{ "schema": "environment-score-release-v1", "release": "public-score-v34.154-billing-freshness-risk-research-v70", "research_version": "risk-research-v70", "observer_version": "0.1.9", "assets": { "research.json": "69afaf6fcf88c69248b1d0868dde290dc031f5c3f0ce150098ca423fa9064861", "site.js": "8274975e00dbc832eea29688fd02409cd97fdc1d4eb9c37a6e915a9922b6e4c0", "site.css": "7f665219fa5a9c4bdf49779cd10aca0a4ca981bde17d29ea7d4996f59cc17996", "claude-web-observer.zip": "27d360f19cf765a9d2c21115a3cdd76e38d29fb181bc5eb80cac3dc2767df164", "index.html": "26d86136082df16a5094fa062a65c33b38877838a5f90c8efc907acbe6744edc", "provenance-summary.json": "162270d774d5d6f78a6eaed6887bda74fa37f1480df63354e5fcbe445239b858", "devices.py": "0ca5b00375f4a288bc8820bfd6c8514b764b7cf2f7a786671014042dd896c096", "device_assessment.py": "aab4eab6e8439544d1de2ba3721a2504c80d97828f5c49b78ff35b5370d2f60d", "agent.py": "205cf29266fc4822d546dd1e79d82cbdb22e95a6178e32f7e0453f45197b8764", "agent_network.py": "b94b6a6ce745a6e368d93d8173b33e5a7bcf69fa3e2d8bdf926a7d58093a690b", "agent_release.py": "9848e5953e2a1c7c96e6a2fd322e2c9c39131cc5f64288c71cfac230499c3083", "environment-score-agent-macos-arm64-v2.zip": "03d41f67e9511a04a901f39e310b7439f9e4412a343ee699f3758208d241711c", "environment-score-agent-macos-arm64-v3.zip": "c24491f7e3cf089e1b7ee7b3b4bc2ae919fd0b8aaa36d99bb29a6ccf3db20b29", "member_cli.py": "1afe23717096d04b3645660b31bfdd3fd616b6a9b93da4f3979daa9d6b479899", "collect_member_cli.py": "8927458ea7021f5a1847efdafbcdc1d1b8d52031752cdf861c26faa5ce69d542", "agent_package.py": "d83312e234c7be527cdc2c3addcb15569724a04414d489a79303913c87560d8e", "environment-score-agent-macos-arm64-v4.zip": "f9da9ff23a4cd043130fa64f59757b354ee745b882067d7150f9404aeace1344", "search-matrix-summary.json": "aa915f1f98f91d3d303a81cf5d2102cd6bf527fedb17cac74c891ab767c330f5", "server.py": "666212d69e95f41af890f82a8c8b5be5f1ff87d444f56c45e6d8722e8c5316cc", "claude-web-observer-firefox-v0.1.4.zip": "b380103e8cfea8784e95b65f49e8ce0bc6a6bd64c4b5433318a2ca9ab9fc7b3d", "accounts.py": "fbb0a0cc632b70f409972a8b2c5883b7cc5d18fab4c52630bad236d66dbe7ead", "claude-web-observer-firefox-v0.1.6.xpi": "6ab732e366c97cfdc3f2d495b5544cb9fe2cc261ddfaabd3f0ee86e0f08bd721", "claude-web-observer-firefox-updates.json": "8b3feba97f9c43d08f6dca52ba90cf276e04f18cf01ec40e5e3985b3aa58e83a", "claude-web-observer-firefox-v0.1.7.xpi": "a898f808407d65ec0268cc748f2f1f364aae2329f5001818e096b8f0fe999d5e", "claude-web-observer-firefox-v0.1.8.xpi": "81f536e1a137d2f4d2e50b5d562cae78a021dbdefecffccf4841fa4042ddbad9", "claude-web-observer-firefox-v0.1.9.xpi": "06bb2e44d0dc2433b6024ad817493c6e86940a68ce3059c9dbd70bd326266a88", "attention_engine.py": "6f5be2a43a2f9e15b4672266fbf5fc0845ec7abfa45d1c90cb4009f73c7479de", "windows_probe.py": "8c46e3018d5385bb2ea5b01cad112fdd0c89308c543ff017479b1d127e41b699", "host_checks.py": "7d98f04cc60f28efd223fe8d30b91ffb0d4b2b69fe4617a705d7fae1a0151d9d", "environment-score-agent-windows-amd64-assistant-release-v6.zip": "5cb3166a2dad5cc7da79de6065064169e31d1bf11e0d5921560b677d52d5631e", "environment-score-agent-macos-arm64-assistant-release-v6.zip": "b7f504bc356050903592c6de47c7a1416d081170f7b576e18d963bcee32f8122", "environment-score-agent-linux-x86_64-assistant-release-v6.zip": "20efe19699daff77abe238c3ae7217570f52f1bdd8f34d61ef5a2d23bff1df33", "environment-score-agent-windows-amd64-assistant-release-v7.zip": "4af771646b3c0636d7a1726bbefffa05ca8950a87c3ffc0b125b901948db099f", "environment-score-agent-macos-arm64-assistant-release-v7.zip": "f7b9b8e9d3fe54e62810d8dbbad0f17db5ed144b5b9c62e9129e776982ae3b31", "environment-score-agent-linux-x86_64-assistant-release-v7.zip": "f23690751ca397440cce28fd3d1b804e99576efb3a33d8fef481b67094e4d4be", "environment-score-agent-windows-amd64-assistant-release-v8.zip": "2bb03f1fbfde463aabc33aa8ce66bb9ffa25831cdc520f35fc86700b4de65df7", "environment-score-agent-macos-arm64-assistant-release-v8.zip": "025af9385d777b2ad8041c5af22d8503087bb78a6e5339c2750dc52034b77a6c", "environment-score-agent-linux-x86_64-assistant-release-v8.zip": "d4c8712d1ceebba6401638a7724cddde5f64bfdac5cb676446ab10ab04426b23", "user_settings_probe.py": "10e4040290c8828b78405c077feb47bbdf08f34dfeb7284679e9c83e170d6813", "environment-score-agent-windows-amd64-assistant-release-v9.zip": "a3f7c5894a0d91f4d92592c11b8c63dfcc9255d0e7cd29cce90eda5d76ae2734", "environment-score-agent-macos-arm64-assistant-release-v9.zip": "936ddbd2c58eafedbf9b225912e65622733502cf743912b384066c236c5123b4", "environment-score-agent-linux-x86_64-assistant-release-v9.zip": "ac5969cb8b54e35e1da0ea69fd38369dcfdd333580341ec1b697d6f79ebcb0a8", "registry_policy_probe.py": "04714395c85f6c3540db94e93a2899ea76fe1afb0492e654247ddeffb7d671f5", "managed_settings_probe.py": "e6fe071c9c397f681ea4b12cf4cb9b104b6d4c3a92f76a14af7f0bc66500de6e", "windows_managed_probe.py": "0b1ace14ef27743c8b5ae78295d2e5587a4853b28bffb87d0a671a0cb3d3cfbf", "environment-score-agent-windows-amd64-assistant-release-v10.zip": "4c80cc9cdd4e7a62ef36f7fbfe239cedfd8905d863f2cc17d1b9cbb4f6edba55", "environment-score-agent-macos-arm64-assistant-release-v10.zip": "b0d0f7bba2821ee23f2cec8a60b337902c890a3c8a24fd7b5c88be14417ad977", "environment-score-agent-linux-x86_64-assistant-release-v10.zip": "abe4e1ee52727b5e4bec079f7588eeaa604f4cbded430df64415f7fa0a70f76c", "environment-score-agent-windows-amd64-assistant-release-v11.zip": "a79b3687eb342783b82ae53f102e7775d5b314be3b6cab59883b3a246d2866b1", "environment-score-agent-macos-arm64-assistant-release-v11.zip": "2afe17d51bae02d49593776d627c594f98762d913a0876de4c5341918f8cc9ce", "environment-score-agent-linux-x86_64-assistant-release-v11.zip": "60f39ca5f5491d170cccc758f4138536318b65c392ff66e480bb90efb2c3def6", "windows_credential_acl.py": "e90368715036d173226985f5772651711380e285a4691ae2dcef0902a62aa7c5", "environment-score-agent-windows-amd64-assistant-release-v12.zip": "72584f62ceaa69928f0d8defdc11bb9a182f355df9e40f15ba38d0f4eefefe81", "environment-score-agent-macos-arm64-assistant-release-v12.zip": "24ce435f02d60f2eabcde0c088fdc98926ae7bcf84cdc5a4beddb7b22855dbf2", "environment-score-agent-linux-x86_64-assistant-release-v12.zip": "e7efa5974a9c26e20c18b099f585cc40641fc79a1c06b0bb88a80c3d1c223eac", "agent_startup.py": "db6c8b9766412f5546717a8a2445055572ac7dbfbc71931834a0b62334c533e4", "environment-score-agent-windows-amd64-assistant-release-v13.zip": "1d15869bb1e31b1bc42314c95e72f3ab7c251265d2234b4b007c865d2f6caf38", "environment-score-agent-macos-arm64-assistant-release-v13.zip": "efd7670c3a5a75942b062ba71714fec52478fef55c0a57ab105b06e17f78edc2", "environment-score-agent-linux-x86_64-assistant-release-v13.zip": "972aafafa1b5e2b5c4c826ed83075f2499e08c44bf55f4dc2073331e4eb904f7", "assessment.py": "6d335c69a672b64b06d05964e1a01e4f7262d9144fbdb500ed8ce82ee9ab1e4a", "product_scope.py": "bdef15655ed982a766bac0c7ee75b3751b70762598b18f4f3e599e889678d883", "environment-score-agent-windows-amd64-assistant-release-v13-onedir.zip": "8cb7a5117a0c77ca7817fed5712fd685398452ae315bb288d7c356f0a3974618", "environment-score-agent-macos-arm64-assistant-release-v14.zip": "fa187181791fa749b1616030b7848f33ef60b9699531e2d3e74571a615658a1d", "environment-score-agent-linux-x86_64-assistant-release-v14.zip": "56d18b437a0962217784cfd3ba0a4824976f2020100e1d2de7602036c585a18a", "environment-score-agent-windows-amd64-assistant-release-v14-onedir.zip": "84927616e6591b2e5111f82efe86e8a7056246986229ac0918077f013309c38f", "seed-alignment.json": "4b6b88990042843f6f49fe0dab48bf15db911e872aa1094fef53a4d4ff99c27f", "network.py": "c34a435a55f1b982d4ea16f06654def2bbd7d93a272aec06cedc407f20eb78aa", "risk-provenance.json": "8b0c0ce80138ef7034c1e83afe6241ab6e1ceb0688744bb5f4061f6f23a2d1b0" }, "notes": [ "研究线索不直接计入技术主分", "网页观测事件仅保存在扩展会话内", "真实Claude网页会话仍需自然使用验收", "设备不适用状态同步评分基数；改善预估与主评分按组最高权重和适用范围保持一致；真实设备现场验收待完成", "Firefox v0.1.8 已签名并通过现有更新通道实际升级，本机开启和共享关闭状态保留。已取得两条自然会话传输样本：终止标志后副本读取失败，同时调用方取消信号触发，原因未确定。无人操作的定时更新及完整会话兼容性未验收。Chrome/Edge 开发预览仍为 v0.1.6。", "设备不适用说明按具体项目显示；IPv6提示仅用于已标记不适用的IPv6默认路由或回退项目，不再把其他不适用状态归因为无IPv6。", "Firefox v0.1.9 已获 Mozilla 签名并发布；仅改进终止标志后副本读取失败的本地解释，不改变评分或权限。v0.1.9 本机升级待验收，未确定取消原因。", "Linux x86_64 runtime bundle built on dorm WSL2 with current assistant-release-v4 source; isolated binary starts, pairing and browser workflow pending. Corrected release metadata from v23 to actual v4 without changing assistant code.", "Linux runtime download instructions now distinguish x86_64 bundles from Python source packages; device pairing and end-to-end scoring acceptance remain pending.", "Browser extension heartbeat status no longer overwrites the selected assistant-device report scope; missing-extension guidance uses the current download version.", "Native Mac arm64, Linux x86_64 and Windows amd64 bundles use assistant-release-v5. Explicit UTF-8 policy reads fix the observed native Windows GBK startup failure. Windows pairing/report/pause/resume and forget on a temporary protected-state copy accepted; original binding retained. No dorm Claude access or network probing; VPS prerequisite remains unverified. Source downloads remain available separately for other architectures.", "Windows device reports now label credential access as unverified Windows ACL evidence; score weights and unknown deductions are unchanged.", "assistant-release-v6 native Windows, macOS arm64 and Linux x86_64 bundles built and startup-checked. Windows system hosts path now uses GetSystemDirectoryW; oversized adapter/DNS/NRPT samples remain unknown instead of reporting truncated counts. Windows isolated binding upload and temporary pause/resume/forget accepted with network probes disabled; host mapping counts remain local diagnostic facts, not extra scoring rules. Dorm OpenVPN dual-stack egress separately verified; no Claude/Anthropic requests made. Full detection coverage, new-user flow and Firefox natural-session acceptance remain incomplete.", "assistant-release-v7 connects native Windows adapter DNS configuration counts to the existing host_dns status. Only fixed state is uploaded, not DNS server identities. Actual Windows report:12/54 observed,0 issue/82 unverified deductions; DNS query/path/callback remain unknown and network probes disabled. Configuration evidence is not DNS leak/route proof. No scoring weights changed. Mac startup eventually returned unpaired after a slow run; startup latency cause not established.", "Fresh successful IPv6 transport establishes applicability even with ULA-only VPN interfaces; missing global interface addresses alone no longer exempt IPv6 route/fallback checks. Route summaries remain separate from actual traffic evidence; stale samples remain unknown. Assistant v7 unchanged.", "assistant-release-v8 reads Windows GetIpForwardTable2 through bounded native ctypes. Default routes and paired /1 routes on the same interface provide host route configuration status only. Native Windows report:14/54 observed,40 unknown; issue0/unverified82 deductions,network disabled. No route winner, egress or Claude session inference; existing score weights unchanged.", "DNS sample coverage follows the assistant diagnostic route: direct/system VPN requires system A/AAAA and system callback; explicit HTTP proxy requires both system and proxy callbacks. Unsupported routes, missing/expired samples remain unknown. Runtime row titles and UI now identify used-path sampling, not full DNS leak or real Claude session verification. No agent changes, weights or added exemptions.", "assistant-release-v9 adds Windows handle-based selected-user settings.json reads with bounded parsing, all-ancestor reparse checks, hardlink rejection and no-write/no-delete sharing guards. Only fixed declarations are projected, no settings content. Native Windows fixture and frozen local-only report acceptance passed; Windows effective ACL and managed-system settings remain unverified. Original account binding/report left intact; no added network probing. Score weights and metadata schema unchanged.", "assistant-release-v10 reads the fixed Windows C:\\Program Files\\ClaudeCode managed-settings.json and nonhidden JSON drop-ins with retained native handles, bounded inventory and no-write/no-delete file sharing. Reparse points, hardlinks, concurrent writers and changing inventory remain unverified. Only per-file declarations are projected; no policy merge or execution. Windows ACL, registry, remote/host policy and effective configuration remain unknown even when the file source is absent. Empty system policy documents are treated as {} per official documentation. No additional network probes or production fixture reports; score weights unchanged.", "assistant-release-v11 adds optional closed Windows 64-bit HKLM/HKCU Settings declaration evidence using bounded native registry reads. Registry link markers, oversized values, key/value changes, duplicate/deep JSON and unresolved REG_EXPAND_SZ templates stay unknown. Empty registry strings are malformed, unlike empty file policies. Per-source rows enter existing correlated group budgets without additive weights; HKLM launch-blocking format guidance differs from HKCU notice/fallback guidance. Old report shapes remain accepted; Windows reports include 22 additional declaration/ACL rows, so coverage counts differ. ACL, 32-bit registry view, remote/host policy and effective Claude settings remain unverified. Frontend distinguishes assistant updates, ACL integration and unparsed declarations.", "assistant-release-v12 adds fixed Windows credential-file and ancestor DACL static signals using READ_CONTROL-only retained handles. Null DACLs and relevant standard broad-group allow ACEs are attention signals; broad deny/unsupported ACEs, changes, links, unreadable or missing paths remain unknown. No broad grant does not prove effective access: custom groups, AccessCheck, system credential stores and actual Claude storage remain unverified. Ancestor creation/traversal alone is not treated as credential exposure. No credential content, SID, path or raw ACL is exported; existing correlated weights unchanged. Native fixture and frozen local-only report accepted; network stays off.", "Optimization details now render each fixed evidence_note alongside remedies, including missing credential files and bounded DACL interpretation. No scoring change.", "assistant-release-v13 pins the selected local state directory in all startup commands. CLI --state-dir validates absolute paths/control characters. Windows uses a current-user, Limited interactive logon task XML, separates executable/arguments to avoid schtasks /TR 261-character limit, disables the default execution time limit, and refuses overwriting an existing named task. A native temporary scheduled task uploaded locally with network off, correct binding and Limited run level; original binding retained and task removed. Guided onboarding separately asks before background installation; foreground remains available. No scoring or network opt-in changes. Actual restart/logon acceptance remains pending.", "Score overview now leads directly to optimization steps and the complete deduction ledger. Supporting capability, status, login and optional collection panels follow under a separate heading. Existing detection, permissions and scoring are unchanged. Reference-only review of GOV.UK Frontend native details pattern; no third-party code or dependency copied.", "Explicit combined/webpage/Claude Code score views use the same closed evidence. Combined scoring remains unchanged; webpage excludes 33 terminal/configuration rules and Windows registry-only policy rows, not shared host DNS or network evidence. CLI view excludes browser checks and browser-only callbacks/stability. Each view includes exact rule IDs, exclusions, normalized group deductions and scope version. Excluded rows are not passed or measured not-applicable. Raw history preserves combined reports; exports label the selected view. No user network opt-in, collector or extension changes.", "History no longer appends literal null for missing comparison. It finds the earlier report from the same nonempty browser session or exact paired device, with matching scoring version, subject scope, budget and rule IDs. Unidentified browser snapshots and different sources are not compared. History introduction correctly includes assistant reports and labels device evidence separately from browser sessions. Raw combined reports unchanged.", "Fresh fixed-target device network samples carry forward across newer local-only snapshots for the same device until the original 900-second TTL expires. Carry-forward preserves the original sample timestamp, does not probe again, excludes per-upload browser/assistant exit alignment, and is shown explicitly in the device UI.", "评分范围默认按 Claude 网页版显示，并在本浏览器记住用户选择；切换仍不改变原始报告、权限或评分规则。", "首页脚本缓存版本同步更新到 v34.111，确保浏览器获取同版本的评分范围默认与本机选择记忆逻辑。", "助手设备报告清楚区分采集时间与页面重新评分时间，避免把读取时刻误当成重新采集；浏览器报告仍显示页面检测时间。评分规则与设备采集未变。", "Windows v13 onedir packaging passed isolated native scheduled-task start/stop/restart with one main process, no surviving process after stop, and loopback sequences1/2. All runtime dependencies are hashed and verified; unchanged assistant v13 source and opt-in. Production restart/logon/reboot recovery remains pending; no new scoring or network default changes.", "Anonymous startup account checks preserve the selected registration/recovery form and typed inputs. Startup responses cannot replace a newer authentication action.", "Device network expiry/absence explains that disabled probing cannot be replenished by refreshing. Device remediation button says read latest report and explicitly does not remotely enable probes. Scoring/probe defaults unchanged.", "Assistant v14 adds current-user interactive logon plus a UTC-anchored five-minute repeated Windows wake, IgnoreNew and unlimited task execution. Native Windows acceptance verifies no duplicate, persistent pause across crash recovery, sequence/binding preservation, uninstall disabling future wakes and no restart across the next five-minute boundary. Network probing defaults unchanged; actual Windows reboot/logon remains untested.", "When no helper device is listed, users who already started one are prompted to confirm the website and helper use the same scoring account; device reports do not cross accounts.", "澄清评分站共享授权与扩展本机观测互相独立；共享关闭时本站不读取或上传，本机观测可继续。", "研究库 v43 增补公开社区及官方仓库用户自述；个案未核实，不作因果推断，0项研究因素启用评分。", "v44 adds three unverified Reddit/V2EX source records and four non-causal claims; no scoring rules enabled.", "v45 adds three unverified Reddit/V2EX self-reports as non-causal, non-scoring research; no score rules enabled.", "v46 adds three unverified Hacker News community self-reports as non-causal, non-scoring research; no score rules enabled.", "v47 corrects two factor references to one existing Reddit source: Claude Code surface, publication/recheck dates, and shared incident linkage. No source or claim added; no scoring enabled.", "v50 adds current official identity-verification scope to R003: rollout is limited to certain use cases/capabilities and routine checks. No automatic identity scoring or score change; verification status is not inferred.", "v51 adds Anthropic 2026 H1 aggregate banned-account/appeal/overturn counts as R061 system context. Aggregate totals do not identify environment-level causes or support individual risk weights; scoring unchanged.", "Local candidate only, not deployed. Adds 8 non-causal community source records grouped into 7 reported narratives under R061; no score rules changed; 0 research factors enabled.", "v53 adds one full-text, single-source Reddit suspension narrative to R061 as unverified background only; no score rule or weight changed.", "v54 adds one Reddit author self-report as unverified non-causal background; comments and causal guesses excluded; no score rule or weight changed.", "v62 records one full-text Chinese commercial anti-ban guide as an unverified marketing source with six atomic assertions; no causal or scoring rule added, and no evasion advice endorsed.", "v63 clarifies commercial marketing claim labels and corrects R062 research-only guidance; scoring rules unchanged.", "Follow-up UI fix: translate Zhihu platform and marketing evidence filter options.", "v64 corrects R018 attribution wording and marks the R001 Hong Kong VPN rumor crosspost as a non-incident discussion; no new event or scoring rule added.", "v65 records one Reddit author report across R016/R018 as a single event, plus a distinct commenter report under R031. VPN and keyword causality remain unverified; no scoring rule or weight changed.", "v66 refreshes the caveat on existing GitHub #51588: private support statements are reported by the author, not independently verified or publicly confirmed. No source, incident, score rule, or weight added.", "v67 adds official aggregate Jan-Jun 2026 account-enforcement and appeal totals to R016 as context only. No environment attribution or score rule is inferred.", "v68 adds GitHub #99907 as one unverified author event linked to R016/R018; VPN causality and product context are unresolved. No score rule or weight changed.", "v69 adds one SCMP secondary report linked to the already catalogued Reddit #1wutq5d event; no additional incident, causal claim, score rule, or weight is added.", "v34.151 shows an actionable Chinese message for pair-preview fetch failures and advances the site.js cache-buster; a failed preview does not pair a device.", "v70 adds GitHub #63421 as one unverified consumer-Claude report and a same-author Reddit duplicate source; no cause, score rule or weight is inferred.", "v34.153 preserves the opt-in host DNS result from the same-device original network snapshot within its original 15-minute TTL; later local-only uploads do not erase that result or replace newer explicit DNS evidence. Policy weights and probe consent remain unchanged.", "v34.154 labels DigitalOcean quota snapshots older than 36 hours as stale and states that displayed remaining data may have changed; no quota records, scoring rules, or device probes changed." ], "assistant_revision": "assistant-release-v14", "firefox_signed_version": "0.1.9", "chrome_edge_preview_version": "0.1.6" }