Our Work

Reproducibility code and data for the manuscript "Selecting Generative Text-to-Image Models for Industrial-Heritage Visualization: a Design-Code-Constrained AHP-TOPSIS Benchmark with Computational Cro…

Reproducibility code and data for the manuscript "Selecting Generative Text-to-Image Models for Industrial-Heritage Visualization: a Design-Code-Constrained AHP-TOPSIS Benchmark with Computational Cross-Validation (Krasny Treugolnik, Saint Petersburg)." Ten generative text-to-image systems are compared on five standardized architectural scenes under an explicit heritage design code, ranked with AHP-TOPSIS (consistency ratio 0.012) and cross-validated with a computational image-analysis pipeline (criterion collinearity, rater-noise bootstrap, weighting stress test, and a leave-one-out stylistic-fingerprint classifier). Includes the expert decision matrix, computational image features, weights, ranking, all Python analysis scripts, the design code and standardized prompts. Source repository: https://github.com/i-SAVE/krasny-treugolnik-mcda

wisp-science v0.24.0 这是继 v0.23.0 之后的研究工作台体验与可靠性版本。Wisp 现在能通过真实浏览器捕获可供视觉模型查看的当前页面截图;研究图谱、数据库中登记的 Artifact 和 Run 产物不再只存在于后台,均可直接在界面中查看、打开和追溯。此版本还加入 SSH 定向信任边的可见与撤销管理、存储空间与会话 token 用量概览,并改善长对话、文献预览、运行时初始…

wisp-science v0.24.0 这是继 v0.23.0 之后的研究工作台体验与可靠性版本。Wisp 现在能通过真实浏览器捕获可供视觉模型查看的当前页面截图;研究图谱、数据库中登记的 Artifact 和 Run 产物不再只存在于后台,均可直接在界面中查看、打开和追溯。此版本还加入 SSH 定向信任边的可见与撤销管理、存储空间与会话 token 用量概览,并改善长对话、文献预览、运行时初始化和 SSH 诊断。 下载 macOS(.dmg,已签名 + 公证):区分 Apple Silicon(aarch64)与 Intel(x86_64),请下载对应架构。 Windows 安装包(MSI / NSIS):尚未签名,可能触发 SmartScreen,可选择「仍要运行」。 见下方 Assets。 新增功能 浏览器、图谱与 Artifact 可见性 真实浏览器截图工具:新增需批准的 web_screenshot,可捕获选中 HTTP(S) 标签页的可见视口并作为图像结果交给视觉模型,适用于图表、canvas/WebGL、二维码和布局问题;无需新增浏览器扩展权限。 研究图谱面板:右侧新增"研究图谱"标签,按节点类型显示项目中的问题、决策、数据与文献及其出边;节点现在还会展示决策理由、DOI、行数等已有元数据。 完整 Artifact 发现路径:Artifact 面板会合并对话中提到的文件与数据库已登记的 Artifact,Run 卡片会显示其产物并可直接打开。由 Run、MCP bridge 或 Agent 写入的结果不再因未出现在聊天文本中而不可见。 更易浏览的生成结果:对话内的 Generated 区块改为缩略图网格;图片显示预览,其余类型保留类型徽标,超过八项可展开查看,点击仍打开右侧预览。 工作区、项目与代码复用 Storage 与 Usage 概览:Settings → Workspace 新增 Storage,展示应用数据目录和数据库、Python 环境、插件、工作区等占用;Usage 按会话汇总已持久化的输入、输出、推理与缓存 token,sub-agent 用量会归入根会话。 项目卡片 Artifact 数量:Projects 首页现在同时显示会话和 Artifact 数量。 Library 键盘入口与指定版本复跑:Ctrl/⌘P 可直接打开 Library;收藏代码或图像生成代码的任一版本都可插入 composer,在发送前审阅并重新运行,不会改写历史版本。 SSH 信任边可审计、可撤销:Settings → Environments 会显示 Agent 创建的定向 SSH trust edge,并提供 Revoke。撤销先移除授权记录,再尽力清理两端的受管密钥;离线主机不会阻止撤销。 修复内容 长内容阅读:超过 12 行或 600 字符的用户消息默认折叠,可切换"显示全部/收起";复制与编辑始终使用完整文本。attempt_completion 的最终答案不再被 4,000 字符预览限制截断。 更快的首次发送:请求路径不再同步等待 uv pip install 下载科学依赖。后台依赖安装尚未完成时,bio-tools MCP 会快速报告缺少依赖而非无限阻塞普通对话或 Python REPL。 SSH 错误可诊断且受控:SSH master 失败时会报告 OpenSSH 的退出状态和 stderr,正确识别认证失败并启用 guard;策略同时禁止以 paramiko、fabric 等客户端库绕过已保存凭据与主机密钥校验。 更可靠的文献预览:本地 PDF 预览上限提高到 100 MB(远端仍为 32 MB),DOCX/PPTX/XLSX 中的 TIFF 媒体会在内存中转为 PNG,避免正文可见而图像空白。 控制本地 trace 占用:.wisp/subagents/ 仅保留最新 20 个 explore trace,避免长期项目无限增长。 兼容性与升级说明 可直接覆盖安装 v0.23.0;项目、会话、模型、连接、ExecutionContext、Run、Artifact、研究图谱、插件和 Agent workflow 数据会继续保留,无需手工迁移。 web_screenshot 仅能操作已有真实浏览器 bridge 连接中的 HTTP(S) 标签页,仍受现有浏览器权限与审批流程约束。 SSH trust edge 的撤销立即移除 Wisp 的直接传输授权;如果任一端无法连接,远端受管公钥的清理会在界面中明确标记为失败,而不会恢复授权。 Storage 只报告目录和分类占用,本版本不迁移应用数据目录,也不显示磁盘可用空间。 已知限制 已登记的远程 ssh:// Artifact 可以显示在面板中,但当前右侧预览尚不能读取其内容;delegated child frame 的 Artifact 仍不显示在父会话的列表中。 远程 PDF/DOCX 等预览仍限制为 32 MB,以避免大文件跨网络传输。 Windows 安装包仍未签名,可能触发 SmartScreen。 需自备模型与第三方科研服务的 API Key;对应费用、配额与可用性由用户及服务商负责。 测试与交付 v0.23.0..v0.24.0 包含 19 个非合并提交,涉及 45 个文件;其中 11 项功能、5 项修复与 UI/E2E 覆盖更新。 新增和更新覆盖浏览器截图、研究图谱与元数据、Artifact/Run 关联、SSH 信任边与错误诊断、token 用量、长消息显示、TIFF 文档预览,以及 PDF 100 MB 上限。 推送 v0.24.0 标签后,GitHub Actions 会构建 macOS Apple Silicon / Intel DMG 与 Windows MSI / NSIS,并附加到本 Release。 反馈 欢迎在 GitHub Issues 反馈问题。浏览器截图问题请说明浏览器 bridge 状态、目标页面类型和脱敏复现步骤;SSH 问题请提供来源/目标 context 与脱敏后的 OpenSSH 错误。API key、token、密码、Cookie、SSH 私钥和未脱敏科研数据请勿上传。 English: wisp-science v0.24.0 brings key research-workbench records into the visible UI and strengthens everyday reliability. The real-browser bridge can now return the visible viewport as an image for vision-capable agents; the right pane exposes the project's research graph and database-registered Artifacts; and Run cards link directly to their outputs. This release also makes directed SSH trust edges visible and revocable, adds workspace Storage and per-session Usage views, and improves long chats, literature previews, runtime startup, and SSH diagnostics. What's Changed Capture a selected real-browser tab's visible viewport with the approved web_screenshot tool and pass it through the standard vision-result path, without adding extension permissions. Browse research graph nodes, relationships, and persisted node metadata in the right pane; merge registered Artifacts into the Artifact pane and open each Run's produced Artifact from its card. Render in-thread generated files as thumbnail tiles with a progressive "more" expansion instead of a long list of cards. Add Workspace Storage breakdown and per-session token Usage views, including sub-agent usage folded into the root session; show Artifact counts on project cards. Open the global Library from Ctrl/⌘P and insert any starred code/figure version into the composer for review-before-send reruns. List and revoke Agent-created directed SSH trust edges from Environments, with immediate authorization removal and best-effort remote-key cleanup. Collapse very long user messages without affecting copy/edit, preserve full attempt_completion answers, avoid blocking the send path while Python dependencies install, and retain only the newest 20 explore traces. Surface OpenSSH's real failure diagnostics and prevent SSH client-library fallbacks that bypass saved credentials and host-key checks. Raise local PDF previews to 100 MB and transcode embedded OOXML TIFF media to PNG so figures remain visible. Compatibility: Installs over v0.23.0 in place with no manual migration. web_screenshot works only through the existing real-browser bridge on HTTP(S) tabs and follows the same approval model. Revoking a trust edge immediately removes Wisp authorization even when a disconnected host prevents remote-key cleanup. Remote file previews remain capped at 32 MB. Full Changelog: https://github.com/xuzhougeng/wisp-science/compare/v0.23.0...v0.24.0

We introduce the term Digital Echopraxia to describe a systemic failure mode present across a broad class of digital systems: the production of approval-optimized output that mimics understanding, ins…

We introduce the term Digital Echopraxia to describe a systemic failure mode present across a broad class of digital systems: the production of approval-optimized output that mimics understanding, insight, or genuine response without the grounding that would make such output reliable or honest. The clinical analogue is echopraxia: the involuntary imitation of another’s actions without volitional comprehension. Unlike its neurological counterpart, Digital Echopraxia is not incidental but architecturally induced, arising wherever digital systems are trained or optimized against human or proxy approval signals. We trace its analytical progression from early engagement-maximizing recommendation systems through contemporary Reinforcement Learning from Human Feedback (RLHF)-trained large language models (LLMs), hypothesizing that each successive form is more fine-grained and harder to detect than the last. We argue that the detection burden falls disproportionately on the people least equipped to bear it, and that this burden grows inversely with the sophistication of the mimicry, with consequences that reach directly into AI alignment, public trust in information, and the reliability of human-machine communication.

Accurate monitoring of crop yield is important for ensuring food security. Current yield estimation methods, such as machine learning models or the assimilation of remotely sensed biophysical variable…

Accurate monitoring of crop yield is important for ensuring food security. Current yield estimation methods, such as machine learning models or the assimilation of remotely sensed biophysical variables into crop growth models, depend heavily on ground observations and involve significant computational costs. To solve these problems, a hybrid framework coupling the World Food Studies Simulation Model (WOFOST) and the Gated Recurrent Unit model (GRU) was proposed for maize yield estimation in Northeast China from 2019 to 2024. The model was trained exclusively on WOFOST-simulated data without using any ground-measured yield labels, and was subsequently transferred to Sentinel-2-derived leaf area index time series to generate spatially explicit yield maps. This dataset provides 10 m annual maize yield maps for Northeast China from 2019 to 2024. Accompanying uncertainty layers (coefficient of variation) for 2023 and 2024 are also provided to support user assessment of local prediction reliability. *** The data files are in ".tif" format *** Temporal Resolution: annually *** Temporal coverage: 2019–2024 (yield maps); 2023–2024 (uncertainty layers; remaining years under production) *** Spatial Resolution: 10 m *** Unit: kg/ha *** Projection information: EPSG: 4326 *** Note: The CV values in the uncertainty layers have been scaled up by a factor of 10,000 for integer storage. Users should divide the pixel values by 10,000 to obtain the original CV values. *** Data access: https://zenodo.org/records/19547014 (Hu et al., 2026)

[no abstract]

Coupled Model Intercomparison Project Phase 7 (CMIP7) ScenarioMIP greenhouse gas (GHG) forcing files.This is ready for CMIP7 production use.The files are (or soon will be) available directly from ESGF…

Coupled Model Intercomparison Project Phase 7 (CMIP7) ScenarioMIP greenhouse gas (GHG) forcing files.This is ready for CMIP7 production use.The files are (or soon will be) available directly from ESGF (make sure to select the version(s) of interest to you as this search will show all versions), this record is simply used to provide a DOI and reproducibility.For a list of the changes made over time, see the source repository.For an overview of the delivery status of all CMIP7 ScenarioMIP forcing datasets, see input4MIPs CVs' delivery summary.For more detail on each CMIP7 dataset, start from input4MIPs CVs' dataset overviews.

This repository contains the data and derived analytical outputs supporting a study of plant–floral visitor interaction networks in potato field margins in southeastern Buenos Aires, Argentina. …

This repository contains the data and derived analytical outputs supporting a study of plant–floral visitor interaction networks in potato field margins in southeastern Buenos Aires, Argentina. Data were collected from 35 independent sampling sites between 2020 and 2023 through standardized transect surveys. A total of 4,371 interactions were recorded between 52 flowering plant taxa and 123 floral visitor morphospecies. The deposited files include quantitative plant–floral visitor interaction records, local field-margin variables, landscape metrics calculated for three non-overlapping spatial bands (0–500 m, 500–1000 m, and 1000–2000 m), observed network metrics, and standardized effect sizes (SES) for network-level specialization (H2′), weighted nestedness (WNODF), and modularity. The repository also contains network metrics and SES recalculated after excluding Apis mellifera, comparisons between complete networks and networks without this species, model-selection outputs, coefficient tables, residual diagnostics, spatial-autocorrelation tests, a variable dictionary, metadata, and quality-control summaries. Local environmental variables include field-margin width, flowering plant richness, floral unit abundance, grass cover, and bare-soil cover. Landscape variables include the proportion of semi-natural habitat, Shannon landscape diversity, and edge density. The landscape variables were calculated separately for the three spatial bands. The files are temporarily under restricted access while the associated manuscript is being prepared or evaluated. They are intended to be made openly available following acceptance or publication of the associated article. The analytical R code is not included in the current version of the repository.

Geometric decoding of spatial transcriptomics in pathology. Cell density correlation analysis across tumor microenvironment compartments.

Latest production release. XDI-Validator is a standalone JSON Schema based validator for XDI files. XDI (*.xdi) is a format used to save XAS data. This validator aiming to be fully compliant with the …

Latest production release. XDI-Validator is a standalone JSON Schema based validator for XDI files. XDI (*.xdi) is a format used to save XAS data. This validator aiming to be fully compliant with the XDI/1.0 specification, as detailed in XAS Data Interchange Format Draft Specification, version 1.0 and Dictionary of XAS Data Interchange Metadata. Changelog highlights: Make Mono.d_spacing conditionally required per XDI/1.0 spec by @smrgeoinfo. From @smrgeoinfo 's PR #6 : The XDI/1.0 specification requires Mono.d_spacing only when the abscissa (Column.1) is monochromator angle or encoder step count — the mono d-spacing is needed only to convert those units into energy: The Mono.d_spacing header field must be specified if the abscissa is conveyed as monochromator angle. The current schema requires Mono.d_spacing unconditionally, which produces spurious errors on the many XDI files whose abscissa is energy or wavelength (where d_spacing is optional). New Contributors @smrgeoinfo made their first contribution in https://github.com/AAAlvesJr/XDI-Validator/pull/6 Full Changelog: https://github.com/AAAlvesJr/XDI-Validator/compare/v1.4.0...v1.5.0