v0.5.0

v0.5.0 — 期权决策研究平台

v0.5.0 将离线学习、研究简报、证据复核和开源验证流程连接起来。面向 BTC 期权研究者，
它回答当前输入支持哪些研究结论、为什么被阻断、下一步需要什么证据。
research_only=true 与 execution_allowed=false 保持不变。

这是本版本的交付说明。发布状态与最终验收记录见
本轮交付记录；只有 GitHub Release
中实际存在的资产才是已发布产物。

使用者可见变化

• crypto-options-report demo 默认进入离线学习导览，先理解有限风险结构，再查看包内真实快照。
  教学点数与实际研究数据、证据提升分开。
• 研究简报、工作台、公开版与 Chrome 侧栏统一解释过期、阻断和评估时刻；历史通过不表示当前资格。
• 加载失败、无效 JSON、合同异常和渲染错误提供恢复入口；信号/序列未配置与样本不足分别说明。
• 新增 完整决策研究指南 和 离线复核案例，
  覆盖公开采集、报告、证据、过期恢复和可复核记录。
• 策略卡要求数字成本和模型身份，最低净权利金真实扣除入场费、滑点与分腿预算；
  损失字段明确是 payoff 边界加冻结成本预算。交割费绝对上界未验证，所有卡片最高为 WATCH。
• 历史路径不再按该路径未来收益推导“相似状态”权重。缺少路径开始前的状态证据时，
  使用 unconditioned_uniform 并明确相似度条件不可用。
• 精确结构历史回放纳入按规则计算的逐腿交割费，并明确风险分母未包含交割费，因此净 R 可能小于 -1。
• 路径不利变动按实际腿的风险方向计算，并区分短腿行权价穿越、价格阈值代理与未实现的动态 Delta。
  压力场景覆盖双向冲击，明确为确定性假设，其权重不代表校准概率。

维护者可见变化

• 分离带类型的分析输入、不可变求值、五组准入条件与惰性旧报告投影，保留公共调用入口。
• 记录输入摘要和构建身份；源码与安装包的来源可区分，Git 状态观测不冒充签名。
• 将公开 JSON 合同集中为固定 OpenAPI/schema，发布前校验允许字段与产物结构。
• 统一 python tools/verify.py：源码检查、Web 构建、公开边界、扩展、最终 wheel 安装和浏览器流程。
  --quick 是明确的子集，不能代替完整验收。
• 对齐 Python、Web、扩展和发布资产的版本，提供无需网络的固定输入复现工具。

安装与升级

从 GitHub Releases 下载此版本的 wheel、
Chrome 扩展 ZIP 与 SHA256SUMS。将资产保存在同一目录，先核对本机摘要与校验和文件的对应行：

Get-FileHash -Algorithm SHA256 .\*.whl, .\*.zip
Get-Content -LiteralPath .\SHA256SUMS

在干净环境安装 wheel（将占位路径替换为实际文件名）：

python -m venv .venv
.\.venv\Scripts\python.exe -m pip install --no-index --no-deps .\<downloaded-wheel>.whl
.\.venv\Scripts\python.exe -m crypto_options_report.cli demo

wheel 包含前端和脱敏演示资源，安装后无需 Node、密钥、网络或仓库测试文件。开发源码请使用
贡献指南 安装受约束的开发依赖并重新构建前端。

v0.4.0 用户必须重新生成报告

原始市场快照可以重新计算，CLI 与公共分析调用入口保留；已有策略卡并非完全向后兼容。
strategy_brief.v1 现在要求 entry.cost_model_id、entry.cost_config_hash、完整数字
entry.cost_breakdown，以及明确的 risk.max_loss_basis 和 delivery_fee_upper_bound_verified。
v0.4.0 仅带 bool 含费标记的旧卡会被拒绝，这是有意收紧合同。

同时升级 Python 引擎、前端与 Chrome 扩展，从保留的原始输入重新生成报告和公开工件。
不要原地修改旧报告的 bool 或补零金额。预测的 selection_binding_key 还绑定成本模型、配置、
预算及净权利金；旧 CALIBRATED 产物不能直接复用，即使腿和到期日相同也会因缺少成本绑定退役。
旧预测文件仍可读取和独立验证；退役指它在当前策略简报中不再保持 CALIBRATED，
需要重新生成绑定当前成本与净权利金的校准证据，并非把历史文件当作损坏文件。

构建和输入身份调整会改变派生 ID 与摘要。用同一版本、输入和时钟生成新复核记录，保留旧记录
用于解释差异。Chrome 扩展解压到新目录并重新加载，避免混用旧 JS 与新 manifest。

信号与序列工件升级

/research/signal 和 /research/series 现在验证 research_only=true、允许的 schema 与合法时间，
并拒绝任意嵌套深度出现的 execution_allowed、manual_execution_allowed、order_instruction、
order_instructions 或 recommended_size 字段。执行字段即使值为 false 也不属于这两类研究工件。

由当前 CLI 生成的 measured、preflight、blocked 和 demo 工件兼容，excluded_snapshots
等来源排除证据保留。旧手工文件缺少研究标记或合同字段时，使用当前 CLI 重新生成，不手工补标记
来绕过校验。例如：

crypto-options-report validate-signal --preflight `
  --snapshot-dir artifacts/snapshots/btc-series `
  --underlying-history-fixture artifacts/history/btc-daily.json `
  --output artifacts/reports/signal-preflight.json --compact

crypto-options-report series-history --snapshot-dir artifacts/snapshots/btc-series `
  --output artifacts/reports/series-history.json --compact

随后将 --signal-artifact 与 --series-artifact 指向新文件。样本不足会产生可解释的阻断工件，
无需制造统计值；文件读取失败和样本不足仍是不同状态。

明确限制

可靠策略卡允许为空。历史胜率只在 VALIDATED 后出现，精确策略预测区间只在 CALIBRATED
后出现。本版本没有补造 cohort、提升模型或授权实盘；软件测试与离线案例不证明市场 edge。
公开静态站是否持续产出合格版次，需要独立的采集、监控和发布验收。

English summary

Version 0.5.0 connects an offline learning tour, the options research brief,
evidence inspection, recovery, and reproducible open-source verification.
Typed analysis inputs and lazy compatibility projection retain public entry
points; public artifacts are checked against fixed schemas. Full local
verification includes the final installed wheel and browser journey.

The platform remains research-only. Missing or stale evidence blocks, historical
rates require VALIDATED, forecast intervals require CALIBRATED, and execution
is always disabled. Cards remain WATCH: modeled costs are not an absolute
fee-inclusive loss cap. Version 0.4.0 boolean-only cost cards no longer validate;
upgrade the engine, frontends, and extension together and regenerate reports.
Old forecasts do not satisfy the new cost and net-credit selection binding.
They remain readable for historical inspection but retire in the current brief
until appropriately rebound evidence exists. Signal/series artifacts now require
research_only=true, a supported schema, and valid timestamps; execution and
sizing fields are rejected at every depth. Regenerate old hand-authored files
using the current CLI. Measured, preflight, blocked, and demo artifacts remain
supported, including snapshot-exclusion evidence.
See the decision workflow
and the delivery record above for the actual verification and publication status.

v0.4.0

LensOS Option v0.4.0

Released 2026-08-30 from the verified public v0.1.0 baseline.

This version integrates the v0.2 market/strategy brief, v0.3 aligned historical
replay, and v0.4 exact-strategy forecast calibration contract. It is a research
release, not a claim that a promoted model or live trading service exists.

One-screen strategy brief

The canonical strategy_brief.v1 contract answers, in one fail-closed surface:

• the current BTC option-market regime and whether reliable opportunities exist;
• zero to three deduplicated, finite-risk structures: Bull Put Credit Spread,
  Bear Call Credit Spread, and Iron Condor;
• exact one-unit legs, expiry/DTE, executable minimum net credit based on short
  bid and long ask, frozen costs, maximum loss, and cancel conditions;
• aligned historical status and exact-strategy forecast status without exposing
  success rates before their respective evidence gates pass.

If no candidate passes every gate, the canonical result is NO_TRADE and the UI
states “今日暂无可靠策略”. execution_allowed is permanently false.

The same brief is projected into the internal Evidence Console, the public
published surface, and the Chrome side panel. Missing runtime provenance,
demo/replay data, stale publication state, or fallback payloads cannot present
themselves as live.

Evidence discipline

• Relative value and absolute expected value remain separate evidence families;
  no blended score was introduced.
• Negative post-cost EV, a better opposite direction, no edge at the touch,
  unbounded/unknown loss, missing/stale/crossed/asynchronous quotes, or unit
  mismatch is a hard rejection.
• Historical metrics are visible only for history.status=VALIDATED, after a
  replay aligned on structure, direction, DTE, leg selection, execution, fees,
  and exit rules. A stable protocol binding prevents results from being reused
  after selection, fill, settlement, or replay-policy drift.
• Forecast intervals are visible only for forecast.status=CALIBRATED, from a
  non-expired exact-scope artifact. Input drift, scope mismatch, expiry, or
  out-of-sample deterioration demotes the result automatically. Exact expiry
  and leg identity are bound separately from changing quotes and timestamps;
  legacy unbound evidence is retired fail-closed.
• Production cvar_95_usdc path-risk evidence is normalized without weakening
  the positive finite-risk gate. Fallback candidates retain each source leg's
  quote timestamp; missing, stale, or desynchronized timestamps are never
  replaced with the report-generation time.
• Artifacts use canonical JSON and deterministic SHA-256 identities.

Deliberately unavailable evidence

No cohort, promotion, or calibration statistic was fabricated for this release.
The pre-existing frozen history boundary covers Call Credit Spread only. Bull
Put Credit Spread and Iron Condor therefore remain INSUFFICIENT or
EXPLORATORY until each has its own frozen protocol and future holdout. Forecasts
remain UNAVAILABLE or SCREENING until a matching, prospectively valid cohort
produces a promotable artifact. The UI renders these states as “暂不可用”.

Distribution status

Package and extension metadata are 0.4.0. The GitHub Release provides the
Python wheel, unpacked Chrome-extension ZIP, and SHA256SUMS; all three web
targets can also be built from source. The release also incorporates the
approved dependency patch set for the Python toolchain, Web build/type tooling,
and Python 3.14 slim container digest after the same release gates pass.

中文摘要

v0.4.0 把 v0.2 市场与策略简报、v0.3 同构历史回放、v0.4 精确策略校准状态机统一到
strategy_brief.v1。首页只给 0–3 张有限风险、可复核的一单位策略卡；所有硬门禁未过时
明确显示“今日暂无可靠策略”。历史与预测数字分别只在 VALIDATED / CALIBRATED 后出现，
execution_allowed 永远为 false。Bull Put 与 Iron Condor 尚无未来冻结 cohort，预测也
尚无可提升的真实精确策略样本，因此当前正确交付是完整状态机、artifact、自动降级和
“暂不可用”，不是制造胜率。

v0.1.0

LensOS Option v0.1.0

Released 2026-08-29.

This is the first public release of the Crypto Options Research Console: a
fail-closed, evidence-first research tool for inspecting crypto option-selling
candidates. It is a research and methodology release, not a validated trading
signal and not an execution system.

Install and run the offline demo

Python 3.12 or newer is required.

python -m pip install https://github.com/Lens-less/LensOS-Option/releases/download/v0.1.0/crypto_options_research_console-0.1.0-py3-none-any.whl
crypto-options-report demo

The installed demo binds to 127.0.0.1, uses a redacted snapshot packaged in
the wheel, and needs no Node.js, API key, repository fixture, or network access
at runtime. The UI remains explicitly marked DEMO, SNAPSHOT DATA, and
RESEARCH_ONLY · NO_TRADE.

Highlights

• A wheel-installed, credential-free local demo with honest snapshot labeling.
• A read-only Web research workbench and Evidence Console whose visible local
  navigation targets, legal pages, and license links are served by the Python
  API.
• A Chrome 114+ Manifest V3 research companion packaged as an unpacked-extension
  ZIP for personal, loopback-only use.
• Deterministic replay, evidence provenance, explicit missing-data states, and
  fail-closed quality gates across the CLI, HTTP API, and browser surfaces.
• Bilingual README guidance, public contribution and conduct policies, private
  vulnerability reporting, and Apache-2.0 / CC BY 4.0 licensing.

Release assets

Asset	Purpose
crypto_options_research_console-0.1.0-py3-none-any.whl	Platform-independent Python package and offline demo
lensos-option-chrome-extension-v0.1.0.zip	Chrome companion; extract and load the folder as an unpacked extension
SHA256SUMS	SHA-256 digests for both distributable files

On a system with sha256sum, verify downloaded assets with:

sha256sum --check SHA256SUMS

Deliberate boundaries

• No order endpoint, automated or semi-automated execution, position sizing, or
  trade instruction is included.
• Trusted output never exceeds an EntryAdmissionDecision with
  execution_allowed=false.
• Pre-registered signal validation is still accumulating at 1/8 settled
  cohorts. This release makes no claim that a trading signal is validated.
• Hosted-site publishing remains SUSPENDED; the release does not imply that a
  public market-data site or external monitoring service is live.

See the detailed changelog,
security policy,
and documentation map.

中文摘要

v0.1.0 是 Crypto Options Research Console 的首个公开版本。它提供可从 wheel
独立运行的离线演示、只读 Web 工作台、Evidence Console 与 Chrome 研究伴侣；运行演示
不需要 Node、API 密钥、仓库 fixture 或外网。所有界面继续明确标注演示/快照数据，并
保留 RESEARCH_ONLY · NO_TRADE 与 execution_allowed=false 的真实边界。

本版本发布的是可审计工具与方法，不是已经验证的交易信号，也不包含任何下单、手数或
自动执行能力。托管站点发布仍保持 SUSPENDED。