verification

双语

Show HN: Tally – check a spreadsheet's numbers against their source, in-browser
5 days ago
- Tally independently recomputes every figure from source data and shows which numbers match or mismatch with exact row proof.
- Users can drop a file or connect a Google Sheet; Tally marks each figure as verified or not with the underlying rows as receipts.
- Tally offers automated re-verification: it re-pulls connected sheets daily/weekly and alerts when data changes or figures stop tying out.
- Every verification generates a shareable 'Verified' link that lets recipients click any figure to see the source rows.
更多...
Before the theorem prover: verification is older than the alphabet
6 days ago
- Verification predates writing; around 3300 BC, sealed clay envelopes with redundant marks evolved into writing itself.
- Ancient civilizations had checksums, audits, inverse-operation verification, acceptance tests, split-key authentication, and signed certificates.
- Athens' 375/4 BC coin law carved a complete verification system into marble: written spec, daily verifier, per-defect actions, and sanctions on the checker.
- Ancient checks had known limits: casting-out-nines misses errors divisible by nine; the Talmud admitted its letter-count expertise had decayed.
更多...
I built an iOS camera that cryptographically verifies media at capture
7 days ago
- Apex 从 iPhone 拍摄流程开始，在源头验证照片和视频。
- 证据从相机内部开始，而不是通过后期的猜测或检测。
- 用户可以将 Apex 相机添加到锁屏，以便更快地访问并拍摄经过验证的媒体。
Now that Huang believes in open source, looking forward to CUDA open sourcing
10 days ago
- Browser verification process starts automatically and takes a few seconds.
- If verification fails, whitelist your extensions, update your browser, or use a modern browser like Firefox, Chrome, or Safari.
- For further assistance, email antibot@xcancel.com with your browser details and reference ID.
Ten ways a check passes while the thing it checks is broken
14 days ago
- 从未失败的检查是不可靠的——故意破坏它，以确认它确实会失败。
- 在后果发生的层面进行检查（例如，渲染后的页面，而非仅源代码），因为简单的检查会遗漏失败。
- 可靠性不等于有效性：测量结果可能一致，但测量的内容有误；测试在不相关变换下的不变性。
- 测试数据若已知假设则缺乏说服力；应使用早于问题产生的材料，以避免确认偏误。
更多...
PopcornGap: Rotten Tomatoes Audience Scores Without the Push Votes
14 days ago
- 2019年之前，烂番茄采用“拉取”系统，用户自愿访问网站对电影进行评分。自2019年起，他们改用“推送”系统，通过Fandango等应用提示购票者进行评分。
- “推送”系统从首周末观众中获得更多正面评分，而2019年前的“拉取”系统则因用户自我选择而给出更苛刻的评分。
- “真实”评分通过从总观众中减去已验证（推送）投票计算得出，揭示未验证（拉取）用户的评分，通常更低。
- 例如，《奥德赛》在同一天显示97%的已验证评分，而来自未验证用户的真实评分仅为56.2%。
更多...
It's Hard to Eval Is a Product Smell
a month ago
- 我们的产品很难评估'的说法是一种产品异味，意味着可能存在对开发者和用户都造成困扰的可用性问题。
- 设计易于验证的产品应优先考虑用户信任，并减少用户为验证AI输出而重新做工作的需求。
- 示例1：AI数据代理应提供可检查的成果物，例如具备渐进式披露功能的交互式笔记本，以便通过可信来源、指标定义和合理性检查来验证。
- 示例2：用于体育教育课程计划的AI工具应将生成的计划锚定在已审核的现有计划基础上，显示差异和变更原因，以减少审查时的认知负担。
更多...

verification

Show HN: Tally – check a spreadsheet's numbers against their source, in-browser

Before the theorem prover: verification is older than the alphabet

I built an iOS camera that cryptographically verifies media at capture

Now that Huang believes in open source, looking forward to CUDA open sourcing

Ten ways a check passes while the thing it checks is broken

PopcornGap: Rotten Tomatoes Audience Scores Without the Push Votes

It's Hard to Eval Is a Product Smell