The workspace for technology assessment.
创新管线
GCRI · 阶段关口 · 示例界面,并非记录
按就绪等级
准备度标准
试点与关口
试点问题
- I-1试点站点的数据许可待核查9 天后到期
- I-2观察员协议已签署已完成8 天后到期
GCRI · 创新管线。 创新管线:示例名称和示意数值,并非记录。
每项评估背后的四个问题。
A fourth question runs through all three: what would make the technology fail, and has that been looked for?
成熟度如何?
Evidence on maturity in the intended setting, not in the most favourable one.
是否经过测试?
针对既定目标的独立测试,并报告测试条件。
What exactly is claimed?
能力声明区分为实测、估算和自称三类。
准备度等级:其衡量的内容以及未涵盖的内容。
技术成熟度等级从一级(观察到基本原理)到九级(系统在其运行环境中得到验证)。NASA 提出了这一概念;ISO 16290:2013 为空间 系统对其作出了定义,许多机构使用改编版本。成熟度等级描述的是在既定环境中的成熟程度。它并不说明成本、在不同场景下的安全性、社会接受度,或某个组织是否有能力运行该技术。
| 问题 | Evidence that answers it | Common gap |
|---|---|---|
| 它是否如所称那样有效? | Independent tests against a pre-registered plan, on held-out data or in a representative environment | Vendor benchmarks on data the system was tuned to |
| 在此适用吗? | 在目标情境中开展试验,涉及本地用户和本地条件 | 从不同的气候、语言或基础设施迁移而来 |
| 它取决于什么? | Bills of materials, supply and energy needs, skills required | 对单一供应商或服务的隐性依赖 |
| 它如何失效? | Adversarial and stress tests; incident records | Testing only the expected cases |
如实报告能力。
- Report performance with confidence intervals and on the population it will serve, not only on aggregate test sets.
- 检查数据污染:泄漏到训练集中的测试数据会抬高得分,这是 AI 基准测试中已知的问题。
- For prediction models in health and similar settings, follow reporting guides such as TRIPOD+AI (2024).
- Separate a demonstration, a pilot and sustained operation; each is a different claim.
The OECD’s Framework for Anticipatory Governance of Emerging Technologies of 2024 and parliamentary technology assessment offices, networked through EPTA, provide models of assessment that look beyond performance to values and institutions.
Assessment reports, not approvals.
An assessment records what was tested, how and with what result. It confers no status and is never a condition of procurement. Assessments are designed for the Labs Evidence series and, when they concern readiness for national deployment, the Foundry Build and Readiness series; evidence would follow NS-11 of the proposed Nexus Standards. Providers can submit test plans; independent researchers can submit replications.
能力如何报告。
- 已测得的能力连同工作负载、数据规模、硬件、配置、软件版本和区间一并发布
- 峰值、持续、可用和已分配容量分别报告
- 设备数量和规格属于描述性信息,而非应用成果
- 对能源、制冷和用水情况进行跟踪,实测数据与估算数据分开保存
- 测试中发现的失败和局限会被如实报告,而不是被删除
Testing is not certification.
An assessment describes what was tested and found. It is not accreditation, certification, endorsement or approval for use. Evaluation never depends on a provider’s contribution. Providers start at technology and infrastructure providers. See also responsible technology and the Labs Evidence report series.
Where this page’s references lead.
阅读本页时所对照的机构和文件。状态和版本会变化;每个链接都指向发布机构。
问题与解答。
供应商能否在营销中引用 GCRI 的评估?
谁支付评估费用?
是否发表负面结果?
GCRI 是否使用就绪度等级?
评估结果可以提出异议吗?
从本页贡献证据。
Reports 是联合体的季刊。您对该主题的了解可以成为下一期中的证据、案例或更正,署名归您,并在属于您本人作品时以 CC BY 4.0 发布。
- 电子邮件网关 GW-GCRI从本页发送证据、更正、案例、方法、审查、翻译或概况投稿。它会带上本页标签送达,由负责的编辑接收。
- ZenodoOpen depositDeposit it and submit to the communityPublish your own record on Zenodo with your ORCID iD, then submit it to the Reports community there. Curators review it; the DOI stays yours.
- 报告GCRI查看所有贡献方式贡献类型、各自的评审方式、您获得的署名、许可协议,以及绝不接受的内容。
切勿发送个人数据、敏感基础设施细节或您无权分享的任何内容。贡献在使用前会经过审查;发送贡献不赋予署名权或编辑权,赞助也绝不能换取研究结论。
法律文件。
本网站上的地图、数据和图表仅作为背景和一般信息展示。它们不是建议、预测或预警;信息的局限见免责声明和数据说明。