test(live): seal AD autonomous Node success and preserved provider failures

This commit is contained in:
Agent Workbench Dev
2026-11-06 12:57:20 +08:00
parent 3b1510f83e
commit f5b0ffee1c
9 changed files with 302 additions and 2 deletions
+2
View File
@@ -1,5 +1,7 @@
# Agent Workbench项目执行规范
AD已封存:正式1.25.8/PID71628,五线10真实预检8通过;jojocode目录ECONNRESET,Claude目录UND_ERR_SOCKET及两次HTTP400缺Messages必填max_tokens原错保留。新DeepSeek Node11请求180554报告tokens、0unknown、原9测试/build/仅三模块Git88bb25cb及owner/index全过,0人工审批/继续,原高性能池已恢复。全批21请求199250报告tokens+2unknown/338本地估算,不补造计费;216份60197251字节封存、原46111行/105导出/工件精确对账,配置仅声明修订/连接编辑时间与顺序变化,9229关闭。见docs/QA-LIVE-VALIDATION-AD.md及live-validation-ad-evidence.json。7次不同文件读取、30工具schema稳定、上游122624缓存输入tokens仅本例观测;内置delivery_report仍in_progress(终端Git未登记delivery_job)如实保留。下一另冻AE React/Python,过后才原版Purge,不续跑AD/旧样本或用户项目,不委派/推送/触Key、不恢复默认额度。
C随1.25.8已正式交付:工具登记/装配/可见原因与schema摘要、前缀v2首次变化段和消息/主线诊断归属,新诊断历史独立保存。来源a6e31dec/候选20261106042247703;原41相关40过1夹具UUID失败保持,最终8/8和升级3/3、717输入重建22/22、安装副本两端各4组及正式两端回看通过。唯一正式安装319241ms,保留PID71628/instance37d6b536-c0dc-4cbf-9c51-095bbe779062/Provider、五条连接Key存在状态和85项目105会话全部原行/导出/工件/模型中心exact,9229关闭。见docs/QA-TOOL-PREFIX-C.md与tool-prefix-c-1.25.8-release-identity.json。下一新AD驱动先本机验证,再冻结五供应商文本/工具兼容及独立Node;Node闭环后才React/Python/Purge,不续跑原样本/用户项目,不委派/推送/触Key。默认unrestricted/full/passthrough不退回有限额度,真实模型/网络/重复读取/缓存/真实预留/Purge未知保持。下方旧版本按历史时点理解。
1.25.7已正式本地交付:程序ebd605f9/候选20261106020800751;旧设置窗口与演示选择入口移除,既有有效连接按真实型号或保存名统一编辑/参数/启停,不改内部ID与凭据归属。保留PID71628/instance37d6b536-c0dc-4cbf-9c51-095bbe779062、Provider、选模和85项目105会话全部原行/导出/工件;公开模型中心仅旧连接生成名称/品牌的声明投影变化。72相关、18发行、713输入重建22/22、安装副本网页/Electron各5组及正式两端通过。见docs/QA-UNIFIED-MODELS-1.25.7.md与unified-models-1.25.7-release-identity.json;下方待发行段按历史时点。只本地提交、不委派/推送,不触真实Key/续跑原项目;真实模型/网络/重复读取/缓存/预留/Purge未知保持。
+2
View File
@@ -1,5 +1,7 @@
# 1.25 计划:任务要求与交付复核
新AD已完成:五线十次真实预检8通过、两次Claude上游400缺必填max_tokens,目录两处网络异常保持;DeepSeek新Node自主11请求、原九测试/build/三源码实际Git交付成功,0人工审批/继续。全批21请求199250报告tokens+2unknown,216份原证据及原状态对账完成,见[AD验收](QA-LIVE-VALIDATION-AD.md)。本例30工具schema保持、7次不同文件读取及122624上游缓存输入tokens不作普遍收益;终端Git未登记内置交付账本的缺口单列。下一另冻AE React/Python,通过后才原版Purge AB/BA;历史失败/真实预留边界/Purge收益未知保持。
C已随1.25.8正式交付:工具可见性/schema/代次、prefixAudit v2首次变化段和消息/主线诊断归属,修复新诊断混入主历史。来源a6e31dec;最终8/8、升级3/3、717输入重建22/22、安装副本与正式网页/Electron通过。原PID71628/Provider、五条连接凭据存在状态和85项目105会话全部原行/导出/工件/模型中心exact。见[C验收与回顾](QA-TOOL-PREFIX-C.md)。下一新AD先验证驱动、冻结供应商兼容及独立Node,之后按前置推进React/Python/Purge;D–F未据此完成,网络/自主交付/重复读取/缓存/真实预留/Purge未知保持。下方为历史时点。
当前正式1.25.7已交付:统一既有模型连接编辑/参数/启停,移除旧窗口与演示入口;原PID71628、Provider、选模和85项目105会话全部原行/导出/工件保持。相关72/72、发行18/18、源码ZIP重建22/22、安装副本与正式网页/Electron通过,见[本次验收](QA-UNIFIED-MODELS-1.25.7.md)。C–F及真实模型/网络/重复读取/缓存/Purge未验证项保持;下方为历史时点。
+7 -1
View File
@@ -22,6 +22,12 @@ Claude两次原400响应、请求ID和完整152字节错误工件均保存;上
## 独立工程与未验证项
Node即将按已冻结契约使用DeepSeek开展一次新工程;不会按其它线路补跑同一失败样本,也不供应实现答案或人工继续。Node通过前,React/Python/原版Purge AB/BA未执行,收益unknown。
Node已按冻结契约使用DeepSeek完成真实自主交付:11请求、171841输入/8713输出/180554上游报告tokens、0unknown,全部完整matched。模型自行读取7个不同文件,提出并自动应用三份源码,规划并运行原检查,再以终端仅提交三个源码文件;实际提交88bb25cbf37739359c991dddac26e296a4d29d18。独立重跑原九测试9/9、build退出0,当前文件字节与提交一致,owner文件及原暂存/未暂存/未跟踪状态全部保留。没有人工批准、实现提示、追加指导或继续。本样本允许原生终端、默认不限额度,与旧16K/2048样本存在多项变化,不作单因素比较。
内置delivery_report仍为in_progress:本次终端Git没有生成内置delivery_job,不能用该报告冒称内置交付账本已闭环。本次工程结论依赖独立Git原件及原测试,不抹掉这个产品整合缺口。11请求实际发送30工具、定义SHA与15121字节保持一致,每请求已逐项核对原HTTP工具schema与prefixAudit v2;读取7次/7不同文件,本例未见重复原输入。DeepSeek原响应报告122624缓存输入tokens(171841输入),这仅是上游观测,不证明C带来的因果节省或普遍不掉缓存。
工程后原状态审计第二轮通过:原46111行、105导出和所有原工件保持;本批合计新增6项目/11会话/877事件及outbox/218执行记录/132去重工件全部归属本批。原高性能池恢复;DeepSeek revision9→11、configRevision110→112、updatedAt及公开编辑移到列表末尾按契约单列,所有其它配置字段/连接内容及凭据存在/修订保持。首审计错误为驱动假设JSON有末尾换行,实际atomicJson无换行;原失败node-ed8ff4fa保留,按真实序列化精确逆算原SHA后新审计node-577d9308通过,没有改动磁盘或重发模型。
本批21次真实推理,19次有用量、合计199250报告tokens;两次unknown的本地估算338只作未知账本、不作计费。证据索引见[live-validation-ad-evidence.json](live-validation-ad-evidence.json)。下一独立冻结React/Python;原版Purge AB/BA仍待它们通过,收益unknown,不在AD原样本中续跑或倒改结论。
本次短预检通过仅说明所观测协议/型号标识/文本及工具结构可用,不证明权重、长期稳定、自主工程交付、普遍重复读取改善、缓存收益或真实异常预留排除边界。所有历史失败和原有限预算样本保持。
+2
View File
@@ -1,5 +1,7 @@
# Tianshu 研究后的 Agent Workbench 采用计划
新AD已完成:五线十次真实预检8通过、两次Claude上游400缺必填max_tokens,目录两处网络异常保持;DeepSeek新Node自主11请求、原九测试/build/三源码实际Git交付成功,0人工审批/继续。全批21请求199250报告tokens+2unknown,216份原证据及原状态对账完成,见[AD验收](QA-LIVE-VALIDATION-AD.md)。本例30工具schema保持、7次不同文件读取及122624上游缓存输入tokens不作普遍收益;终端Git未登记内置交付账本的缺口单列。下一另冻AE React/Python,通过后才原版Purge AB/BA;历史失败/真实预留边界/Purge收益未知保持。
C已随1.25.8正式交付:工具可见性/schema/代次、prefixAudit v2首次变化段和消息/主线诊断归属,修复新诊断混入主历史。来源a6e31dec;最终8/8、升级3/3、717输入重建22/22、安装副本与正式网页/Electron通过。原PID71628/Provider、五条连接凭据存在状态和85项目105会话全部原行/导出/工件/模型中心exact。见[C验收与回顾](QA-TOOL-PREFIX-C.md)。下一新AD先验证驱动、冻结供应商兼容及独立Node,之后按前置推进React/Python/Purge;D–F未据此完成,网络/自主交付/重复读取/缓存/真实预留/Purge未知保持。下方为历史时点。
当前正式1.25.7已交付:统一既有模型连接编辑/参数/启停,移除旧窗口与演示入口;原PID71628、Provider、选模和85项目105会话全部原行/导出/工件保持。相关72/72、发行18/18、源码ZIP重建22/22、安装副本与正式网页/Electron通过,见[本次验收](QA-UNIFIED-MODELS-1.25.7.md)。C–F及真实模型/网络/重复读取/缓存/Purge未验证项保持;下方为历史时点。
+246
View File
@@ -0,0 +1,246 @@
{
"summary": {
"schema": 1,
"driverCommit": "ed78f02c032f2968e326fb8f696eba1281e0785b",
"version": "1.25.8",
"realModelRequests": 21,
"reportedRequests": 19,
"reportedTokens": 199250,
"unknownRequests": 2,
"unknownLocalEstimate": 338,
"node": {
"success": true,
"commit": "88bb25cbf37739359c991dddac26e296a4d29d18",
"requests": 11,
"usage": {
"requests": 11,
"inputTokens": 171841,
"outputTokens": 8713,
"totalTokens": 180554,
"unknownRequests": 0,
"estimatedUnknownTokens": 0
},
"independentChecks": {
"test": {
"status": 0,
"signal": null,
"error": null,
"stats": {
"tests": 9,
"pass": 9,
"fail": 0,
"cancelled": 0,
"skipped": 0,
"todo": 0
},
"stdoutSha256": "c3bbf67e62ae04805face4ea745574b8d64a243ac421553271950215c9ba140c",
"stderrSha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
},
"build": {
"status": 0,
"signal": null,
"error": null,
"stats": null,
"stdoutSha256": "9307ee4e4f72392dd2f29f26c8e9fd3c8e55353bbe21f93627f260f108be9c2c",
"stderrSha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
}
},
"allIdentitiesMatched": true,
"ownerBytesAndIndexPreserved": true,
"reads": [
{
"path": "test/expense.test.mjs"
},
{
"path": "src/amount.mjs"
},
{
"path": "src/summary.mjs"
},
{
"path": "src/report.mjs"
},
{
"path": "scripts/build.mjs"
},
{
"path": "README.md"
},
{
"path": "package.json"
}
],
"duplicateReadInputCount": 0,
"reportedCachedTokens": 122624,
"inputTokens": 171841,
"terminalGit": true,
"internalDeliveryReportStatus": "in_progress",
"internalReportIsNotIndependentGitProof": true,
"supervisorApprovals": 0,
"supervisorContinuations": 0,
"elapsedMs": 60424
},
"prefixProofs": [
{
"attemptId": "d5da6c54-4537-4203-9077-65dcd10edb8b",
"planId": "055d792e-15ae-4216-a4a3-4bd8b0626eb5",
"toolCount": 30,
"toolSha256": "8f2ad087b3da842978ca5195ef3cad7fa1941803a1ccd48813ec55dd6af104bb",
"commonPrefixBytes": 0,
"firstChangedPath": "model"
},
{
"attemptId": "e4d9710f-3007-499c-8678-4f1a67d07068",
"planId": "becb4720-b770-4ca7-a277-3c245e5d8b9c",
"toolCount": 30,
"toolSha256": "8f2ad087b3da842978ca5195ef3cad7fa1941803a1ccd48813ec55dd6af104bb",
"commonPrefixBytes": 18586,
"firstChangedPath": "messages[2]"
},
{
"attemptId": "c21c8b76-ee68-4b33-9ca3-6555014b2b18",
"planId": "249b2250-5ac8-4ae9-b1bd-10cf71090637",
"toolCount": 30,
"toolSha256": "8f2ad087b3da842978ca5195ef3cad7fa1941803a1ccd48813ec55dd6af104bb",
"commonPrefixBytes": 23559,
"firstChangedPath": "messages[5]"
},
{
"attemptId": "597a345a-4b39-4f38-a03c-61188ef2b426",
"planId": "4ff062ed-5a5b-466b-86db-445ee84526d0",
"toolCount": 30,
"toolSha256": "8f2ad087b3da842978ca5195ef3cad7fa1941803a1ccd48813ec55dd6af104bb",
"commonPrefixBytes": 27011,
"firstChangedPath": "messages[10]"
},
{
"attemptId": "5dfd0750-fc44-41d7-b82f-5e1e31a56d4b",
"planId": "e5254528-9e58-48e6-90d6-031478be837f",
"toolCount": 30,
"toolSha256": "8f2ad087b3da842978ca5195ef3cad7fa1941803a1ccd48813ec55dd6af104bb",
"commonPrefixBytes": 43312,
"firstChangedPath": "messages[13]"
},
{
"attemptId": "b1e39d27-4eee-4664-b3e2-25da202a4b05",
"planId": "39ec66b9-6165-4d48-8b78-9010738b88d3",
"toolCount": 30,
"toolSha256": "8f2ad087b3da842978ca5195ef3cad7fa1941803a1ccd48813ec55dd6af104bb",
"commonPrefixBytes": 50213,
"firstChangedPath": "messages[17]"
},
{
"attemptId": "fab4161c-0390-4f8e-8fb8-6ac3dbe5d3b8",
"planId": "d33dacb4-8602-4f70-a0b2-5ffb72ed3038",
"toolCount": 30,
"toolSha256": "8f2ad087b3da842978ca5195ef3cad7fa1941803a1ccd48813ec55dd6af104bb",
"commonPrefixBytes": 54721,
"firstChangedPath": "messages[19]"
},
{
"attemptId": "f1134982-11e9-476f-9f85-c85640acd501",
"planId": "f68447bb-a090-4465-af6c-b8b73ede55fa",
"toolCount": 30,
"toolSha256": "8f2ad087b3da842978ca5195ef3cad7fa1941803a1ccd48813ec55dd6af104bb",
"commonPrefixBytes": 57143,
"firstChangedPath": "messages[21]"
},
{
"attemptId": "7b49e6a2-0e9c-4390-a8d7-1cbe4b2df083",
"planId": "73f6e14b-afaf-4489-9ceb-a46acded939d",
"toolCount": 30,
"toolSha256": "8f2ad087b3da842978ca5195ef3cad7fa1941803a1ccd48813ec55dd6af104bb",
"commonPrefixBytes": 58746,
"firstChangedPath": "messages[23]"
},
{
"attemptId": "68bad6fe-e9dc-4543-ad22-618800c07355",
"planId": "be94f73d-c2bc-473e-a127-05b34bc967d3",
"toolCount": 30,
"toolSha256": "8f2ad087b3da842978ca5195ef3cad7fa1941803a1ccd48813ec55dd6af104bb",
"commonPrefixBytes": 62324,
"firstChangedPath": "messages[25]"
},
{
"attemptId": "edbb488b-1524-4fd3-8196-281abd17840c",
"planId": "ce67447f-1e6b-40c1-849c-aab8780b9421",
"toolCount": 30,
"toolSha256": "8f2ad087b3da842978ca5195ef3cad7fa1941803a1ccd48813ec55dd6af104bb",
"commonPrefixBytes": 67475,
"firstChangedPath": "messages[27]"
}
],
"originalStateAudit": {
"success": true,
"stage": "node",
"version": "1.25.8",
"pid": 71628,
"originalProjects": 85,
"originalSessions": 105,
"addedProjects": 6,
"addedSessions": 11,
"addedRows": {
"events": 877,
"metadata": 0,
"outbox": 877,
"projects": 6,
"records": 218,
"sessions": 11,
"sqlite_sequence": 0
},
"newArtifactCount": 132,
"checkedOriginalRows": 46111,
"originalSessionExports": "exact",
"originalArtifactBytes": "exact",
"originalRows": "exact except explicitly reconciled project positions, event sequence and four public metadata rows",
"configComparison": {
"original": 110,
"current": 112,
"allOtherFields": "exact"
},
"registryChanges": {
"connection": {
"id": "1a7a55a9-29f2-4b9f-8ea2-5a55afd7eacd",
"originalRevision": 9,
"currentRevision": 11,
"originalUpdatedAt": "2026-11-06T01:18:17.613Z",
"currentUpdatedAt": "2026-11-06T04:53:05.522Z",
"explicitIdentityDefaults": [],
"originalOrder": [
"1a7a55a9-29f2-4b9f-8ea2-5a55afd7eacd",
"5b4daed9-b2f7-4de0-b949-573eba70c7b3",
"93d0cf6d-4f32-4918-9da8-6f125aa8a6e5",
"09b0293b-faad-45c3-832d-7c8b615813aa",
"9ae348a5-d0a6-42ed-8664-0da5f9effe34"
],
"currentOrder": [
"5b4daed9-b2f7-4de0-b949-573eba70c7b3",
"93d0cf6d-4f32-4918-9da8-6f125aa8a6e5",
"09b0293b-faad-45c3-832d-7c8b615813aa",
"9ae348a5-d0a6-42ed-8664-0da5f9effe34",
"1a7a55a9-29f2-4b9f-8ea2-5a55afd7eacd"
]
}
},
"oldCredentialRow": "exact hash, never decrypted or copied",
"oldModelAttempts": "every original entry exact",
"newModelAttempts": 21,
"inspectorClosed": true,
"keysReadOrResaved": false,
"reactPythonExecuted": false,
"purgeExecuted": false,
"nodePassed": true
},
"reactPythonExecuted": false,
"purgeExecuted": false,
"cacheCausalBenefit": "unknown",
"historicalSamplesReplayed": false,
"keysReadOrResaved": false
},
"evidence": {
"root": "D:\\Ai-project\\REA逆向\\Agent-Workbench\\.qa\\live-validation-ad",
"files": 216,
"bytes": 60197251,
"manifestSha256": "1c21f6b82bd88c1860b77b783d2556a449a76f0b5e21751ea710ab40f9419440"
}
}
+3
View File
@@ -1299,3 +1299,6 @@ 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
第二候选b8完成三归档、713输入ZIP重建22/22和开放SDK4组;完整便携发现REA非法detail=full,改为真实协议complete并新增真实REA分析/导出,32相关/build-18通过。实际安装10生命周期后要求UI路由自造死锁,等待自动刷新并只持有一次手动请求后同安装副本7/7通过;旧失败与两次QA卸载数据exact收据保持。当前编译结果完整便携预验退出0,仅两个server编译文件变化。下一提交新来源、独立发行/实际安装,再执行已授权正式升级/full/open/实际重启;before不重做,原976轮958/18和历史真实模型/Purge未知保持。
# AD真实结果
Node新样本已真实交付:11请求180554tokens,7次读取分别属于不同文件,30工具schema原HTTP与prefixAudit v2逐次保持,上游报告122624缓存输入tokens。不能据此归因C收益或普遍不重复读。终端Git成功且owner/index保持,但内置delivery_report仍in_progress,未登记delivery_job;实际Git独立证据与产品报告状态分别记录。Claude原生Messages要求max_tokens,完全省略字段会原样400;不加任意额度或改写历史。两网络目录失败保留。完整阶段见QA-LIVE-VALIDATION-AD。
+2
View File
@@ -1938,3 +1938,5 @@ c9c77c7e首完整候选完成三归档、710输入ZIP离线重建22/22和实际S
正式安装session7380退出0,319241ms;原PID71628/instance/Provider及原数据、模型中心exact。网页/Electron正式只读回看均过,之后再次原状态exact。83份46924092字节阶段证据逐SHA核对,未重打包历史原件,真实模型0。
AD驱动初次本机夹具失败在连接health由unverified转responded的错误exact断言;原结果ad-driver-5e252008保留,尚未冻结/执行真实AD。
AD现已完成并封存216份60197251字节。真实兼容8/10,jojocode/Claude目录网络异常及Claude缺max_tokens的两次400保持。DeepSeek新Node11请求180554tokens、9/9/build/三模块Git88bb25cb、全部owner/index通过,0人工审批/继续;原高性能池恢复。首Node审计末尾换行假设错误保留,按真实atomicJson精确逆算原SHA后审计node-577d9308通过,原46111行/105导出/工件保持。下步另冻AE React/Python,Purge仍待前置。
+1 -1
View File
@@ -94,7 +94,7 @@ try {
if (original.path === 'config.json' && actual.sha256 !== original.sha256) {
const actualConfig = JSON.parse(fs.readFileSync(path.join(host.identity.dataRoot,original.path),'utf8'));
assert.deepEqual(actualConfig,metaAfter.config);
assert.equal(sha(JSON.stringify({...actualConfig,configRevision:metaBefore.config.configRevision},null,2)+'\n'),original.sha256,'Original config bytes reconstructed exactly after the declared revision increment.');
assert.equal(sha(JSON.stringify({...actualConfig,configRevision:metaBefore.config.configRevision},null,2)),original.sha256,'Original config bytes reconstructed exactly with atomicJson serialization after the declared revision increment.');
} else assert.equal(actual.sha256,original.sha256);
}
const newArtifactCount = after.persistent.artifacts.length-before.persistent.artifacts.length;
+37
View File
@@ -0,0 +1,37 @@
import fs from 'node:fs';
import path from 'node:path';
import assert from 'node:assert/strict';
import { root, read, save, sha } from './common.mjs';
const node = read('node/result.json'), state = read('node/session.json'), compatibility = read('compatibility/result.json');
const audit = read('audits/node-577d9308-d5b6-4bd2-85d3-0ad8cb5255bf/result.json'); assert.equal(audit.success,true);
const prefixes = compatibility.results.flatMap(row=>row.preflights.map(value=>'compatibility/'+row.connectionId+'/'+value.scope)).concat(['node']);
const all = [], proofs = [];
for (const prefix of prefixes) {
const rows = read(prefix+'/attempts.json'); all.push(...rows);
for (const row of rows) for (const kind of ['request','response']) {
const names = fs.readdirSync(path.join(root,prefix,'captures')).filter(file=>file.startsWith(row.id+'-'+kind)); assert.equal(names.length,1);
const name=prefix+'/captures/'+names[0],bytes=fs.readFileSync(path.join(root,name)); assert.equal(sha(bytes),row[kind+'Sha256']);
if (kind==='request') {
const body=JSON.parse(bytes); assert.ok(['max_tokens','max_output_tokens','max_completion_tokens'].every(key=>!Object.hasOwn(body,key)));
if (prefix==='node') {
const plan=state.execution.contextPlans.find(value=>value.id===row.requestId),variant=plan.prefixAudit.variants.find(value=>value.connectionId===row.connectionId);
assert.equal(sha(JSON.stringify(body.tools)),variant.toolsSha256); assert.equal(Buffer.byteLength(JSON.stringify(body.tools)),variant.toolBytes);
proofs.push({attemptId:row.id,planId:plan.id,toolCount:body.tools.length,toolSha256:variant.toolsSha256,commonPrefixBytes:variant.commonPrefixBytes,firstChangedPath:variant.firstChange.path});
}
}
}
}
const nodeAttempts=read('node/attempts.json'),reads=state.events.filter(row=>row.tool==='read_file'&&row.phase==='completed').map(row=>row.input);
const summary={schema:1,driverCommit:read('current.json').driverCommit,version:'1.25.8',realModelRequests:all.length,reportedRequests:all.filter(row=>row.usageStatus==='reported').length,
reportedTokens:all.reduce((sum,row)=>sum+(row.usage?.total_tokens??0),0),unknownRequests:all.filter(row=>row.usageStatus==='unknown').length,
unknownLocalEstimate:compatibility.results.flatMap(row=>row.preflights).filter(row=>row.usageStatus==='unknown').reduce((sum,row)=>sum+read('compatibility/'+compatibility.results.find(value=>value.preflights.includes(row)).connectionId+'/'+row.scope+'/invocation.json').result.usage.estimatedUnknownTokens,0),
node:{success:node.success,commit:node.commit,requests:node.requests,usage:node.usage,independentChecks:node.final,allIdentitiesMatched:node.identitiesPassed,ownerBytesAndIndexPreserved:node.allOriginalOwnerBytesAndIndexPreserved,
reads,duplicateReadInputCount:reads.length-new Set(reads.map(row=>JSON.stringify(row))).size,reportedCachedTokens:nodeAttempts.reduce((sum,row)=>sum+(row.usage?.prompt_tokens_details?.cached_tokens??0),0),inputTokens:node.usage.inputTokens,
terminalGit:true,internalDeliveryReportStatus:'in_progress',internalReportIsNotIndependentGitProof:true,supervisorApprovals:0,supervisorContinuations:0,elapsedMs:Date.parse(state.execution.run.finishedAt)-Date.parse(state.execution.run.startedAt)},
prefixProofs:proofs,originalStateAudit:audit,reactPythonExecuted:false,purgeExecuted:false,cacheCausalBenefit:'unknown',historicalSamplesReplayed:false,keysReadOrResaved:false};
save('summary.json',summary);
const files=[];function walk(dir){for(const entry of fs.readdirSync(dir,{withFileTypes:true})){const file=path.join(dir,entry.name);if(entry.isDirectory())walk(file);else if(entry.isFile()){const bytes=fs.readFileSync(file);files.push({path:path.relative(root,file).replaceAll('\\','/'),bytes:bytes.length,sha256:sha(bytes)});}else throw Error('Nonregular evidence entry.');}}
walk(root);files.sort((a,b)=>a.path.localeCompare(b.path));save('FINAL-MANIFEST.json',{schema:1,files,bytes:files.reduce((sum,row)=>sum+row.bytes,0),scope:'All original AD source/requests/responses/snapshots/audits, including failed audits. Previous historical evidence is not repackaged.'});
for(const row of files){const bytes=fs.readFileSync(path.join(root,row.path));assert.equal(bytes.length,row.bytes);assert.equal(sha(bytes),row.sha256);}
const manifest=fs.readFileSync(path.join(root,'FINAL-MANIFEST.json'));fs.writeFileSync('docs/live-validation-ad-evidence.json',JSON.stringify({summary,evidence:{root,files:files.length,bytes:files.reduce((sum,row)=>sum+row.bytes,0),manifestSha256:sha(manifest)}},null,2)+'\n',{flag:'wx'});
console.log(JSON.stringify({requests:summary.realModelRequests,reportedTokens:summary.reportedTokens,unknown:summary.unknownRequests,nodePassed:node.success,files:files.length,bytes:files.reduce((sum,row)=>sum+row.bytes,0)}));