading numbers. No media, raw evidence, annotations or PDF changed.
Bilingual and links / 双语链接	Definitions, five table rows, paper focuses, example and candidate status agree across languages. Bilingual/local-link/media evidence checker and git diff --check pass. No external URL was introduced by this two-file revision.
Current CI	Push and PR Windows/Ubuntu CPU jobs pass on this head: audit tools, tests, evidence/docs checks and synthetic smoke runs. Push · PR. Author-method jobs were intentionally skipped.
Isolation / 隔离	Only the maintenance worktree was edited. Main stays clean at the base above. No other window's branch, files, environment or run directory was changed.

No blocking documentation issue found in self-review. This is not independent scientific review. Keep draft; no auto-merge; explicit owner approval is required. Any later modification requires a new head-bound review.

[p20030920p]

Current-head self-review / 当前提交自审

Head: 78122808e029af84fb6f5e2ed9f37e88fc1f3285
Main base: dfd4a05f02352d32b4f0a2bf522e189708c9e2b2

This supersedes the 8a886a5 review for the current head. Earlier dependency/evidence work is unchanged.

• Scope: the latest commit changes only README.md and README.zh-CN.md. Only the related-work media blocks and the five-viewer details block differ from 8a886a5; the SLAM explanation, question, scores, counterevidence, H1, gates and branch content remain identical.
• Captions: every non-hero GIF now stands alone with a dataset, run scope, visible task and video link. KITTI-00 141-scan runs are distinguished from the 21 displayed scans; Replica room0 40-frame/20-display and 8-frame cores are distinguished from the 400-frame author-viewer map. These descriptions were checked against paper cards and recording/source records.
• Scientific scope: rendered replays, recorded 3D viewing, final maps and saved incremental snapshots remain distinct. Query candidates are unverified; HOV-SG hierarchy/navigation are not claimed. No new recording or experiment was performed.
• Preservation: all ten GIF paths and their order, including the hero, remain unchanged. All previous URLs remain; four existing MP4s are newly linked directly. There are no GIF layout tables. No media, evidence, annotation or PDF bytes changed.
• Validation: bilingual/local-link/media checker, the one-off README scope/media comparison and git diff --check pass. Current-head Windows/Ubuntu CPU jobs pass for push and PR, including audits, tests, evidence/docs checks and smoke runs. Author-method jobs were intentionally skipped.
• Isolation: only this maintenance worktree was edited. Main remains clean at the base above; other branches and their run directories/environments were not changed.

No blocking documentation issue found in self-review. This is not independent scientific review. Keep draft; no auto-merge; explicit owner approval is required. Later changes require a new head-bound review.

中文：相关动图均改为单独展示，逐个解释数据集、运行内容和画面。两种语言同步，指标与实验结论保留。本轮只改首页，未改主分支。

[p20030920p]

Current-head self-review / 当前提交自审

Head: f42bfc0cd39292452505c28fce3b9566a4395df9
Main base: dfd4a05f02352d32b4f0a2bf522e189708c9e2b2

This supersedes the 7812280 review for the current head. The latest commit exports clearer previews; it does not introduce new scientific results.

• Code: one serial two-pass FFmpeg exporter validates every MP4 against its recorded SHA-256 before creating output. It uses native dimensions, explicit intervals and a 256-color palette. GIF duration, dimensions, looping, nonblank samples, full decoding and a 10 MiB per-file budget are checked. WSL uses low-priority single-thread CPU encoding; no package/environment installation or GPU work.
• Negative checks: a mismatched author-video hash is rejected before output creation. An existing output directory is rejected, with every existing GIF hash preserved. The first oversized continuous ConceptGraphs attempt remains unpublished in ignored local runs; the final version samples all nine recorded stages rather than dropping later queries.
• Media/provenance: nine GIFs total 18.98 MiB, largest 4.20 MiB. Four core replays and five viewer clips derive from unchanged, hash-bound MP4s. ConceptGraphs RViz has chronological three-second excerpts of every one of its nine stages; other time windows and commands are recorded. Original MP4s, historical GIFs, raw evidence and the first hero remain unchanged.
• Visual review: first/middle/last frames of all nine GIFs, all nine ConceptGraphs stages and before/after text/point-cloud crops inspected. Portable comparison/stage sheets are hash-bound in the new record. This is AI visual inspection, not independent human review. No frame interpolation or AI image generation was used.
• Scientific and bilingual scope: supplied-pose replays/viewer inspection, query uncertainty and full benchmark/navigation limits remain explicit. Research, scores, counterevidence and H1/gates are unchanged. Both READMEs point to the same nine new GIFs; concise regeneration instructions are bilingual.
• Validation: source/snapshot hashes, original-media copies, nine added media registrations, existing-slot preservation, local links, bilingual pairs, ruff, Vulture and 37 portable evidence records pass. Current-head Windows/Ubuntu CPU jobs pass for push and PR, including audits, tests, evidence/docs checks and smoke runs. Transcoding ran locally; CI verifies committed evidence. Author-method jobs were intentionally skipped.
• Isolation: main remains clean at the base above. Only this maintenance worktree and the new local media-handoff folder were written; other branches, environments and ongoing runs were not changed.

No blocking issue found in self-review. Keep draft; no auto-merge; explicit owner approval is required. Any later changes require another head-bound review.

中文：九个 GIF 已从原 MP4 重导出，文字与点云更清晰，ConceptGraphs 九个阶段完整覆盖。原视频和新版 GIF 另存本地使用包；源文件、旧实验与主分支不改。