Spaces:
Running
Running
| export const LANGS = ['zh', 'en', 'ja', 'ko', 'de', 'uk', 'in', 'br']; | |
| export const T = { | |
| headline: '把隐私交给 AI 之前，先脱敏', | |
| headlineEn: 'Redact PII before it reaches an AI', | |
| trust: '🔒 全部在你的浏览器里运行，数据不上传 · runs locally, nothing uploaded', | |
| loading: '加载中… · loading', | |
| ready: '● 已就绪 · ready', | |
| failed: '⚠ 加载失败 · failed: ', | |
| tryLabel: '试试 · Try:', | |
| redact: '一键脱敏 · Redact', | |
| yourText: '你的原文 · Your text', | |
| aiSees: 'AI 看到的 · What the AI sees', | |
| restoredOk: '✓ 还能一字不差地还原 · restored exactly', | |
| howTitle: '怎么做到的 · How it works', | |
| steps: ['① 找出隐私信息 · find personal info', '② 换成安全替身 · swap in safe look-alikes', '③ 回复里再换回来 · restore from the reply'], | |
| badges: ['78 类隐私信息', '8 种语言', '可逆还原', '开源 Apache-2.0', '身份证/银行卡校验'], | |
| limitNote: '⚠ 脱敏移除显式 PII，但不能保证抵御从残留上下文的推理式重识别 · removes explicit PII, not a guarantee against inference-based re-identification', | |
| llmProof: '真实大模型，这个场景零泄漏 · Real AIs, zero leaks in this case', | |
| llmProvenance: '来源 · Source: ', | |
| llmCaptionTail: '。非对抗输入下的缓存结果，非绝对保证 · cached reference run, not a guarantee against adversarial input.', | |
| devFold: '开发者选项 · For developers', | |
| devInput: '在这里输入或粘贴文本 · type or paste text', | |
| devStrategyTitle: '每类信息的处理方式 · Per-type handling', | |
| devNames: '已知姓名（逗号分隔）· known names', | |
| devLang: '语言 · Languages', | |
| devSeed: '种子 · seed', | |
| randomize: '🎲 随机 · Randomize', | |
| devRun: '脱敏 · Redact', | |
| keyJson: '密钥 · Key (JSON)', | |
| copy: '复制 · Copy', | |
| redacted: '脱敏后 · Redacted', | |
| restored: '还原 · Restored', | |
| streamTitle: '流式处理 · Streaming', | |
| streamRun: '开始流式 · Stream it', | |
| // Guarded restore panel — see demo/js/guarded.js (initGuarded). No live LLM | |
| // is wired up in this demo: the panel drives the guard with a SIMULATED | |
| // echoed reply (happy path) and lets the visitor edit/clear it to see the | |
| // fail-closed path, never implying a real model round-trip. | |
| guardedTitle: '带防护的还原（模拟回复）· Guarded restore (simulated reply)', | |
| guardedNote: | |
| '⚠ 这里没有接入真实大模型：下面的"模拟回复"演示两条路径——正常回显后完整还原，或验证令牌缺失/被篡改后安全拒绝还原 · No live LLM here — the "simulated reply" below demonstrates both paths: a normal echo restores everything, a missing/tampered verification token safely refuses the restore.', | |
| guardedBuild: '脱敏并生成验证提示 · Redact & build the verification prompt', | |
| guardedLblRedacted: 'AI 看到的（脱敏后）· What the AI sees (redacted)', | |
| guardedLblPrompt: '发给大模型的验证提示（可复制）· Verification prompt to send your LLM (copyable)', | |
| guardedLblReply: | |
| '模拟的大模型回复 — 编辑或清空来试试安全拒绝 · Simulated LLM reply — edit or clear it to try the fail-closed path', | |
| guardedRun: '带防护地还原 · Guarded restore', | |
| guardedResultLbl: '还原结果 · Restored result', | |
| guardedWithheldLbl: '被保留（未还原）的项 · Withheld items', | |
| // Copy keyed by the guard's stable event-kind vocabulary (identical strings | |
| // the PyO3 and wasm bindings both emit — see guard_event_kind_str in | |
| // crates/argus-redact-wasm/src/lib.rs). Structured events carry no prose of | |
| // their own by design; this is where the demo owns that prose. | |
| guardKind: { | |
| guard_no_anchor: | |
| '没有提供验证锚点，出于安全考虑拒绝还原 · No verification anchor was supplied, so the restore was refused for safety', | |
| provenance_failed: | |
| '回复中没有找到验证令牌，无法证明这是同一次对话，还原被拒绝 · The verification token was not found in the reply, so provenance could not be proven and the restore was refused', | |
| empty_key_with_scope: | |
| '提供了授权范围但密钥为空，没有可还原的内容 · A scope was supplied but the key was empty, so there was nothing to restore', | |
| out_of_scope_pseudonym: | |
| '回复中出现了超出本次授权范围的替身编码，已保留未还原 · The reply contained a pseudonym code outside the scope authorized for this restore, so it was withheld', | |
| alias_collision: | |
| '检测到别名冲突，为安全起见保留了该项 · An alias collision was detected, so that item was withheld for safety', | |
| }, | |
| // Copy keyed by the guard's stable outcome vocabulary (restore_outcome_str). | |
| guardOutcome: { | |
| blocked: '❌ 已阻止 — 没有任何原文被还原 · Blocked — nothing was restored', | |
| partial: '⚠ 部分还原 — 超出授权范围的项被保留 · Partial — out-of-scope items were withheld', | |
| complete: '✓ 完全还原 · Complete — everything was restored', | |
| }, | |
| }; | |