Skip to content

v2.6.6 Beta

Choose a tag to compare

View all tags
@Horace-Maxwell Horace-Maxwell released this 11 Jun 21:20
· 28 commits to main since this release
v2.6.6
ff69e89

Horosa 2.6.6 Beta

Beta release / Beta 测试版：this build is published for hands-on verification before it is promoted as the stable release.
Beta 测试版：本构建用于稳定版前的真实安装与使用检查；安装包、runtime 与 manifest 已完成签名、公证和端到端验证。

Horosa is a local-first metaphysics workstation for Apple Silicon, spanning Western astrology, Chinese traditional systems, AI-assisted analysis, and a notarized desktop delivery stack.
Horosa 是面向 Apple Silicon 的本地优先玄学工作站，覆盖西方占星、中国传统术数、AI 辅助分析，以及正式公证的桌面交付链路。

当前版本亮点 / Release Highlights

  • 当前发布版本 / Current release: v2.6.6 Beta

本版更新 / What's new in this release

v2.6.6 —— AI 报告打磨 + 桌面体验修复：主限法（Primary Directions）全面升级（显示窗精确化 + 宿命点应星 + 时间钥匙修真 + 年数上限 3000 + 世俗行集修复）；修复调用 Gemini 时采样参数封装错误导致的偶发 400「格式错误」；修复聊天点「发送」按钮时所提问题可能被 [object Object] 占位串顶替的错误；聊天高级参数（停止序列/惩罚/JSON 模式）对 OpenAI 兼容接口真正生效；AI 报告「思考档」新增「极高 / 最大」两档；界面缩放在重启后保持（不再每次打开都复位）；另含一批白屏防护与暗色可读性加固。主限法以外技法的命盘计算与 v2.6.5 完全一致。安装旧版的用户启动后自动收到 v2.6.6 更新提示。本版运行时组件已更新（2.6.6-runtime1），升级时自动下载新运行时。

  • 主限法全面升级：行星对显示窗改为单一判据（弧归一化前原值）、世俗（In Mundo）此前缺失的行星对方向行现全数列出；黄道向运新增宿命点（Vertex）应星行；Simmonite / Kepler / Brahe 时间钥匙改为逐盘真算太阳日速，并新增 Kündig 与 太阳弧（黄经）两把钥匙；推算年数上限 360 → 3000（跨圈方向按整圈复发自动延展）。默认设置（Alchabitius + Ptolemy + 100 年）行为不变。
  • Gemini 参数封装修复：调用 Gemini 生成时，temperature / maxOutputTokens 等采样参数此前可能漏到请求顶层（它们只应在 generationConfig 内），导致上游返回 400「格式错误」。本版组装请求时把这些键从顶层剔除（对照其它 provider 路径补齐防漏），消除该偶发报错。
  • 聊天「发送」按钮修复：点击圆形发送按钮时，所提问题可能被一个 [object Object] 占位串顶替后发出（按 Enter 发送不受影响），AI 因此收不到真实问题。本版修正按钮传参并加固为只接受文本，问题内容不再丢失。
  • 聊天高级参数真正生效：「停止序列 / 频率·存在惩罚 / JSON 输出模式」此前对 OpenAI 兼容类接口（OpenAI / DeepSeek / Moonshot 等）因接口家族判定有误而未实际下发，本版起全部生效；「思考档」对 gpt-5.5 / gpt-6 / gpt-7、o6 / o7 等新推理模型同样生效。
  • 报告生成错误透出：上游在流式输出中途报错时，此前该章节会带着半截内容被标为「完成」；现在按错误处理并自动重试，重试无果则明确标错（已生成的部分内容保留展示）。
  • 稳定性与暗色一批加固：阅读器 / 六爻卦辞 / 八字设置等对本地损坏配置自愈（不再白屏）；中点盘、农历月历、经纬度输入等空值/畸形值防护；排盘失败弹窗、更新按钮、寿主星选中态等暗色下可读性修复；多处列表渲染键稳定化（不再丢输入焦点）。
  • AI 报告「思考档」加高档位：思考档原本只有 关 / 低 / 中 / 高，本版新增 极高 / 最大。推理模型自动适配——OpenAI 系（reasoning_effort 仅认 low/medium/high）封顶为 high，Anthropic / Gemini 提高思考预算，并按输出上限 clamp 防止预算超额再触发 400。档位只作用于主章节生成（辅助节保持关闭），所选档位本地记忆、下次默认。
  • 界面缩放持久化：放大 / 缩小界面后，缩放值现写入偏好、重启自动恢复（此前每次打开都复位成默认大小）。
  • 运行时：本版后端组件（Java）已更新 → runtime 2.6.6-runtime1，升级时自动下载。

v2.6.6 — AI report polish + desktop UX fixes: a comprehensive Primary Directions upgrade (exact display-window semantics + Vertex significator + true per-chart time keys + 3000-year range + In-Mundo row-set fix); fixes an occasional upstream 400 ("format error") when calling Gemini caused by sampling params not being wrapped correctly; fixes the chat send button occasionally replacing your question with an [object Object] placeholder; chat advanced params (stop sequences / penalties / JSON mode) now actually apply on OpenAI-compatible providers; the AI report "thinking effort" selector gains two higher tiers (Extra-High / Max); the interface zoom now persists across restarts (no longer resets every launch); plus a batch of blank-screen guards and dark-mode readability fixes. Outside Primary Directions, every technique's chart computation is bit-for-bit identical to v2.6.5. Users on older builds receive v2.6.6 automatically on next launch. The runtime is updated this release (2.6.6-runtime1) and downloads automatically on upgrade.

  • Primary Directions upgrade: the planet-pair display window now uses a single criterion (the arc's pre-normalization value) and previously-missing In-Mundo planet-pair rows are now all listed; zodiacal directions gain Vertex significator rows; Simmonite / Kepler / Brahe time keys now compute the natal Sun's true daily motion per chart, with new Kündig and Symbolic Solar Arc keys; the year selector extends from 360 to 3000 (directions beyond one revolution recur per revolution). Default settings (Alchabitius + Ptolemy + 100 years) behave identically.
  • Gemini parameter-wrapping fix: when calling Gemini, sampling params such as temperature / maxOutputTokens could leak to the top level of the request (they belong only inside generationConfig), and the upstream returned a 400 "format error". This release strips those keys from the top level before sending (mirroring the guard already present on the other provider paths), eliminating the intermittent error.
  • Chat "send" button fix: clicking the round send button could replace your question with a literal [object Object] placeholder before sending (sending with Enter was unaffected), so the AI never received the real question. The button wiring is corrected and the handler hardened to accept text only.
  • Chat advanced params now take effect: stop sequences / frequency & presence penalties / JSON output mode were silently dropped for OpenAI-compatible providers (OpenAI / DeepSeek / Moonshot, etc.) due to a wrong protocol-family check; they now apply. The "thinking effort" tier also now covers newer reasoning models (gpt-5.5 / gpt-6 / gpt-7, o6 / o7).
  • Report errors surfaced: when the upstream errored mid-stream, the section used to be marked "done" with truncated content; it now retries automatically and is clearly marked failed if retries are exhausted (partial content stays visible).
  • Stability & dark-mode hardening: self-healing guards for corrupted local settings (reader / hexagram text / BaZi options — no more blank screens); empty/malformed-value guards (midpoints, lunar calendar, lon/lat inputs); dark-mode readability fixes (chart-service error dialog, update buttons, lifespan method pills); stable list keys (no more lost input focus).
  • Higher "thinking effort" tiers for AI reports: the selector previously offered Off / Low / Medium / High; this release adds Extra-High / Max. Reasoning models adapt automatically — OpenAI's reasoning_effort (which only accepts low/medium/high) is capped at high, while Anthropic / Gemini raise the thinking budget, clamped against the output limit so the larger budget can't re-trigger a 400. The tier applies to main-section generation only (auxiliary sections stay off); the chosen tier is remembered locally and pre-selected next time.
  • Interface-zoom persistence: after you zoom the interface in/out, the zoom factor is now saved to preferences and restored on restart (previously it reset to the default size every launch).
  • Runtime: the back-end component (Java) is updated this release → runtime 2.6.6-runtime1, downloaded automatically on upgrade.
  • v2.6.6 是桌面发布线的 beta 扩展版，重点覆盖新命法/卜法后端、本地数据管理、结构化 AI 导出、设置持久化与明暗主题。
  • v2.6.6 is a beta expansion of the desktop release train, focused on new traditional-method engines, local data management, structured AI export, persistent settings, and light/dark UI polish.
  • 新增并规范接入太乙、金口诀、皇极经世、五兆、太玄、荆诀、神易数、Kin Astro、七政四余、奇门等命法与卜法后端。
  • Added and normalized backend integrations for Taiyi, Jin Kou, Huangji/Wangji, Wuzhao, Taixuan, Jingjue, Shenyishu, Kin Astro, Qizheng, Qimen, and related specialty methods.
  • 三式合一中奇门与太乙固定走 kentang2017 后端口径，六壬保留现有本地六壬实现。
  • Sanshi United routes Qimen and Taiyi through the kentang2017 backend while LiuReng remains on the existing local LiuReng implementation.
  • 奇门和三式合一页面已移除后端不支持的月家奇门选项，不再静默回退到旧本地算法。
  • Unsupported Qimen month-chart selection was removed from Qimen and Sanshi surfaces instead of silently falling back to the old local calculation.
  • 管理命盘与管理事盘会保留新技法输入、标签、快照、后端原始结构化数据、JSON 导入导出与重开恢复行为。
  • Chart management and case management preserve new-method inputs, tags, snapshots, raw backend payloads, JSON import/export, and reopening behavior.
  • AI 导出直接读取结构化后端数据，并为每个支持技法、tab 与页面提供可勾选导出分段。
  • AI export reads structured backend data and exposes selectable export groups for each supported technique, tab, and page.
  • 用户设置、桌面窗口大小与必要 UI 选项会在关闭、重开和版本更新后继续沿用。
  • User settings, desktop window size, and necessary UI choices persist across close/reopen and app updates.
  • 启动控制台采用 Daily / Offline Ready / Failed 共用骨架，实时跟随后端进度，并统一使用新版星阙 icon。
  • The startup console now uses one shared Daily / Offline Ready / Failed skeleton, follows real backend progress, and consistently uses the new Xingque icon.
  • 窗口恢复已按首个可见帧验收，避免重开时先大后小或大小跳动。
  • Window restoration is verified from the first visible frame to prevent launch-size flashing or bouncing.
  • 全局明暗主题、下拉层、弹层、加载态、管理列表与导出控件已再次检查并打磨。
  • App-wide light/dark mode contrast, dropdowns, overlays, loading states, management lists, and export controls were audited and polished.
  • 保留此前奇门遁甲一致性与桌面交付修复。
  • Previous Qimen Dunjia parity and desktop delivery fixes remain preserved.

下载 / Download

  • 推荐安装包 / Recommended download: Horosa-Installer-macos-arm64-offline.pkg
  • 适合普通用户、中国大陆、弱网和离线转发场景。
  • Best for ordinary users, weak-network environments, and offline forwarding.
  • 当前平台目标 / Platform target: Apple Silicon (arm64)

安装步骤 / Installation

  1. 下载 / Download Horosa-Installer-macos-arm64-offline.pkg
  2. 双击 .pkg 开始安装 / Double-click the .pkg to start installation
  3. 安装完成后直接打开 /Applications/星阙.app / Open /Applications/星阙.app after install
  4. 如系统提示安全确认，请在“系统设置 -> 隐私与安全性”中放行 / If macOS asks for confirmation, allow it in System Settings -> Privacy & Security

仓库入口 / Repository Entry Points

自动更新与桌面交付 / Auto-Update And Desktop Delivery

  • 自动更新继续依赖 horosa-latest.json、桌面安装包与独立 runtime 资产，不要求改变客户端协议。
  • Auto-update continues to rely on horosa-latest.json, desktop release assets, and the separate runtime artifact without changing the client protocol.

已知限制 / Known Limitations

  • 仓库中的一部分 legacy 模块仍保留较强的本地运行和内部依赖假设。
  • Some legacy modules in the repository still assume a strongly local runtime environment and internal dependency chain.

技术资产 / Technical Assets

此 Release 中其余资产是安装器与自动更新器使用的内部支持文件，普通用户可以忽略。
The remaining assets in this release are internal support files for the installer and auto-updater. Ordinary users should ignore them.

Assets 5
Loading