OpenThesis 2.5.0

What's Changed

English

• Added automatic keyless market snapshots with dual-source quotes, caching, validation, circuit breaking, and bounded concurrency.
• Added reverse DCF using an auditable FCFE proxy against equity market value, with FX-date handling, stale-data protection, and dual-listing safeguards.
• Closed the 2.5.0 QA legacy items for recovery state, report projection, bounded synthesis, and tri-lingual market status.
• Added official OT package updates, secure section-repair prompting, regression coverage, and Windows packaging metadata.
• Included an unsigned Windows x64 portable test package (unsigned-test).
• Portable artifact SHA-256: EF081A576C300FE57E6AFA3FB51E0255F55AC4376171FBB13DC35DBB7DB4C455.

中文

• 新增免 Key 自动行情快照，支持双来源报价、缓存、校验、熔断和有界并发。
• 新增可审计的反向 DCF，以 FCFE proxy 对权益市值，并处理汇率日期、陈旧行情和双重上市安全规则。
• 关闭 2.5.0 QA 遗留的恢复状态、报告投影、有界合成和三语行情状态问题。
• 更新官方 OT 包、安全的章节修复提示词、回归测试和 Windows 打包元数据。
• 提供无签名 Windows x64 便携测试包（unsigned-test）。
• 便携包 SHA-256：EF081A576C300FE57E6AFA3FB51E0255F55AC4376171FBB13DC35DBB7DB4C455。

OpenThesis 2.4.2

What's Changed

English

• Fixed derived-liability semantics for issuers with non-controlling interests.
• Kept the official Liabilities-first path and made fallback derivation auditable with explicit formulas and source facts.
• Added KO 2017–2025 regression coverage and cross-period, cross-currency, and conflict-path rejection tests.
• Included an unsigned Windows x64 portable test package (unsigned-test).
• Portable artifact SHA-256: E7A1C9F9E17F5A267036149DB97BBECC8EA2A2D86564E6D748AA861686BE1EB0.

中文

• 修复含少数股东权益发行人的派生负债语义问题。
• 保留官方 Liabilities 优先路径，并通过明确公式和来源事实实现可审计的兜底派生。
• 新增可口可乐 2017–2025 回归覆盖，以及跨期、跨币种和冲突路径拒绝测试。
• 提供无签名 Windows x64 便携测试包（unsigned-test）。
• 便携包 SHA-256：E7A1C9F9E17F5A267036149DB97BBECC8EA2A2D86564E6D748AA861686BE1EB0。

OpenThesis 2.3.0

What's Changed

English

• Added a single financial recognition coordinator for initial research, retries, and full rebuilds.
• Added bounded parallel PDF processing, hard timeouts, cancellation, content-addressed parse caching, single-flight, and atomic publication.
• Added secure cloud visual completion with deterministic task IDs, minimal page uploads, resumability, and canonical quality-gate merging.
• Added cold/warm benchmark validation, cross-market golden-corpus regression coverage, and tri-lingual per-filing progress UI.
• Included an unsigned Windows x64 portable test package (unsigned-test).
• Portable artifact SHA-256: 46027BB2676C2D354409C006B1F9F3CA116CD8AD1974009D44AEE78E270E9AA3.

中文

• 新增统一财报识别协调器，统一初次研究、重试和完整重建路径。
• 新增有界并行 PDF 处理、硬超时、取消、内容寻址解析缓存、single-flight 和原子发布。
• 新增安全云端视觉补全，支持确定性任务 ID、最小页面上传、断点恢复和统一质量门合并。
• 新增 cold/warm 性能基准、跨市场 Golden Corpus 回归覆盖和三语逐报告进度界面。
• 提供无签名 Windows x64 便携测试包（unsigned-test）。
• 便携包 SHA-256：46027BB2676C2D354409C006B1F9F3CA116CD8AD1974009D44AEE78E270E9AA3。

OpenThesis 2.2.1

What's Changed

English

• Replaced company-specific patches with a unified financial fact compiler and quality gate.
• Added cross-market five-year golden-corpus validation, dual-equity semantics, conflict auditing, and safe bounded retries.
• Updated research, reporting, growth evidence, tri-lingual desktop UI, packaging, and runtime verification.
• Included an unsigned Windows x64 portable test package (unsigned-test, NotSigned).
• Portable artifact SHA-256: A425A8EACFBACA08C2403929622BE499171FA78E63BBF8453DB6B657D5D3E847.

中文

• 以统一财务事实编译器和质量门替代逐公司补丁。
• 新增跨市场五年 Golden Corpus 验证、双权益语义、冲突审计与安全有限重试。
• 更新研究、报告、增长证据、三语桌面端界面、打包和运行时验证。
• 提供无签名 Windows x64 便携测试包（unsigned-test、NotSigned）。
• 便携包 SHA-256：A425A8EACFBACA08C2403929622BE499171FA78E63BBF8453DB6B657D5D3E847。

OpenThesis 2.1.0

OpenThesis 2.1.0

简体中文

OpenThesis 2.1.0 将财报研究流程收敛为可复核、可恢复的确定性工作流：从采集、证据边界到报告结论，都保留清晰的来源与状态。

• 财报采集提升性能，并采用有界并发控制资源使用。
• 支持非自然财年与连续性处理，减少跨期比较中的断点。
• 执行严格同比与字段级质量校验；无来源数字不进入确定性报告。
• 对失败或缺失的财报节点执行重试，并支持确定性恢复，不消耗模型 Token。
• 通过 EvidenceSnapshot 保留证据边界，分离结论周期与证据历史，并兼容 .ot 研究包。

OpenThesis 不自动交易，也不保证收益；它用于组织、验证和复核研究证据。

English

OpenThesis 2.1.0 turns financial-report research into a reviewable and recoverable deterministic workflow, with explicit provenance and state from collection through evidence boundaries to conclusions.

• Faster financial-report collection with bounded concurrency for controlled resource use.
• Support for non-calendar fiscal years and continuity across reporting periods.
• Strict year-over-year checks and field-level quality validation; unsupported numbers do not enter deterministic reports.
• Retry failed or missing financial-report nodes with deterministic recovery that does not consume model tokens.
• EvidenceSnapshot preserves evidence boundaries, separates conclusion cycles from evidence history, and remains compatible with .ot research packages.

OpenThesis does not trade automatically or guarantee returns; it organizes, validates, and preserves research evidence.

OpenThesis 2.0.1

OpenThesis 2.0.1 (2.01)

版本标识 / Version: 2.0.1（部分场景简写为 2.01 / sometimes written as 2.01）

简体中文

OpenThesis 2.0.1 让模型配置与 .ot Studio 工作流更稳定、更易验证。

• 内置离线模型候选，合并在线目录，并支持手动模型兜底。
• 支持连接级和已配置模型级测试，由用户主动触发。
• API Key 通过 Windows Credential Manager 本地保存，不写入 SQLite、报告、日志或 .ot 包。
• .ot Studio 提供更一致的操作按钮、参数提示，以及确定性编译、导入校验、内容哈希和重启持久化。
• 兼容用户已安装的 Ollama；OpenThesis 不捆绑 Ollama 或模型权重。

Windows x64 便携包：OpenThesis-2.0.1-windows-x64-portable.zip

SHA-256：19786C955CDD6463E9E1AC660DEF77A0FCE142F8A28F102A428289FCD625718B

边界：当前正式发行包为 Windows 版本；OpenThesis 不提供自动交易，也不保证投资收益。

English

OpenThesis 2.0.1 makes model configuration and the .ot Studio workflow more dependable and easier to verify.

• Includes offline model candidates, stable online-directory merging, and a manual-model fallback.
• Adds user-triggered connection-level and configured-model-level tests.
• Stores API keys locally through Windows Credential Manager, never in SQLite, reports, logs, or .ot packages.
• Improves .ot Studio controls and parameter guidance, with deterministic compilation, import validation, content hashing, and persistence after restart.
• Works with an Ollama installation already on the user's machine; OpenThesis does not bundle Ollama or model weights.

Windows x64 portable package: OpenThesis-2.0.1-windows-x64-portable.zip

SHA-256: 19786C955CDD6463E9E1AC660DEF77A0FCE142F8A28F102A428289FCD625718B

Boundary: the formal binary release is currently Windows-only; OpenThesis does not provide automatic trading or guarantee investment returns.

OpenThesis 1.5.1

OpenThesis 1.5.1

繁體中文

OpenThesis 1.5.1 擴充了完整的語言與本地化基礎，新增繁體中文介面、報告與等待狀態文字，並讓介面語言可跟隨系統或由使用者手動選擇。介面語言與研究報告語言可以獨立設定，舊版設定也會安全相容。

本版本同時強化財報證據流程：結構化來源與官方 PDF 解析會按期間、合併口徑、幣種、單位及證據位置驗證；品質不足或有矛盾的資料會隔離，不會交給模型補造。可選的雲端視覺財報備援只在本地解析失敗且使用者明確同意後上傳必要表格頁，最多 20 頁／10 MB，不包含本地模型或訓練流程。

報告投影、確定性財務計算、DCF、比較、增長與 HTML/Markdown 匯出均維持穩定的協定鍵與證據 ID，同時依語言顯示友善標籤。繁體中文非技術檢視不會顯示內部欄位、錯誤碼或機密；技術檢視仍可用於審計。API Key、SEC 聯絡信箱及視覺 Token 僅存於工作階段，不會寫入設定或報告。

English

OpenThesis 1.5.1 adds a complete language-registry foundation with a first-class Traditional Chinese interface, report vocabulary, progress messages, help content, and system/manual language selection. UI language and research-report language remain independent, with safe compatibility for existing settings.

The financial evidence path now validates structured sources and official PDF facts by period, consolidated scope, currency, unit, statement context, and evidence location. Insufficient or contradictory groups are quarantined and never handed to a model as facts. The optional cloud vision fallback is available only after local extraction fails and explicit user consent; it uploads only relevant financial pages, capped at 20 pages/10 MB, and does not train or bundle a local model.

Deterministic financial calculations, DCF, comparison, growth, and Markdown/HTML exports keep stable protocol keys and evidence IDs while presenting localized labels. Nontechnical reports hide internal fields, error codes, and secrets; technical mode retains audit details. API keys, SEC contact email, and vision tokens are session-only and are not persisted in settings or reports.

Download / 下載

Download the Windows portable archive from the project Releases page. Extract it and run OpenThesis\\OpenThesis.exe. The first launch does not call AI; choose a provider and start a study only when you are ready to send the selected research context to that provider.

從專案 Releases 頁面下載 Windows portable 壓縮檔，解壓後執行 OpenThesis\\OpenThesis.exe。首次啟動不會呼叫 AI；只有在您選擇提供者並開始研究後，選定的研究內容才會傳送給該提供者。

This is a test build. Packaging, platform runtime smoke tests, and release publication remain separate acceptance steps.

這是測試版；打包、平台 runtime smoke test 與正式發布仍是獨立的驗收步驟。

OpenThesis 1.4.1

OpenThesis 1.4.1

简体中文

本版本重点提升研究过程的可观察性、财务数据可靠性和报告恢复能力。

• 研究过程改用稳定阶段展示，并增加单调进度、真实子步骤和正计时；下载完成后会明确进入财报识别与校验阶段。
• 重构 A 股、港股与美股财务数据摄取：优先使用官方结构化数据，PDF 坐标表格作为可审计回退，并统一经过期间、口径、单位、来源和会计勾稽质量门。
• 支持由用户自行配置的云端视觉财报兜底；默认关闭，只上传本地识别失败且经用户逐次批准的财务表页，不训练或捆绑本地视觉模型。
• 修复比亚迪季度同比：保留上一年度相同期间的官方披露，严格执行 Q1 对 Q1、H1 对 H1、Q3 对 Q3，不与完整财年混算。
• 修复增长机会空章节：综合报告为空时回退独立增长研究产物；模型空响应可只重试增长阶段及最终综合，不重复下载财报或运行其他 Agent。
• 报告重试、阶段性状态与导出状态统一到单一状态区，消除提示重叠，并进一步清理非技术模式中的内部字段和证据 ID。
• Windows 版本继续提供免安装便携 ZIP；解压后运行 OpenThesis\OpenThesis.exe。AI 与云端视觉功能默认均不启用。

English

This release focuses on observable research progress, reliable financial data, and resilient report generation.

• Research now uses stable stages with monotonic progress, real sub-step counts, and an elapsed-time clock. Filing downloads transition explicitly into parsing and validation.
• A-share, Hong Kong, and U.S. financial ingestion has been rebuilt around official structured data where available, with auditable coordinate-based PDF tables as fallback and one shared period, scope, unit, provenance, and accounting-consistency quality gate.
• Users may configure an optional cloud-vision fallback. It is disabled by default and uploads only locally failed financial-statement pages after explicit per-request approval; no local vision model is trained or bundled.
• BYD interim growth now retains the prior-year same-period official filing and compares Q1 with Q1, H1 with H1, and Q3 with Q3—never an interim period with a full fiscal year.
• Empty growth sections now fall back to the independent growth artifact. Retryable model failures can rerun only growth analysis and final synthesis without redownloading filings or rerunning other agents.
• Report retry, partial-report, and export feedback now share one status area, removing overlap while further hiding protocol fields and evidence IDs from non-technical views.
• Windows remains available as a no-install portable ZIP. Extract it and run OpenThesis\OpenThesis.exe. AI and cloud vision are both disabled by default.

OpenThesis 1.3.0

OpenThesis 1.3.0

中文

• 更可靠地摄取 A 股、港股和美股官方财报，优先结构化来源并保留官方文档证据。
• 按报告期间、合并/母公司口径和币种分组验证；异常组隔离，缺失字段不会伪装成 0。
• 核心质量不足时在 AI 调用前明确失败并关闭，避免不完整事实进入研究流程。
• 修订报告优先；表格单位只在当前报表或已验证续表内生效，不跨无关页面污染。
• 季度/中期报告与全年报告分离，仅进行同期比较。
• 报告字段本地化，技术详情可隐藏，结论按置信度分组展示。
• 并行研究最多两个 agent；临时失败可单独重试。
• 新研究失败时不再继续显示旧公司的报告。

下载 OpenThesis-1.3.0-windows-x64-portable.zip，解压后运行 OpenThesis\\OpenThesis.exe。默认不调用 AI。

English

• More reliable official financial-report ingestion for mainland China, Hong Kong, and US markets, with structured data preferred and document evidence retained.
• Facts are validated by fiscal period, consolidated/parent scope, and currency; bad groups are quarantined and missing values are never treated as zero.
• Insufficient core coverage fails closed before any AI call.
• Revised filings are preferred, and table units cannot leak across unrelated pages.
• Quarterly/interim reports stay separate from full-year reports and use like-for-like comparisons.
• Report fields are localized, technical details can be hidden, and conclusions are grouped by confidence.
• Parallel research uses at most two agents, with isolated retry for transient failures.
• A failed new research run no longer leaves the previous company report visible.

Download OpenThesis-1.3.0-windows-x64-portable.zip, extract it, and run OpenThesis\\OpenThesis.exe. AI calls are disabled by default.

OpenThesis 1.2.3

OpenThesis 1.2.3

中文

此版本重点提高中国 A 股财报研究的可靠性，并完善报告阅读与本地历史管理。

• 巨潮资讯的完整年度报告与季度、半年度公告分开查询，A 股公司优先使用本市场完整年报。
• “综合收益”不再被识别为营业收入；负收入、极端年度变化和资产负债异常会在 AI 调用前被拦截。
• 重新解析同一披露文件时事务性替换旧事实，避免错误缓存继续影响新研究。
• 非技术模式递归隐藏 Evidence ID、Agent ID、机会 ID 和响应元数据，并本地化常见结构化字段。
• 技术详情按钮加入明确的选中状态。
• 研究历史支持安全删除；用户手动保存的投资逻辑不会随自动研究记录删除。
• Windows 便携版解压即可运行，不显示命令行窗口。

下载 OpenThesis-1.2.3-windows-x64-portable.zip，完整解压后运行
OpenThesis\OpenThesis.exe。首次启动默认不调用 AI。

English

This release focuses on more reliable China A-share filing research and improves
report reading and local history management.

• CNInfo full annual reports are queried separately from quarterly and interim
  announcements, with native A-share annual reports preferred.
• Comprehensive income can no longer become revenue. Negative revenue, extreme
  year-over-year changes, and balance-sheet anomalies stop the run before any AI call.
• Re-parsing an official filing transactionally replaces stale extracted facts.
• Non-technical reports recursively hide evidence, agent, opportunity, and response
  protocol identifiers while localizing common structured fields.
• The technical-details control now has an unmistakable active state.
• Research history can be deleted safely while preserving theses saved manually by users.
• The Windows portable build runs after extraction without opening a console window.

Download OpenThesis-1.2.3-windows-x64-portable.zip, extract it completely, and run
OpenThesis\OpenThesis.exe. AI remains disabled by default on first launch.