这套系统拒绝“应该没问题”这种模糊判断，风险必须被规则命名。 This system rejects the vague comfort of “probably fine”; risk has to be named by rule.
// IMMUNE SYSTEM + DETERMINISTIC GOVERNANCE
DATA GOVERNANCE PIPELINE
数据如何被安检 How Data Gets Cleared
DataVault 的价值不只是“帮 AI 扫描文件”，而是把数据治理从临场发挥变成 SOP。它让每个 agent 在面对文件、隐私和上链注册时都遵守同一套确定性规则，而不是每次都即兴判断。 DataVault matters not because it “helps AI scan files,” but because it turns data governance from improvisation into SOP. It makes every agent follow the same deterministic rules when dealing with files, privacy, and on-chain registration instead of making it up each time.
// ORACLE AI 可以参与判断价值，但不该参与决定什么叫“也许安全”。 AI may help judge value, but it should not decide what counts as “probably safe.”
扫描、分类、隐私检查全部发生在本地，链只接收被确认过的结果。 Scan, classify, and privacy checks all happen locally; the chain only receives confirmed results.
真正上链前必须显式 `--confirm`，把治理责任从自动化里重新拉回到人手里。 Real registration still requires explicit `--confirm`, pulling governance responsibility back into human hands.
标准流程 STANDARD SOP
$ datavault scan ~/Documents $ datavault classify $ datavault privacy $ datavault report --format json # user reviews the result $ datavault register --confirm
四阶段管线 FOUR-STAGE PIPELINE
01 SCAN
扫描与指纹 Scan and fingerprint
sha256 · size · mime · mtime 第一步只做事实记录。递归遍历目录、生成 SHA-256 指纹、提取元数据，让每个文件先拥有不可争议的身份。 The first step only records facts. It traverses directories, generates SHA-256 fingerprints, and extracts metadata so every file first gains a non-negotiable identity.
02 CLASSIFY
分类与归档 Classify and inventory
59 extensions · 9 categories 第二步用确定性映射而不是模型猜测，把文件放进稳定分类系统。不同 AI session 不应对同一个文件得出不同资产身份。 The second step uses deterministic mapping rather than model guesswork to place files into a stable classification system. Different AI sessions should not derive different asset identities for the same file.
03 CHECK
隐私与风险检查 Privacy and risk check
6 pii patterns · 5 risk levels 第三步把“可能安全”替换成可复现的 PII 检查。邮箱、手机号、身份证、信用卡、IP、API key 都按规则匹配，而不是靠模型心情。 The third step replaces “probably safe” with reproducible PII inspection. Emails, phone numbers, ID cards, credit cards, IPs, and API keys are matched by rule rather than model mood.
04 REGISTER
显式上链注册 Explicit on-chain registration
--confirm required 最后一步不是自动发生。只有 SAFE / LOW / MEDIUM 才能进入候选集，且需要显式 `--confirm` 才会把哈希与元数据送入 Oasyce 网络。 The last step never happens implicitly. Only SAFE / LOW / MEDIUM enter the candidate set, and explicit `--confirm` is required before hashes and metadata are sent into the Oasyce network.
治理规则 GOVERNANCE RULES
高风险永不上链 High-risk never goes on-chain
`HIGH` 和 `CRITICAL` 文件被硬编码拦截。这个规则不该依赖提示词、团队习惯或人类记性。 `HIGH` and `CRITICAL` files are hard-blocked. This rule should not depend on prompts, team habits, or human memory.
前三步全部本地执行 First three steps stay local
扫描、分类、隐私检查都是本地且免费的。链只接收最终确认过的结果，不接收你的原始文件。 Scanning, classification, and privacy checks are local and free. The chain only receives the final confirmed result, never the raw file itself.
每个判断都能复现 Every judgment is reproducible
相同输入、相同环境、相同结果。DataVault 不是为了替代 AI，而是为了给 AI 一个不会漂移的操作标准。 Same input, same environment, same output. DataVault is not there to replace AI, but to give AI a non-drifting operational standard.
生态交接 ECOSYSTEM HANDOFF
DataVault
先做扫描、分类、风控 Scan, classify, and gate risk first
oasyce-net
负责注册与调用流程 Handle registration and invocation flow
oasyce-chain
把产权、清算和收益分配写进共识 Write property, settlement, and payout into consensus
AI Buyer
调用经过治理的数据资产 Consume governed data assets
设计拆解 DESIGN BREAKDOWN
DataVault 被我定义成整个生态的“免疫系统”，而不是一个孤立工具。它的价值不在扫描文件，而在让数据进入经济系统之前先经过一道确定性的安检。 I define DataVault as the immune system of the ecosystem rather than an isolated utility. Its value is not file scanning itself, but giving data a deterministic checkpoint before it enters the economy.
页面按 stage / rules / ecosystem 三层组织，是为了同时说明它怎么工作、凭什么可信、以及它在更大产品系统里的位置。 The page is organized as stage / rules / ecosystem to explain how it works, why it can be trusted, and where it sits in the larger product system.
我保留命令行语言，但把命令行从“演示道具”变成 SOP 证据，用户能直接读懂系统每一步的责任边界。 I keep the command-line language, but shift it from demo prop into SOP evidence so users can directly read the responsibility boundary of each step.
界面继续坚持极度克制的控制台风格，让安全和合规感来自规则结构，而不是来自夸张警告色。 The interface keeps an extremely restrained console style so safety and compliance feel grounded in rule structure rather than exaggerated warning color.
唯一动势仍然留给 signal decode，让这页看起来像一个治理系统启动，而不是一张安全软件广告图。 The only motion stays in the signal decode so the page feels like a governance system booting, not a security software advertisement.