01
O/S/C/AOSCA definitionOSCA 定义
Plain text that answers four questions about a job: objective, steps, interfaces, timing. Drop it into a runtime and it's a working AI agent.一份纯文本，写清一项工作的四个答案：目标、步骤、接口、时机。放进运行框架，它就是一个能干活的 AI Agent。
OSCAWARE / OPEN SOURCE
AI workflows that evolve with human feedback.让 AI 工作流程，
随人的反馈不断进化。
The more you use it, the more it fits — because it stays aligned with your world as it changes.越用越顺手——因为它始终贴合你当下的环境。
AI WORKFLOW
HUMANFeedback反馈
NEW PATH ↗
01 / WHY
Most AI deployed inside organizations today is chatbots and scattered one-step skills: passive, fragmented, unloved — and it doesn't change the outcome of the work.
今天落地在组织里的 AI，多半是问答机器人和零散的单步小技能：被动、碎片、没人爱用，也不改变工作结果。
What's missing isn't a stronger model — it's that no one brings in how the work should actually be done, and nothing keeps it.
缺的不是更强的模型，而是没有人把「怎么干才对」带进去，也没有东西把它留下来。
02 / WHAT
What Oscaware isOscaware 是什么
Oscaware is a definition for AI cognitive workflows (OSCA), a contribution ledger continuously supplied by human experts, and a distillation mechanism that turns human feedback into workflow adjustments at runtime. Oscaware 是一套 AI 认知工作流程的定义（OSCA）、一本由人类专家持续供给的贡献账本，和一套在运行中把人的反馈变成流程调整的蒸馏机制。
02
LEDGERContribution ledger贡献账本
A set of entries distilled from feedback. Each records who gave it, which case it came from, and how many times it's been confirmed.反馈沉淀成的条目集。每条写明：谁给的、依据什么案例、被验证过几次。
03
DISTILLDistillation蒸馏机制
The process that turns scattered feedback into ledger entries. AI drafts the generalization; a human always makes the call.把零散反馈变成账本条目的工序。归纳交给 AI，拍板永远在人。
What a ledger entry looks like · a real entry from the public ops-diagnosis sample pack账本条目长这样 · 运维诊断公开样例包中的一条真实条目
Contribution ledger · entry (JUDGMENT)贡献账本 · 条目（JUDGMENT）
J-0417
If a travel-expense swing overlaps the unit's scheduled-maintenance window, treat it as normal fluctuation and leave it out of the report — unless the increase also exceeds the unit's three-year peak for the same account during maintenance periods.差旅费异动若与该单位检修计划期重叠，视为正常波动，正文不报——除非涨幅同时超过该单位近三年检修期同科目峰值。
From the public sample pack 出自公开样例包 examples/oper-diagnosis.osca/judgments/J-0417.yaml
03 / EXAMPLE
One example to grasp the ledger一个例子看懂账本
Supermarket fresh food / how a ledger entry is born超市生鲜 / 一条账本条目怎样长出来
The system follows its routine系统照常规办
This month 10 new SKUs arrive; the system routes them into near-expiry discounting and past-expiry write-off, and confirms with the store manager.这个月新上 10 个品，系统把它们排进临期打折和过期报废流程，并向店长确认。
The manager gives one piece of feedback店长给一条反馈
“No — 2 of these should go back to the supplier.”「不对，其中 2 个要退回供货商。」
Recorded; live next month入账，下月自动生效
From then on those 2 SKUs auto-route to supplier return — no need to say it again.从此这 2 个品自动走退供流程，不用再说第二遍。
One piece of feedback changed the workflow itself.一条反馈，改了流程本身。
A knowledge base is something you look up; a ledger acts on its own — when the conditions match, it fires.知识库是你去查它；账本是它自己生效——条件命中就出手。04 / HOW
How it runs怎么运转
RUN × FEEDBACK × EVOLVE
A human stands at the key points — not because people judge better than AI, but because only a person knows the outside world has changed, and can carry that change into the pipeline as a single piece of feedback. 关键位置要站人——不是因为人比 AI 更会判断，而是因为只有人知道外面的世界变了，并能把这个变化，用一条反馈带进流水线。
Runs as-is放进来就能跑
One text file is a fully running AI agent.一份文本，就是一个正常运行的 AI Agent。
Composable能互相调用
Agents can call and combine with one another.Agent 之间可以互相调用、互相组合。
Works with people能与人协作
It asks when it should; the calls that are a human's stay a human's.该问人的时候问人；该人拍板的，永远人拍板。
Evolvesmost important能进化最重要
It keeps adjusting itself in response to human feedback.根据人的反馈，不断调整自己。
“Evolve” is a neutral word: no promise it gets smarter or better — only that it stays a better fit, because it keeps aligning with your environment as it is now.进化是中性词：不承诺越用越好，只承诺越用越顺手——因为它始终在贴合当下的环境。
05 / FOR WHOM
Built for people who actually know the work为真正懂工作的人而做
Frontline experts一线专家
Learn nothing new: work as usual, edit as usual. Your experience gets recorded — under your name.不用学任何新东西：照常干活、照常改稿。你的经验被记下、记你的名。
↗Service providers & delivery teams服务商与交付团队
A recurring-revenue delivery model: the client's ledger grows thicker, the service goes deeper.一套能续费的交付模式：客户的账本越用越厚，服务越做越深。
↗Organizations & managers组织与管理者
Controllable, auditable, real; swap the model without starting over.可控、可审计、真落地；换模型不报废。
↗06 / TRUST
Trust built into the mechanism信任写进机制里
Controllable · Auditable · Yours可控 · 可审计 · 归你所有
Every conclusion is traceable每条结论可追溯
Back to who decided it, on what basis, and from which case.追到谁定的、凭什么、当时的案例。
●The ledger is plain text账本是纯文本
Printable, signable, yours to keep.可打印、可签名、归你所有。
●Permissions enforced by the system权限由系统硬管
Not on the AI policing itself; irreversible actions always need a human.不靠 AI 自觉；不可逆操作永远人拍板。
●Model swaps come with a checkup换模型有体检单
Re-verify the ledger entry by entry, spelling out what still holds and what needs patching.逐条重验账本，明明白白列出哪些还灵、哪些要补。
●The method has a pedigree方法有出处
Judgment capture follows Klein’s Critical Decision Method — five decades of naturalistic decision research. Provenance and mapping in the whitepaper below.判断采集依循 Klein 关键决策法（CDM）——五十年自然决策研究的方法论背书；出处与对照见下方白皮书。
●07 / OPEN STANDARD
OSCA is an open standard.OSCA 是开放规范。
OObjectWhat's the objective目标是什么
SStructureHow many steps分几步走
CConnectorWhere the data comes from数据从哪来
AAwareWhen to act什么时候动手
The spec text is open under CC BY 4.0; the tools and samples are open-source under Apache-2.0. We're opening it up; building your own platform on the same mechanism is where this is headed.规范文本以 CC BY 4.0 开放，配套工具与样例以 Apache-2.0 开源。我们正在把它开放出来；基于同一套机制构建你自己的平台，是这条路的方向。
v1.0 is out — a mechanism-verifiable release: the open spec (v0.4), a reference Runtime, and a replayable desensitized sample you can osca replay to watch a judgment move the output from before-edit to after-edit. Real-world validation is the 1.x journey.v1.0 已发布——机制可验：开放规范（v0.4）+ 参考运行框架 + 可回放脱敏样例（osca replay 亲眼看判断把输出从改前移向改后）。真实业务验证属 1.x 旅程。
OSCA Whitepaper v1.1 — the full rationale, including the methodological grounding: why we capture corrections instead of interviewing experts — built on Klein’s Critical Decision Method, five decades of naturalistic decision research.OSCA 白皮书 v1.1——完整的来龙去脉，含方法论出处：为什么采集纠错、而不访谈专家——基于 Klein 关键决策法（CDM），五十年自然决策研究的背书。