docs(qa): seal bounded Y outcome and reconcile retained state
This commit is contained in:
@@ -266,3 +266,4 @@ docs/reliability-1.25.2-evidence/** binary
|
||||
docs/reliability-1.25.2-release-identity.json -text
|
||||
docs/reliability-1.25.2-offline.txt binary
|
||||
docs/reliability-1.25.2-git-index.txt binary
|
||||
docs/LIVE-DELIVERY-X-PREFREEZE-REFUSAL.txt binary
|
||||
|
||||
@@ -1,5 +1,7 @@
|
||||
# Agent Workbench项目执行规范
|
||||
|
||||
Y已结束未闭环:正式1.25.2/apikey GPT SSE,预检1请求110tokens;Node8请求64500tokens全部matched、原九基线失败/build通过、三源码模型提案封存并监督审查应用,最后inspect_session后step_limit,未做修改后检查/Git,不续跑。原收尾在恢复配置后因3条外部诊断追加失败;独立只读逐行/全原导出/工件/配置对账通过,开场快照与驱动配置冻结之间外部加DeepSeek入池(6→7)及两次复核错误完整保留,不回滚外部变更。69项目85会话6607记录13944事件/outbox、3857工件,原PID36552/Provider、凭据1、原身份来源保持,9229关闭。见docs/QA-LIVE-DELIVERY-Y.md。用户最新要求尝试当前apikey-Claude和DeepSeek;另冻每线路单次严格身份预检,通过才新Node,旧Y/X/W等不续跑,预算和原失败不改。
|
||||
|
||||
最新真实实验选择:X在预冻结时发现原jojocode被外部停用(revision12),在创建样本目录/诊断/模型之前拒绝,0请求。用户明确改用当前启用的apikey并“另冻新实验,不沿用X”,执行docs/LIVE-DELIVERY-Y-PLAN.md及scripts/live-y。连接93d0cf6d/Chat SSE/gpt-6.1-sol/凭据修订1,保留原身份来源https://api.apikey.fan;目标/8请求90000tokens/16K2048保持,不启用jojocode、不续跑X或旧样本。下面“下一X”按历史计划理解。
|
||||
|
||||
1.25.2已完成独立本地发行、真实隔离安装和保留原PID36552/instance/Provider的正式升级,产品来源d510cb7049aea9ca61ddc4cf2c2e21dfc0153f1d、候选20261104134035379;正式耗时289669ms,旧1.25.1的156369ms和1.25.0的774098ms保持。67项目83会话6477记录13663事件/outbox及3773原工件、全部原行/导出/公开配置在网页两视口及实际原生回看后再次exact,窗口0/9229关闭、hasKey=true。617冻结输入、源码ZIP622输入独立重建22/22、三归档、便携八套50组及文件9/结果16/读取10、实际安装12组和安装副本十套75组通过。最终完整测试898/899:重启用例在重启前首次审批8秒等待超时;相同源码/原断言/原期限精确补验1/1通过,不宣称整轮全绿。旧TLS/浏览器失败、两次并发背景Harness启动超时与迟到ready均保留;唯一超时根因/高负载稳定性仍未知。1339份126224544字节证据离线通过,见docs/QA-RELIABILITY-1.25.2.md、docs/RETROSPECTIVE-RELIABILITY-1.25.2.md及发行身份。下一按X计划只新冻一次诊断/身份预检/Node,不续跑旧样本;Node闭环后才React/Python及独立原版Purge。真实自主交付、网络间歇根因、真实异常剔除收益和Purge仍未验证。不委派、不推送、不读取/复制/解密/重存Key,不重启正式Provider。以下“验收中/正式1.25.1”等按历史时点理解。
|
||||
|
||||
@@ -1,19 +1,19 @@
|
||||
node:internal/modules/run_main:107
|
||||
triggerUncaughtException(
|
||||
^
|
||||
|
||||
AssertionError [ERR_ASSERTION]: Expected values to be strictly equal:
|
||||
|
||||
false !== true
|
||||
|
||||
at file:///D:/Ai-project/REA%E9%80%86%E5%90%91/Agent-Workbench/scripts/live-x/initial.mjs:15:8
|
||||
at process.processTicksAndRejections (node:internal/process/task_queues:104:5) {
|
||||
generatedMessage: true,
|
||||
code: 'ERR_ASSERTION',
|
||||
actual: false,
|
||||
expected: true,
|
||||
operator: 'strictEqual',
|
||||
diff: 'simple'
|
||||
}
|
||||
|
||||
Node.js v24.22.1
|
||||
node:internal/modules/run_main:107
|
||||
triggerUncaughtException(
|
||||
^
|
||||
|
||||
AssertionError [ERR_ASSERTION]: Expected values to be strictly equal:
|
||||
|
||||
false !== true
|
||||
|
||||
at file:///D:/Ai-project/REA%E9%80%86%E5%90%91/Agent-Workbench/scripts/live-x/initial.mjs:15:8
|
||||
at process.processTicksAndRejections (node:internal/process/task_queues:104:5) {
|
||||
generatedMessage: true,
|
||||
code: 'ERR_ASSERTION',
|
||||
actual: false,
|
||||
expected: true,
|
||||
operator: 'strictEqual',
|
||||
diff: 'simple'
|
||||
}
|
||||
|
||||
Node.js v24.22.1
|
||||
|
||||
@@ -1,12 +1,14 @@
|
||||
# Y:使用当前启用的 apikey 的新有界样本
|
||||
|
||||
执行已结束:预检1请求/110tokens通过;Node8请求/64500tokens、三文件模型提案与整批应用完成,最后inspect_session后step_limit,修改后检查和Git交付未执行,未闭环。原收尾因3条外部网络诊断追加触发断言失败,独立只读逐行/配置/工件对账通过并保留开场池变化和原失败;见[验收与回顾](QA-LIVE-DELIVERY-Y.md)。不续跑Y,后续按用户新指定的apikey-Claude/DeepSeek另冻实验。
|
||||
|
||||
用户明确选择“改用当前启用的apikey,另冻新实验,不沿用X”。正式1.25.2独立发行及保留状态升级已完成(产品来源d510cb7049aea9ca61ddc4cf2c2e21dfc0153f1d,候选20261104134035379)。X在预冻结第15行发现jojocode已停用(revision12),在创建样本目录、诊断、模型或工程之前拒绝;0请求且不重放,不启用该连接。Y是另一个样本,不能与W/X合并为同条件收益比较。
|
||||
|
||||
## 冻结因素与限额
|
||||
|
||||
使用当前已启用的apikey连接93d0cf6d-4f32-4918-9da8-6f125aa8a6e5:https://api.apikey.fan/v1、Chat Completions、gpt-6.1-sol、streaming=true、revision5、credentialRevision1、16K上下文/2048输出/单并发。保留它的流式设置及原身份record/空别名/来源https://api.apikey.fan;在新预检快照及Node临时策略中严格require-match、无别名,随后精确恢复。原Provider内存连接直接使用,不读取/复制/解密/重存Key。默认高性能池及其成员保持,工程阶段只临时选中本连接,恢复原选择;不切池、换线或自动重试。
|
||||
|
||||
scripts/live-y独立复制已审查X驱动;完整工程目标、原三模块/九测试/build/11文件、owner staged/unstaged/untracked保护和人工边界均不改。Y与W相同目标SHA应为ea3eaee0a94112085b6e41f528397f32fb2640a963a40add9bd3549b9eb642e0;实际冻结后再次核对。Node8请求/90000报告tokens/600000ms,16K/2048/1600,不增加预算、实现提示或催写。
|
||||
scripts/live-y独立复制已审查X驱动;完整工程目标、原三模块/九测试/build/11文件、owner staged/unstaged/untracked保护和人工边界均不改。Y与W源码目标声明片段SHA应为ea3eaee0a94112085b6e41f528397f32fb2640a963a40add9bd3549b9eb642e0;实际目标运行字节SHA为b5421d0bb502126f49a5153729027f13ef82db86ab19a9e5b0c23a49070bb905。Node8请求/90000报告tokens/600000ms,16K/2048/1600,不增加预算、实现提示或催写。
|
||||
|
||||
只执行一次:initial实际安装清单与来源冻结→network所属DNS/TCP/TLS(0HTTP/模型,仅观察不阻断)→preflight单次无工具身份请求(最多256输出/16000tokens/单请求,报告型号严格匹配)→预检通过才Node。预检失败即封存,网络/未知消费无自动重试或切换。工程只由模型实现,监督仅批准原test/build、审查应用模型完整三文件变更集和确认模型Git计划;审查续接仍属于原8请求预算。部分提案、不完整批次、身份不符、失败/无进展/预算停止均保留原结果,不代写或追加指导。
|
||||
|
||||
|
||||
@@ -0,0 +1,38 @@
|
||||
# Y:apikey GPT SSE 真实工程样本
|
||||
|
||||
结论:预检通过,工程未闭环。正式1.25.2上的新Node样本8次请求全部完成且上游型号matched,累计64500报告tokens、0未知消费;模型提出并封存三个模块,监督审查后按原指纹应用。最后一次请求只调用inspect_session,随后step_limit暂停;修改后检查、模型Git计划和交付均未执行。不能把三文件已应用写成自主交付成功,也不补跑或代写实现。
|
||||
|
||||
## 来源与预算
|
||||
|
||||
- 用户明确选择当前已启用apikey另冻Y,不沿用X;X因jojocode已停用在预冻结前拒绝、0请求,原日志保持。
|
||||
- 正式来源d510cb7049aea9ca61ddc4cf2c2e21dfc0153f1d、候选20261104134035379;驱动冻结af89f019,计划[LIVE-DELIVERY-Y-PLAN.md](LIVE-DELIVERY-Y-PLAN.md)。PID36552、原instance/Provider保持。
|
||||
- 连接93d0cf6d,https://api.apikey.fan/v1,Chat Completions/SSE/gpt-6.1-sol,凭据修订1;工程临时require-match、无别名,原record/[]/https://api.apikey.fan恢复。
|
||||
- 原目标运行字节SHA256:b5421d0bb502126f49a5153729027f13ef82db86ab19a9e5b0c23a49070bb905,与W逐字节相同;源码目标声明片段(含JS拼接语法)的SHA为ea3eaee0a94112085b6e41f528397f32fb2640a963a40add9bd3549b9eb642e0。Node上限8请求/90000tokens/600000ms,16K/2048/1600;审查续接只剩1请求,无新增目标或实现提示。
|
||||
|
||||
## 实际结果
|
||||
|
||||
| 阶段 | 原结果 |
|
||||
|---|---|
|
||||
| 网络 | 唯一所属诊断10104ms、partial;DNS/auto和IPv6 TLS通过,IPv4 TCP_CONNECT_TIMEOUT;0HTTP/模型/凭据访问。按预注册只记录,不阻断预检 |
|
||||
| 身份预检 | 单请求110tokens(85输入/25输出),matched/complete,nonce通过,0工具;9229关闭 |
|
||||
| 基线 | 模型规划并执行原test/build;九项测试0通过9失败,build通过,原失败保留 |
|
||||
| 读取 | 原2584字符测试与三原模块均读全;原readProgress.repeated=0,不能外推为所有任务不再重复 |
|
||||
| 修改 | 第6次请求提出三个精确补丁,第7次prepare_change_set;监督逐份审查原测试、BigInt精确分/安全汇总/格式及owner状态后批准,批次94b7c60a…实际completed |
|
||||
| 终止 | 第8次inspect_session后step_limit;62712输入+1788输出=64500tokens,0未知。墙钟327秒包含约212秒审查等待 |
|
||||
| 交付 | 修改后三文件保留;没有修改后的检查、Git计划或交付。原HEAD a71148ca…不变,owner原暂存/工作区/未跟踪字节保持 |
|
||||
|
||||
全样本合计9次真实请求、64610报告tokens。上游字段一致不认证模型权重。八份请求/响应均按原工件SHA取回,未重试或换池。v3参与全部8次工程请求,预留10158–14400tokens,excludedSources均0;未触发真实异常剔除边界,不能声称已证明其收益。Y更换供应商及流式设置,不能单独归因1.25.2策略改进。
|
||||
|
||||
## 收尾断言与独立对账
|
||||
|
||||
原驱动在恢复连接/预算/选择之后,发现modelNetworkDiagnostics行不同而失败退出;node/model-outcome.json已保存,node/result.json与原restoration.json没有产生。原日志及不完整收尾保持,未改成原驱动成功。
|
||||
|
||||
只读reconcile.mjs核对原行哈希,证明原11条诊断完整前缀仍与原metadata行SHA完全一致,仅追加3条其它诊断:jojocode partial、DeepSeek passed、apikey-Claude passed;均0HTTP/模型/凭据访问,非Y驱动发起,不推断操作者。另发现开场顺序快照与稍后配置冻结之间,高性能池由revision6增加DeepSeek成为revision7;原驱动冻结的是7并已保持。首次只读复核误以为池未变化而拒绝,第二次对“变化发生在收尾之后”的猜测也被原快照否定;两次错误及准确时间关系均独立保留,未回写任何外部配置。
|
||||
|
||||
独立核对通过:84个原导出、3777个原工件文件、原sessions/records/events/outbox每行、234个原modelAttempts精确保持;原项目行仅逆算本样本一次前插位置后精确保持。磁盘配置除单调修订一致,连接registry除声明修订/时间/排序一致;身份来源及凭据修订1保持。三条外部诊断和开场池改动保留。当前69项目/85会话/6607记录/13944事件及outbox、3857工件文件。窗口0、9229关闭、原PID36552、hasKey=true。
|
||||
|
||||
证据位于.qa/live-delivery-y,144份79372960字节,封存清单MANIFEST.json;node/reconciliation.json是独立只读复核结果,不替代原驱动失败。首次离线复核误把源码声明片段SHA用于运行目标字节而拒绝;核对W/Y实际目标1196字节完全相同后,同时验证两个不同对象的SHA,通过。已封存的初版报告/验证器原字节保持;本报告纠正其对象命名,身份见[live-y-evidence-identity.json](live-y-evidence-identity.json)。React/Python/Purge因Node未闭环未执行。用户随后指定尝试apikey-Claude与DeepSeek,另冻新样本,不继续Y。
|
||||
|
||||
## 阶段回顾
|
||||
|
||||
Y证明该路线本次能返回完整模型响应并生成三文件实现;网络波动没有在这8次请求形成上游失败。它没有证明工程自主交付或接口长期稳定。保持原8次预算和真实停止结果,后续按用户指定独立验证两条新路线;网络根因、真实异常剔除收益、高负载启动稳定性、Purge仍未知。
|
||||
@@ -0,0 +1,20 @@
|
||||
{
|
||||
"schemaVersion": 1,
|
||||
"sample": "Y",
|
||||
"manifestPath": ".qa/live-delivery-y/MANIFEST.json",
|
||||
"manifestSha256": "90016bc37213990084146859bc85bf7428b5a61cbe801533232a2bfc9dc168c3",
|
||||
"files": 144,
|
||||
"bytes": 79372960,
|
||||
"modelRequests": 9,
|
||||
"reportedTokens": 64610,
|
||||
"engineeringSuccess": false,
|
||||
"driverFinalizationFailed": true,
|
||||
"readOnlyReconciliationPassed": true,
|
||||
"goalRuntimeBytesSha256": "b5421d0bb502126f49a5153729027f13ef82db86ab19a9e5b0c23a49070bb905",
|
||||
"goalSourceDeclarationSha256": "ea3eaee0a94112085b6e41f528397f32fb2640a963a40add9bd3549b9eb642e0",
|
||||
"firstOfflineVerificationError": {
|
||||
"expectedSourceDeclarationHashOnRuntimeBytes": true,
|
||||
"correctedByVerifyingBothObjectsSeparately": true,
|
||||
"originalSealedVerifierPreserved": true
|
||||
}
|
||||
}
|
||||
@@ -1,5 +1,7 @@
|
||||
# 汇总发现
|
||||
|
||||
Y:1预检110tokens与8Node64500tokens全部matched,0上游失败/未知,三源码模型提案并封存应用,最后inspect_session/step_limit,未检查或Git交付。原Node收尾已恢复配置但因3条其它网络诊断追加失败;原11条诊断前缀按完整metadata行SHA精确重建。开场顺序快照池6、稍后驱动冻结池7(外部加DeepSeek)均保留;两次只读复核误判和准确时点对照不覆盖。独立对账84原导出/3777原工件/234原尝试/原行及配置通过;当前69/85/6607/13944、3857工件,原PID36552/Provider、凭据1、身份来源、9229关闭保持。用户指定apikey-Claude/DeepSeek另冻,不续跑Y。
|
||||
|
||||
## 1.25.2 起点
|
||||
|
||||
最终发行已落地:source d510/候选20261104134035379,正式289669ms、PID36552及Provider保持;67项目83会话6477记录13663事件/3773工件的全部原状态在网页/原生后exact。完整898/899唯一失败在api.test:101首次审批8秒等待,原断言同源1/1通过;不写成整轮全绿,唯一超时根因/高负载稳定性保持未知。新候选主要验收串行全部通过,622独立输入/22编译、12安装及75安装UI。1339份126224544字节证据离线通过;封存器首次不能识别失败日志的多行JSON,保留未封存的partial于.qa并补解析,未重放产品操作。真实样本X尚未执行。
|
||||
|
||||
@@ -1810,3 +1810,4 @@ V调用前检查:新增11文件SHA断言首语法检查发现多一个右括
|
||||
最终Git索引首核对因`*.log`漏暂存81份封存日志而拒绝,保留原错误;仅为本版主/补充证据目录加例外后,包含两份manifest的1428份158780733字节逐项索引原字节通过,成功收据docs/task-agreement-1.25-git-index.txt。AGENTS、1.25与Tianshu计划、更新记录、验收/回顾和本清单更新为1.25已本地交付。收尾598冻结输入当前SHA、发行身份精确一致、主manifest不变及30处本地文档链接检查通过,最终diff核对后随本次本地提交封存。Tianshu C–F/G仍独立后续;真实模型预算0,V网络partial、T/R/P/O原失败、重复读取、真实自主交付与Purge未知保持,不续跑、不委派、不推送。
|
||||
|
||||
用户要求整理长期未闭环问题供其他AI独立审查。本次从ddefe5f1干净基线只读核对T/R/P/O/V、U、E/H、N及1.25原更新耗时,形成docs/OPEN-PROBLEMS-HANDOFF.md:固定Node交付/重复读取、网络与诊断门槛、用量估算/累计预算、原版受限Purge收益、更新重复核验五项,逐项列事实、已做修复、未知与可证伪问题。明确保留E单个Python成功、E/H原版已加载及N已确认GPT分组,避免把历史已解决事项误列为当前卡点;U/1.25之后没有新真实工程结果,不宣称旧T缺陷在当前版仍复现。文档引用与关键数字复核、diff审查后仅本地提交;无模型/网络/安装复跑,无正式服务或Key操作。
|
||||
Y阶段结束:预检110tokens通过;Node8请求64500tokens全部matched,模型三源码整批应用,最后inspect_session后step_limit,未修改后检查/Git,未闭环。原收尾诊断metadata断言失败和两次只读复核错误封存;精确证明原诊断前缀、3外部新诊断、开场池6→7时间关系及原状态保持,不回滚外部编辑。用户新指定apikey-Claude/DeepSeek另冻实验,先单次严格身份预检。
|
||||
|
||||
@@ -0,0 +1,149 @@
|
||||
// Read-only reconciliation after the original driver stopped in its final
|
||||
// metadata assertion. Never resumes the task, applies files, or writes config.
|
||||
import fs from 'node:fs';
|
||||
import path from 'node:path';
|
||||
import assert from 'node:assert/strict';
|
||||
import { DatabaseSync } from 'node:sqlite';
|
||||
import { execFileSync } from 'node:child_process';
|
||||
import { canonical, digest, publicSnapshot } from './state.mjs';
|
||||
import { inspectorClosed } from './inspector.mjs';
|
||||
import { root, origin, dataRoot, save, client, meta, sha, connectionId } from './common.mjs';
|
||||
|
||||
const read = name => JSON.parse(fs.readFileSync(path.join(root, name)));
|
||||
assert.equal(fs.existsSync(path.join(root, 'MANIFEST.json')), false);
|
||||
assert.equal(fs.existsSync(path.join(root, 'node/reconciliation.json')), false);
|
||||
const before = read('node/before.json'), original = read('node/original-public-metadata.json');
|
||||
const outcome = read('node/model-outcome.json'), source = read('node/source-identity.json');
|
||||
const session = read('node/session.json');
|
||||
assert.equal(outcome.success, false);
|
||||
assert.equal(outcome.stopReason, 'step_limit');
|
||||
assert.equal(outcome.usage.requests, 8);
|
||||
assert.equal(fs.existsSync(path.join(root, 'node/result.json')), false, 'Preserve the original incomplete finalization.');
|
||||
const { api, boot, close } = await client();
|
||||
try {
|
||||
const host = await api('/host/status');
|
||||
assert.equal(host.identity.pid, 36552); assert.equal(host.activity.active, false);
|
||||
assert.equal(host.desktopWindows.count, 0); assert.equal(boot.version, '1.25.2');
|
||||
await inspectorClosed();
|
||||
const after = fs.existsSync(path.join(root, 'node/reconciled-after.json'))
|
||||
? read('node/reconciled-after.json') : await publicSnapshot(origin, dataRoot());
|
||||
if (!fs.existsSync(path.join(root, 'node/reconciled-after.json'))) save('node/reconciled-after.json', after);
|
||||
const currentSession = await api('/sessions/' + outcome.sessionId);
|
||||
assert.deepEqual(currentSession.execution.task.usage, outcome.usage);
|
||||
assert.equal(currentSession.status, 'paused');
|
||||
assert.equal(currentSession.execution.deliveries.length, 0);
|
||||
for (const item of before.exports) assert.deepEqual(after.exports.find(row => row.id === item.id), item);
|
||||
for (const table of ['sessions', 'records', 'events', 'outbox']) {
|
||||
const rows = new Map(after.database.tables[table].rows.map(row => [row.id, row]));
|
||||
for (const row of before.database.tables[table].rows) assert.deepEqual(rows.get(row.id), row, table + '/' + row.id);
|
||||
}
|
||||
const artifacts = new Map(after.persistent.artifacts.map(row => [row.path, row]));
|
||||
for (const item of before.persistent.artifacts) assert.deepEqual(artifacts.get(item.path), item);
|
||||
// The sequential opening snapshot and the driver's subsequent config freeze
|
||||
// straddle an external pool edit. Check both identities explicitly.
|
||||
const current = read('node/final-public-metadata.json');
|
||||
assert.deepEqual(canonical({ ...current.config, configRevision: original.config.configRevision }), canonical(original.config));
|
||||
assert.deepEqual(after.modelCenter.selected, before.modelCenter.selected);
|
||||
assert.deepEqual(read('node/after.json').modelCenter.pools, original.registry.pools);
|
||||
assert.deepEqual(after.modelCenter.pools, original.registry.pools);
|
||||
const poolBefore = before.modelCenter.pools[0], poolAfter = after.modelCenter.pools[0];
|
||||
assert.equal(after.modelCenter.pools.length, before.modelCenter.pools.length);
|
||||
assert.deepEqual(poolAfter, { ...poolBefore, revision: poolBefore.revision + 1,
|
||||
members: [...poolBefore.members, { connectionId: '1a7a55a9-29f2-4b9f-8ea2-5a55afd7eacd', weight: 1 }] });
|
||||
const observedMetadata = { config: meta('config'), registry: meta('modelRegistry') };
|
||||
assert.deepEqual(canonical(observedMetadata.config), canonical(current.config));
|
||||
const laterRegistry = structuredClone(observedMetadata.registry);
|
||||
assert.equal(laterRegistry.revision, current.registry.revision);
|
||||
assert.deepEqual(canonical(laterRegistry), canonical(current.registry));
|
||||
save('node/external-pool-edit-reconciliation.json', { observedMetadata, openingSnapshotPool: poolBefore,
|
||||
driverFrozenPool: original.registry.pools[0], observedPool: poolAfter,
|
||||
interpretation: 'DeepSeek was added between the sequential opening snapshot and the subsequent driver config freeze. The original driver preserved revision 7. The first reconciliation timing inference was incorrect; this comparison corrects it without overwriting that error or restoring away the edit.' });
|
||||
const registry = structuredClone(current.registry);
|
||||
assert.ok(registry.revision > original.registry.revision);
|
||||
registry.revision = original.registry.revision;
|
||||
const initialConnection = original.registry.connections.find(row => row.id === connectionId);
|
||||
const finalConnection = registry.connections.find(row => row.id === connectionId);
|
||||
assert.equal(finalConnection.revision, initialConnection.revision + 2);
|
||||
finalConnection.revision = initialConnection.revision;
|
||||
finalConnection.updatedAt = initialConnection.updatedAt;
|
||||
for (const value of [registry, original.registry]) value.connections.sort((a, b) => a.id.localeCompare(b.id));
|
||||
assert.deepEqual(canonical(registry), canonical(original.registry));
|
||||
const previousAttempts = read('node/original-attempts.json'), attempts = meta('modelAttempts');
|
||||
for (const item of previousAttempts) assert.deepEqual(attempts.find(row => row.id === item.id), item);
|
||||
const publicConnection = after.modelCenter.connections.find(row => row.id === connectionId);
|
||||
assert.equal(publicConnection.credentialRevision, 1); assert.equal(publicConnection.hasKey, true);
|
||||
assert.equal(publicConnection.modelIdentityPolicy, 'record');
|
||||
assert.equal(publicConnection.modelIdentityAliasSource, 'https://api.apikey.fan');
|
||||
assert.deepEqual(publicConnection.modelIdentityAliases, []);
|
||||
assert.equal(after.modelCenter.connections.find(row => row.name === 'jojocode').enabled, false);
|
||||
|
||||
// Match the exact original metadata-row hash by retaining a raw array prefix.
|
||||
// This proves every old diagnostic field and its order survived. New external
|
||||
// diagnostic jobs are observations, never rolled back or attributed to Y.
|
||||
const db = new DatabaseSync(path.join(dataRoot(), 'workbench.sqlite'), { readOnly: true });
|
||||
let externalDiagnostics;
|
||||
try {
|
||||
const row = db.prepare("SELECT * FROM metadata WHERE key='modelNetworkDiagnostics'").get();
|
||||
const jobs = JSON.parse(row.value);
|
||||
const expected = before.database.tables.metadata.rows.find(value => value.id === row.key).sha256;
|
||||
const matches = [];
|
||||
for (let count = 0; count <= jobs.length; count++) {
|
||||
if (digest({ ...row, value: JSON.stringify(jobs.slice(0, count)) }) === expected) matches.push(count);
|
||||
}
|
||||
assert.equal(matches.length, 1, 'Every original diagnostic must match its original complete row hash.');
|
||||
externalDiagnostics = jobs.slice(matches[0]);
|
||||
assert.equal(externalDiagnostics.length, 3);
|
||||
for (const job of externalDiagnostics) {
|
||||
assert.ok(job.finishedAt); assert.equal(job.httpRequests, 0);
|
||||
assert.equal(job.modelRequests, 0); assert.equal(job.credentialAccess, false);
|
||||
}
|
||||
save('node/diagnostic-reconciliation.json', { originalRowSha256: expected, originalJobs: matches[0],
|
||||
originalRowExactAfterRemovingOnlyAppendedJobs: true, externalJobs: externalDiagnostics,
|
||||
attribution: 'Not started by the Y driver; preserve separately. No inference about the operator.' });
|
||||
const previousProjects = new Map(before.database.tables.projects.rows.map(value => [value.id, value]));
|
||||
const rows = db.prepare('SELECT * FROM projects ORDER BY rowid').all();
|
||||
for (const row of rows.filter(value => value.id !== outcome.projectId)) {
|
||||
assert.equal(digest({ ...row, position: row.position - 1 }), previousProjects.get(row.id)?.sha256,
|
||||
'Original project row, after reversing only the one owned prepend: ' + row.id);
|
||||
}
|
||||
assert.equal(rows.length, previousProjects.size + 1);
|
||||
} finally { db.close(); }
|
||||
const metadataRows = new Map(after.database.tables.metadata.rows.map(row => [row.id, row]));
|
||||
for (const row of before.database.tables.metadata.rows) {
|
||||
if (!['config', 'modelRegistry', 'modelAttempts', 'modelNetworkDiagnostics'].includes(row.id)) assert.deepEqual(metadataRows.get(row.id), row);
|
||||
}
|
||||
const project = outcome.sourceRoot;
|
||||
for (const [name, expected] of Object.entries(source.files).filter(([name]) => !name.startsWith('src/'))) {
|
||||
assert.equal(sha(fs.readFileSync(path.join(project, name))), expected, name);
|
||||
}
|
||||
const staged = execFileSync('git', ['-C', project, 'show', ':owner-staged.txt'], { encoding: 'utf8', windowsHide: true });
|
||||
assert.equal(staged, source.stagedBefore);
|
||||
for (const change of session.changes) assert.equal(sha(fs.readFileSync(path.join(project, change.path))), change.sha256);
|
||||
const plans = session.execution.contextPlans.map(plan => ({ id: plan.id, runId: plan.runId,
|
||||
requestBytes: plan.requestBytes, estimatedInputTokens: plan.estimatedInputTokens,
|
||||
reservation: plan.tokenReservation.reservedTokens, method: plan.tokenReservation.method,
|
||||
excludedSources: plan.tokenReservation.routes.reduce((sum, route) => sum + route.excludedSources.length, 0),
|
||||
actualUsage: plan.actualUsage }));
|
||||
save('node/offline-analysis.json', { plans, reads: session.execution.task.readProgress,
|
||||
tools: session.events.filter(event => event.kind === 'tool').map(event => ({ tool: event.tool, phase: event.phase })),
|
||||
checks: session.execution.checks.map(job => ({ id: job.id, status: job.status,
|
||||
results: job.results.map(value => ({ name: value.name, status: value.status, tests: value.tests })) })),
|
||||
applied: session.execution.changeSets.map(value => ({ id: value.id, status: value.status, fingerprint: value.fingerprint })),
|
||||
postApplyChecksExecuted: false, gitDeliveryCompleted: false, supervisorRanPostApplyChecks: false,
|
||||
outlierExclusionTriggered: plans.some(plan => plan.excludedSources > 0),
|
||||
noExtraModelRequests: true, implementationSuppliedBySupervisor: false });
|
||||
save('node/driver-original-log.txt', fs.readFileSync('.qa/reliability-1.25.2/live-y-node.log'));
|
||||
const reconciliation = { passed: true, engineeringSuccess: false, originalDriverCompleted: false,
|
||||
originalErrorPreserved: true, configRestorationHadAlreadyCompletedBeforeOriginalAssertion: true,
|
||||
originalExportsExact: before.exports.length, originalArtifactFilesExact: before.persistent.artifacts.length,
|
||||
originalProjectRowsExactAfterOneOwnedPrependReversal: true, originalRowsExact: ['sessions', 'records', 'events', 'outbox'],
|
||||
originalAttemptsExact: previousAttempts.length, rawConfigExceptRevisionExact: true,
|
||||
registryExceptMonotonicRevisionsTimestampOrderExact: true, otherMetadataRowsExact: true,
|
||||
appendedExternalDiagnostics: externalDiagnostics.map(value => value.id), noExternalChangesRolledBack: true,
|
||||
budgetsSelectionPoolsAndIdentityRestoredAtOriginalFinalization: true,
|
||||
externalPoolAdditionAcrossOpeningSnapshotAndConfigFreezePreserved: 'DeepSeek; pool revision 6 to 7', credentialRevision: 1, hasKey: true,
|
||||
ownerBytesAndStagingExact: true, pid: after.pid, version: after.version, inspectorClosed: true,
|
||||
observedAt: new Date().toISOString(), realModelRequests: 0, commandsExecutedInSample: 0 };
|
||||
save('node/reconciliation.json', reconciliation);
|
||||
console.log(JSON.stringify(reconciliation));
|
||||
} finally { await close(); }
|
||||
@@ -0,0 +1,43 @@
|
||||
// Offline only: validates sealed bytes and the reported outcome without
|
||||
// contacting the host or executing the model, project checks, or Git delivery.
|
||||
import fs from 'node:fs';
|
||||
import path from 'node:path';
|
||||
import assert from 'node:assert/strict';
|
||||
import { createHash } from 'node:crypto';
|
||||
const root = path.resolve('.qa/live-delivery-y');
|
||||
const read = name => JSON.parse(fs.readFileSync(path.join(root, name)));
|
||||
const hash = bytes => createHash('sha256').update(bytes).digest('hex');
|
||||
const manifest = read('MANIFEST.json');
|
||||
for (const row of manifest.files) {
|
||||
const file = path.resolve(root, row.path);
|
||||
assert.ok(file.startsWith(root + path.sep));
|
||||
const bytes = fs.readFileSync(file);
|
||||
assert.equal(bytes.length, row.bytes, row.path);
|
||||
assert.equal(hash(bytes), row.sha256, row.path);
|
||||
}
|
||||
const preflight = read('preflight/result.json'), result = read('node/model-outcome.json');
|
||||
assert.equal(preflight.passed, true); assert.equal(preflight.requests, 1);
|
||||
assert.equal(preflight.usage.total_tokens, 110);
|
||||
assert.equal(result.success, false); assert.equal(result.stopReason, 'step_limit');
|
||||
assert.equal(result.usage.requests, 8); assert.equal(result.usage.totalTokens, 64500);
|
||||
assert.equal(result.usage.unknownRequests, 0); assert.equal(result.identitiesValid, true);
|
||||
assert.equal(result.applications.length, 1);
|
||||
assert.equal(fs.existsSync(path.join(root, 'node/result.json')), false);
|
||||
assert.equal(fs.existsSync(path.join(root, 'node/restoration.json')), false);
|
||||
for (const capture of [...preflight.captures, ...result.captures]) {
|
||||
const bytes = fs.readFileSync(path.join(root, capture.path));
|
||||
assert.equal(bytes.length, capture.bytes); assert.equal(hash(bytes), capture.sha256);
|
||||
}
|
||||
const state = read('node/session.json');
|
||||
assert.equal(state.execution.deliveries.length, 0);
|
||||
assert.equal(state.execution.checks.length, 1);
|
||||
assert.equal(state.execution.checks[0].results.find(value => value.name === 'test').tests.failed, 9);
|
||||
assert.equal(state.execution.changeSets[0].status, 'completed');
|
||||
assert.equal(read('node/reconciliation.json').passed, true);
|
||||
assert.equal(hash(fs.readFileSync(path.join(root, 'node/goal.txt'))), 'b5421d0bb502126f49a5153729027f13ef82db86ab19a9e5b0c23a49070bb905');
|
||||
const sourceText = fs.readFileSync(path.join(root, 'driver-source/node.mjs'), 'utf8');
|
||||
const sourceGoal = sourceText.slice(sourceText.indexOf(' const goal = '), sourceText.indexOf(" save('node/goal.txt'"));
|
||||
assert.equal(hash(sourceGoal), 'ea3eaee0a94112085b6e41f528397f32fb2640a963a40add9bd3549b9eb642e0');
|
||||
console.log(JSON.stringify({ passed: true, files: manifest.files.length, bytes: manifest.totalBytes,
|
||||
requests: 9, reportedTokens: 64610, engineeringSuccess: false, originalFinalizationFailed: true,
|
||||
readOnlyReconciliationPassed: true, modelRequestsExecutedByVerifier: 0 }));
|
||||
+2
-1
@@ -8,7 +8,8 @@
|
||||
2. [complete] inventory恢复每文件treeFile检查,扫描后新增硬链接被拒,原身份/字节一致;12/12安装通过,关联11项通过与升级3/3复验。10971文件只读ABBA全校验旧8.15–8.45s、新9.96–11.08s,未恢复逐文件祖先遍历。
|
||||
3. [complete] 混合进展按同轮实际新增区间结算,独立无变化保护保持;新v3读取批次释放小来源剩余预算给文件/日志、变更操作原序不动。联合43/43、补强2/2及吞吐最终38/38与构建通过,首失败保留。
|
||||
4. [complete] 1.25.2/source d510不可变独立重建22/22、便携/隔离安装12组+75UI、正式保留PID36552升级及网页/原生后全部原状态exact。完整898/899+原断言1/1,原超时保留;1339份证据离线通过,阶段回顾及限定本地提交。
|
||||
5. [in_progress] X预冻结因原jojocode外部停用而拒绝(0请求);用户明确改用已启用apikey、另冻Y。先按Y计划单次诊断/身份预检/有界Node,通过后才React/Python及Purge AB/BA;不改提示/预算筛成功。
|
||||
5. [complete] X预冻结拒绝0请求保持;用户选择另冻Y,预检通过、Node8请求64500tokens三源码模型封存并应用,step_limit前未检查/交付。原诊断元数据收尾失败、开场外部池变化及两次只读复核错误保持;独立对账通过,不续跑Y。
|
||||
6. [in_progress] 用户新指定apikey-Claude/DeepSeek,另冻逐连接单次身份预检;通过才相同原题新Node。旧样本不续跑,不增加预算筛成功;Node闭环后才React/Python和Purge。
|
||||
|
||||
细节和阶段回顾见docs/RELIABILITY-1.25.2.md。开始时session-catchup无待同步输出;计划文件首次补丁因findings标题上下文不符被拒且无文件变化,已按原标题修正。以下按历史时点理解。
|
||||
|
||||
|
||||
Reference in New Issue
Block a user