Our Work

No description provided.

Работа исследует три связанные, но не образующие обязательной последовательности области: условия исследовательского хода, складывание различимости, участие и продолжение. Рассматривается, как исследо…

Работа исследует три связанные, но не образующие обязательной последовательности области: условия исследовательского хода, складывание различимости, участие и продолжение. Рассматривается, как исследование может сохранять свободу пробного хода, не расширяя вывод дальше имеющихся оснований; как различимость складывается в связи с накопленным опытом, движением внимания, масштабом, положением и используемыми средствами; и как уже сложившиеся формы, различия и отношения участвуют в текущем ходе и получают или не получают дальнейшее продолжение.Отдельное внимание уделяется неравным условиям участия, историческим переходам, судьбе различий, локальной достаточности представления относительно требования и тому, как само исследование может войти в последующий ход. Работа не предлагает общей модели познания или обязательного метода анализа. Введённые различения используются локально — там, где они меняют допустимый вывод, его основание или область применимости. Аппарат самой работы также не получает привилегированного статуса: его различения могут перейти в приобретённый способ работы, изменить функцию или отойти, когда перестают быть необходимыми.

<!-- release-title: v1.15.0: Portable Projects · 项目独立数据库 --> wisp-science v1.15.0 Each project carries its own database, so a project folder can move and sync on its own. Starting with v1.15.0, …

<!-- release-title: v1.15.0: Portable Projects · 项目独立数据库 --> wisp-science v1.15.0 Each project carries its own database, so a project folder can move and sync on its own. Starting with v1.15.0, a project's conversations, artifacts, runs and research records live in its own folder at .wisp/project.sqlite, not in the app-wide database. Copy or import the folder on another machine and the project comes with it. Remove the project from Wisp and the records stay in the folder. On top of this, a project folder can sit in a cloud drive (Nutstore, Baidu Netdisk, OneDrive, iCloud Drive, Dropbox) and move between devices through safe, versioned snapshots. The work shipped in #1352, #1353 and #1356. Downloads macOS (.dmg, signed + notarized): Apple Silicon (aarch64) and Intel (x86_64); compatible versions can update in-app. macOS SwiftUI Preview (.dmg, signed + notarized): new this release. Wisp-Science-SwiftUI-Preview_1.15.0_aarch64.dmg and _x86_64.dmg, each with a .sha256 file. See the Compatibility section before installing. Windows (MSI / NSIS): installers remain unsigned while SignPath release signing is disabled. Linux (deb / AppImage): x86_64 and aarch64. Installers will appear in Assets as the platform builds finish. The website download page also provides platform-specific installer links backed by the configured download mirror. Portable Projects Each project owns its database Project records are stored in the project folder at .wisp/project.sqlite, identified by .wisp/project.json. The app database keeps only machine settings, credentials, execution contexts and a list of project locations. Existing projects migrate automatically on first launch. Wisp builds and verifies each project database in a staging copy before switching over, so a failed or interrupted migration leaves the project working and is retried at the next start. Removing a project from Wisp keeps its folder and records. Opening the folder again, even on a fresh install, brings back its full history. Secrets stay in the OS keyring and are never written into project folders. Move a project as a folder Export project now offers Export directory next to Export ZIP. Both contain the same complete project: manifest.json, metadata/project.sqlite and workspace/. Import project folder restores that package and opens its workspace in place. Ordinary folders, packages with missing or damaged metadata, and duplicate project IDs are refused with an error. Wisp no longer registers an empty project in those cases. Opening an existing folder in place now shows its Files panel right away, even when the folder has no conversations yet. Keep a project in a cloud drive In Project Settings, press Save safely for cloud-drive sync. The live database moves to this device's app data, and Wisp publishes complete, checksummed snapshots to .wisp/revisions/. The drive client only ever sees finished files. Wisp publishes about 20 seconds after a turn ends, and never while a turn is running. Project cards show Saved to folder, Changes not yet saved to folder, Newer version in folder, Waiting for the cloud drive or Folder version conflict. Sync now publishes immediately. On another device, wait for the drive to finish, then use Import project → Import project folder. Later, opening the project adopts a newer version from the folder if this device has nothing unsaved. If two devices changed the project from the same version, Wisp overwrites nothing and opens the conflict dialog: Use this device or Use remote version. A snapshot whose size or checksum does not match is treated as still downloading and is never applied. Folder projects need no relay, sync key or device code. Details: docs/project-sync.md. One offline project no longer blocks the rest An unplugged disk or a moved folder used to make the project list, project creation, schedules, run monitoring and startup recovery fail for every project. Now Wisp skips that project with a warning and keeps it listed. Opening it still reports the error. Cross-project searches and safety checks still fail explicitly instead of returning partial results. Subscription Sign-in ChatGPT Plus/Pro (Codex): in Settings → Models, sign in through the browser or with a device code (which works over SSH and WSL). Requests go to the Codex Responses API instead of an OpenAI Platform key. From the CLI: wisp-science login codex, then WISP_PROVIDER=openai_codex. SuperGrok / X Premium+ (xAI): sign in with an xAI device code from Settings → Models → SuperGrok. Requests go to api.x.ai, and the token is only sent to x.ai hosts. From the CLI: wisp-science login xai, then WISP_PROVIDER=xai_oauth. Tokens are stored in the OS keyring and refreshed before they expire. A saved sign-in from the CLI can be reused in Settings on the same machine. Other Improvements and Fixes Browser automation runs for several projects at once. shared and workspace sessions no longer block each other, and each project gets its own isolated workspace browser profile (up to three at a time). The figure-duplicate-audit Skill is now research-integrity-audit. Besides the image checks, it screens Source Data and tables for copied blocks, fixed offsets, digit and Benford anomalies, GRIM/GRIMMER failures, and p-values that don't match their reported statistics. Command palette arrows stay inside the visible window. Tutorials open from Help and the projects home. The session group name field no longer triggers browser autofill. The Models settings toolbar wraps on narrow windows. The app logs each step of its exit sequence to help diagnose hangs on quit. The native SwiftUI shell (macOS) and WinUI 3 shell (Windows), still in preview, now cover more of the WebView app. This includes project creation and archive import, Files, session groups with sort and multi-select, the shared library, research calendar and journey, publications, a scratch chat, file attachments, one queued follow-up during a turn, privacy mode, conversation Markdown with tables and code copy, approvals, and native settings pages. Compatibility and Scope Do not downgrade after upgrading. Migration moves each project's records out of the app database into the project folder. v1.14 and earlier read only the app database and would show your projects without their conversations. Migration needs a volume that supports hard links. On a volume without them, the project keeps working from the app database and migration is retried later. A registered project whose database is missing reports an error. Wisp never replaces it with an empty one. Don't copy a live project folder while Wisp is running. Use Export project, Sync now or cloud-drive mode instead. Cloud-drive mode is opt-in per project and cannot be turned off yet. Enable it while Wisp is closed on the other devices. It marks the folder as version 2, which older Wisp versions refuse to open. Removing the project leaves its local cache in the app data directory. Subscription sign-in uses the public Codex and Grok CLI OAuth clients for your personal subscription. xAI has been reported to reject some tiers with HTTP 403 after a successful sign-in; use an xAI API key in that case. Model IDs are typed by hand. SwiftUI Preview requires macOS 13 or later and installs next to the regular app as Wisp Science Preview.app. It uses the same project data, settings and keyring as the WebView app, so it does not isolate your data. It updates only by manual download and is not yet feature-complete. The WebView app and its in-app updater are unchanged. Full Changes All commits since v1.14.0 Release Verification Release checks cover the workspace test targets, frontend WebAssembly compilation, and Rust formatting. Installer builds start after this GitHub Release is created. wisp-science v1.15.0 每个项目自带数据库,项目文件夹可以单独搬走、单独同步。 从 v1.15.0 起,项目的会话、产物、运行记录和研究记录保存在项目自己的文件夹里(.wisp/project.sqlite),不再放在应用的全局数据库中。把文件夹复制或导入到另一台电脑,项目会完整跟过去;从 Wisp 移除项目,记录仍留在文件夹里。在此基础上,项目文件夹可以放进坚果云、百度网盘、OneDrive、iCloud Drive 或 Dropbox,通过安全的版本快照在多台设备间切换。实现见 #1352、#1353 和 #1356。 下载 macOS(.dmg,已签名 + 公证):Apple Silicon(aarch64)与 Intel(x86_64);兼容版本可在应用内更新。 macOS SwiftUI 预览版(.dmg,已签名 + 公证):本版本新增。Wisp-Science-SwiftUI-Preview_1.15.0_aarch64.dmg 与 _x86_64.dmg,各附 .sha256 校验文件。安装前请先阅读"兼容性与适用范围"。 Windows(MSI / NSIS):SignPath 正式签名仍处于停用状态,安装包暂未签名。 Linux(deb / AppImage):x86_64 与 aarch64。 各平台构建完成后,安装包将陆续附加到下方 Assets。官网下载页 也提供按平台选择的安装包入口,使用已配置的下载镜像。 项目独立数据库 每个项目拥有自己的数据库 项目记录保存在项目文件夹的 .wisp/project.sqlite 中,由 .wisp/project.json 标识。应用数据库只保留本机设置、凭据、执行环境和项目位置列表。 已有项目在首次启动时自动迁移。Wisp 先在暂存副本中构建并校验每个项目的数据库,再切换过去;迁移失败或中断时项目照常可用,下次启动重试。 从 Wisp 移除项目时,文件夹和记录都会保留。重新打开这个文件夹(即使是全新安装)即可恢复完整历史。 密钥仍保存在系统密钥环中,不会写入项目文件夹。 以文件夹形式搬移项目 导出项目新增导出目录,与导出 ZIP 并列,二者内容相同:manifest.json、metadata/project.sqlite 和 workspace/。 导入项目文件夹会恢复这个导出包,并直接在原位置打开其工作区。普通文件夹、元数据缺失或损坏的包、重复的项目 ID 都会报错拒绝,不再登记一个空项目。 原位打开已有文件夹时,即使还没有会话,也会立即显示文件面板。 把项目放进网盘 在项目设置中点击启用网盘安全保存。实时数据库移到本机应用数据目录,Wisp 把完整且带校验和的快照发布到 .wisp/revisions/,网盘客户端只会看到写完的文件。 一轮对话结束约 20 秒后发布,对话进行中不会发布。项目卡片显示已保存到文件夹、有更改尚未保存到文件夹、文件夹中有更新的版本、等待网盘同步完成或文件夹版本冲突。立即同步会马上发布。 在另一台设备上,等网盘同步完成后,使用导入项目 → 导入项目文件夹。之后打开项目时,如果本机没有未保存的更改,会自动采用文件夹中的新版本。 两台设备从同一版本各自修改时,Wisp 不覆盖任何内容,而是弹出冲突对话框:采用此设备版本或采用远端版本。大小或校验和不匹配的快照视为仍在下载,绝不应用。 文件夹项目不需要中继、同步密钥或设备码。详见 docs/project-sync.zh-CN.md。 单个项目离线不再影响其他项目 过去一个项目所在的磁盘拔出或文件夹被移动,会导致所有项目的列表、新建项目、定时任务、运行监控和启动恢复全部失败。现在 Wisp 会跳过该项目并给出警告,项目仍显示在列表中,打开时报告错误。跨项目搜索和安全检查仍会明确报错,不返回不完整的结果。 订阅登录 ChatGPT Plus/Pro(Codex):在设置 → 模型中通过浏览器或设备码登录(设备码适用于 SSH 和 WSL),请求发往 Codex Responses 接口,无需 OpenAI 平台 API 密钥。命令行:wisp-science login codex,然后设置 WISP_PROVIDER=openai_codex。 SuperGrok / X Premium+(xAI):在设置 → 模型 → SuperGrok 中用 xAI 设备码登录,请求发往 api.x.ai,令牌只会发送到 x.ai 域名。命令行:wisp-science login xai,然后设置 WISP_PROVIDER=xai_oauth。 令牌保存在系统密钥环中,过期前自动刷新。在同一台电脑上,命令行保存的登录可以直接在设置中复用。 其他改进与修复 浏览器自动化支持多个项目同时使用:shared 与 workspace 会话互不阻塞,每个项目使用独立的 workspace 浏览器配置(最多同时三个)。 figure-duplicate-audit Skill 更名为 research-integrity-audit(学术诚信审查)。除图片检查外,还会筛查 Source Data 和表格中的重复数据块、固定差值、末位数字与 Benford 异常、GRIM/GRIMMER 不一致,以及与所报告统计量不符的 p 值。 命令面板的箭头不再超出可见窗口;可从帮助和项目首页打开教程;会话分组名称输入框不再触发浏览器自动填充。 窄窗口下模型设置的工具栏按钮会自动换行;应用退出时逐步记录日志,便于排查退出卡住的问题。 仍处于预览阶段的原生 SwiftUI(macOS)与 WinUI 3(Windows)界面进一步向 WebView 版本看齐,覆盖:新建项目与导入归档、文件页、会话分组(排序、多选)、共享文献库、研究日历与研究历程、发表工作区、临时聊天、附加文件、对话进行中排队一条后续消息、隐私模式、支持表格和代码复制的对话 Markdown、审批操作,以及原生设置页。 兼容性与适用范围 **升级后请不要降级。**迁移会把每个项目的记录从应用数据库移到项目文件夹中;v1.14 及更早版本只读取应用数据库,会看到项目但看不到其中的会话。 迁移要求项目所在磁盘支持硬链接;不支持时项目继续使用应用数据库,稍后重试迁移。已登记项目的数据库缺失时会明确报错,Wisp 绝不会用空数据库代替。 Wisp 运行时不要直接复制项目文件夹,请使用导出项目、立即同步或网盘模式。 网盘模式按项目单独开启,目前无法关闭;请在其他设备上的 Wisp 关闭时开启。开启后文件夹标记为版本 2,旧版 Wisp 会拒绝打开;移除项目后,本机应用数据目录中的缓存文件会保留。 订阅登录使用 Codex 与 Grok CLI 的公开 OAuth 客户端,仅用于个人订阅。有报告称 xAI 会对部分订阅档位在登录成功后返回 HTTP 403,此时请改用 xAI API 密钥。模型 ID 需手动填写。 SwiftUI 预览版需要 macOS 13 或更高版本,以 Wisp Science Preview.app 的名称与正式版并存。它与 WebView 版本共用项目数据、设置和密钥环,并不隔离数据;只能手动下载更新,功能尚不完整。WebView 版本及其应用内更新不受影响。 完整变更 自 v1.14.0 以来的全部提交 发布验证 发布检查覆盖工作区测试目标、前端 WebAssembly 编译和 Rust 格式检查。各平台安装包在本次 GitHub Release 创建后开始构建。

EcoBase v0.2.2 Diff since v0.2.1 Merged pull requests: Keep one cell coordinate per cell for float cell sizes (#27) (@mkborregaard)

Closing keynote at a small focussed conference on Molecular, Imaging and Digital Biomarkers for Parkinson's Disease and Related Disorders. European specialists in clinical care, mechanistic research, …

Closing keynote at a small focussed conference on Molecular, Imaging and Digital Biomarkers for Parkinson's Disease and Related Disorders. European specialists in clinical care, mechanistic research, diagnosis and prognosis of disease, prediction and monitoring of therapies, AI, and molecular, imaging and digital biomarkers shared their experiences, needs and potential solutions. I added the focus on multimodal data and AI, highlighting lessons-learned in multi-omics approaches and potential approaches using agentic AI, with a strong advice to validate and confirm scientific literature that is now being poluted with problematic publications ad AI-generated false data. We had a good discussion afterwards ...

La Nueva Agenda Urbana publicada por ONU-Habitat establece el objetivo de proteger y salvaguardar el patrimonio cultural y natural del mundo, especialmente el patrimonio local y regional que corre rie…

La Nueva Agenda Urbana publicada por ONU-Habitat establece el objetivo de proteger y salvaguardar el patrimonio cultural y natural del mundo, especialmente el patrimonio local y regional que corre riesgo de ser marginado frente a las tendencias globales de aculturación. Conscientes del incremento de la consulta del patrimonio en plataformas digitales y de la oportunidad de conservación que este formato implica, UNESCO redactó en 2015 la Recomendación de Naciones Unidas sobre la Preservación y el Acceso al Patrimonio Documental incluido en forma digital. En este artículo se propone aprovechar este proceso para el uso didáctico del patrimonio digitalizado. Se propone como modelo el diseño de actividades didácticas incluidas en el patrimonio editorial digitalizado por la Residencia de Estudiantes por el proyecto PLATA de la universidad UDIT, incorporando herramientas de inteligencia artificial y servicios en la nube que permiten profundizar en su uso didáctico explotando las posibilidades que permite la digitalización.

В статье анализируются нормативно-правовые основы управления операционным риском в коммерческих банках и применение современных аналитических (статистических и основанных на машинном обучении) моделей…

В статье анализируются нормативно-правовые основы управления операционным риском в коммерческих банках и применение современных аналитических (статистических и основанных на машинном обучении) моделей. Проведён сравнительный анализ подходов Базельского комитета к расчёту требований к капиталу под операционный риск и введённого с 2023 года единого Стандартизированного подхода к измерению (SMA). На основе международных исследований обобщены модели машинного обучения, применяемые для классификации операционных потерь, выявления мошенничества, кибербезопасности и противодействия легализации доходов, а также показатели их точности. С учётом того, что само внедрение искусственного интеллекта может выступать новым источником операционного риска, предложен четырёхуровневый аналитический механизм, включающий блоки данных, классификации, специализированного выявления и управления.

Open source framework for representing and manipulating geometric models

Vibe Researching is an agent skill, methodology, and reproducible-research framework designed for researchers in humanities and social-science disciplines (philosophy, political theory, comparative ju…

Vibe Researching is an agent skill, methodology, and reproducible-research framework designed for researchers in humanities and social-science disciplines (philosophy, political theory, comparative jurisprudence, and institutional analysis) transitioning from qualitative research questions to empirical quantitative analysis.While frontier AI systems rapidly generate technical scaffolding (folder structures, cleaning scripts, econometric model calls, and data visualizations), they risk quietly substituting default assumptions for the researcher's substantive theoretical judgments. Vibe Researching introduces mandatory epistemic checkpoints to maintain researcher intellectual sovereignty:Mandatory Gate (Step 1): Collaborative formalization of research_question.md (defining question, claim, model justification, core assumptions, falsification criteria, and reproducibility requirements) before any code is executed.Numbered-Script Pipeline (Step 3): Strictly ordered execution workflow reconstructing raw-to-processed provenance.Granular Decision Cards (Step 4): Transparent documentation of substantive modeling choices, alternatives, and sensitivity.Timestamped Research Log (Step 5): Session-by-session contemporaneous decision tracking.Anti-Default Constraint (Step 6): Prohibition of silent defaults; surfacing 2–3 defensible options with concrete empirical implications.The package contains the formal SKILL.md specification, blank project templates, and a complete worked demo empirical analysis examining income–lifespan associations with full diagnostic leverage testing.Research Stance: Challenging the unchecked power, epistemic asymmetries, and monopolized governance of non-state actors across frontier technologies.GitHub: https://github.com/giabaohuynhasu/vibe-researchingHugging Face: https://huggingface.co/datasets/giabaohuynhasu/vibe-researching

GWexpy is an open-source Python library that extends the GWpy ecosystem with workflow-level abstractions and I/O adapters tailored for detector commissioning and laboratory-scale experimental diagnost…

GWexpy is an open-source Python library that extends the GWpy ecosystem with workflow-level abstractions and I/O adapters tailored for detector commissioning and laboratory-scale experimental diagnostics. GWexpy provides TimeSeriesMatrix containers, commissioning-oriented analysis primitives including transfer-function estimators and coherence-ranking utilities, and lossless format adapters for common experimental data formats.