可实际运行的宿主接入样例，用不同语言和目录检验显式登记、DELTA 选测和统一入口。
backend/calc/ 是 Go 库，checks/go-contract/ 是独立 Go 测试；client/view/ 是 TS 库，checks/ui/ 是独立测试；jobs/normalize.py 与 checks/scripts/normalize_test.py 是独立脚本及其测试。contracts/invoice.json 显式连接 Go 和 TS 两个消费者。
需要 Git、Python ≥3.9、curl 和可下载固定发布/SDK 的网络。harness 使用 chrono-harness v0.1.0-beta.20 的预构建二进制；.chrono-harness/distribution.json 锁定发布清单、平台与摘要，生成的 install.py 执行统一安装。宿主不保留 harness 源码包，不需要 Rust。宿主 Go/Node/TypeScript SDK 独立登记在 bootstrap.json，支持 macOS arm64、Linux x86_64。
python3 .chrono-harness/bootstrap.py . --profile rates
.chrono-harness/bin/chrono-harness check --config .chrono-harness/ci/check.json --base <完整基线SHA> --candidate <完整候选SHA> --unit rates候选必须是当前干净提交。首个无父提交使用 --initial --candidate <SHA>，代替 --base。单元报告由工具写入 .chrono-harness/state/units/<单元ID>/check.json。CI 的检测步骤使用完全相同的 harness argv；平台和 SDK 读数由 bootstrap-result.json 记录，指令相同不宣称环境完全等价。
每个登记单元各有独立 workflow、checkout、检查状态与重跑入口；全部工作流由同一配置生成。启动 SDK 由 bootstrap.json 的同名 profile 显式选择，不按语言或目录推断。上面的命令以一个单元为例，把 --profile 与 --unit 换成同一个登记 ID 即可。
| 单元 ID | 完整测试计划 | Workflow |
|---|---|---|
rates |
test:rates-tests |
.github/workflows/chrono-ci-rates.yml |
labels |
test:labels-tests |
.github/workflows/chrono-ci-labels.yml |
normalize |
test:normalize-tests |
.github/workflows/chrono-ci-normalize.yml |
harness |
test:bootstrap-tests |
.github/workflows/chrono-ci-harness.yml |
chrono / collection 等待并下载同一候选、事件和基线的各单元原始结果，然后用与本地相同的命令判定完整性：
python3 .chrono-harness/bootstrap.py . --profile collection
.chrono-harness/bin/chrono-harness check --config .chrono-harness/ci/check.json --base <完整基线SHA> --candidate <完整候选SHA> --collect .chrono-harness/state/collection/manifest.json本地在独立 checkout 并发运行单元，复制原始报告并显式登记 manifest 的单元、路径、报告摘要与 runner/judge 摘要；完整格式见 CI 单元合同。汇总不重跑业务测试；缺失、重复、陈旧或失败结果不能通过。无 DELTA 的单元标明无需运行产品测试，不代表其它单元通过。
一个单元失败不取消其它 workflow。修复并重跑该单元后，再重跑 collection；其它单元保持原 run。自动收集限 push/PR，手动 dispatch 组合使用显式 manifest。各单元准备输入期间若 integration 基线移动导致不一致，汇总明确失败，不复用错配结果。
backend/calc/rate.go→test:rates-tests；client/view/label.mts→test:labels-tests；contracts/invoice.json→ 两者；jobs/normalize.py→test:normalize-tests。- README.md 修改 → 不运行产品测试；未登记文件 → 失败；实际测试非零 → harness 非零。
- 项目配对、命令参数和依赖均在
.chrono-harness/登记；不从扩展名、import、module 或目录推断选测。 CLAUDE.md由原子规则生成；AGENTS.md是其相对符号链接。修改指令源后运行.chrono-harness/bin/chrono-instructions generate --host-root .。- CI 修改入口是
.chrono-harness/ci/units.json；运行.chrono-harness/bin/chrono-ci generate --host-root . --config .chrono-harness/ci/units.json。受管工作流由chrono-ci verify检查。push_baselines显式登记 integration 分支的所有 push 以 origin/dev 为基线，后续修复推送仍覆盖整个分支差异；dev 保持事件的 before/after。
本仓运行现役 chrono-ci-check/v3 scoped profile，验证实际命令、显式选测、干净快照和退出结果。它不冒充完整七判官治理，不证明工具链输入闭包完整或测试依赖无遗漏。该版本的通用 projects 判官已允许省略 manifest/lock/root 并接受任意显式 action；语言专属检查由可选适配器承载。本仓完整治理的接入仍未完成。
本地与 CI 通过相同 bootstrap 从公开 Release 校验安装；不需要跨仓凭据。安装结果见 .chrono-harness/state/distribution.json，语言 SDK 版本另见 bootstrap-result.json。成本暂未测量，登记为 unknown。
主库与示例索引 列出 Go、TS 和混合语言宿主。
The pinned release also installs chrono-worktree. The host policy is
.chrono-harness/worktree.json; shared registries explicitly declare dev,
feature/, integration/, file ownership and artifacts. Project and FILEMAP
registrations have one owner; directory names and languages do not select work.
The pinned beta.20 release supports the adopted v2 policy. Successful creation
or reconstruction publishes the finalized original report and
.chrono-harness/state/origin.json in the new destination. Both branch kinds use
the integration check role; local context and collection manifest paths are
.chrono-harness/state/local/context.json and
.chrono-harness/state/collection/manifest.json. The source and fetched target
dev must carry the same committed policy before starting a lane. Existing lanes
without an origin receipt need reconstruction from an adopted source; no origin
evidence is synthesized. Full short-command checks require separate full-host
activation.
.chrono-harness/bin/chrono-worktree start --host-root . --config .chrono-harness/worktree.json --kind feature --name change --path ../my-changeThe destination must not exist. reconstruct uses the same arguments plus
--plan .chrono-harness/state/reconstruction.json, with explicit fixed base and
candidate OIDs and a complete carry/retire path list. See the
worktree contract.
Creation/reconstruction records actual Git results and preserves old work;
reconstruction stages changes and requires a new commit and the canonical check.
PR creation and merge remain caller-owned. Shared full-governance
registries are proposed; worktree use does not activate them or establish full
input closure, freshness certification or deterministic local/CI parity.
The same installed chrono-worktree consumes explicit maintenance plans under
.chrono-harness/state/ with the registered .chrono-harness/worktree.json:
.chrono-harness/bin/chrono-worktree recover --host-root . --config .chrono-harness/worktree.json --plan .chrono-harness/state/recover.json
.chrono-harness/bin/chrono-worktree cleanup --host-root . --config .chrono-harness/worktree.json --plan .chrono-harness/state/cleanup.json
.chrono-harness/bin/chrono-worktree cleanup-fetch --host-root . --config .chrono-harness/worktree.json --plan .chrono-harness/state/cleanup-fetch.jsonRecovery validates the original failure, reconciled HEAD/index and owned lock; staged changes still need a candidate commit and the canonical check. Cleanup requires an explicit retained branch/commit and selected disposable artifacts. Fetch-ref cleanup requires its original failed receipt and a fixed local branch preserving the expected commit. Reports retain failures and distinguish verified removal from unverified partial effects. See the pinned maintenance contract for complete plan formats. Lost recovery identity, damaged Git metadata and PR/merge orchestration remain separate obligations.
The installed tool writes separate immutable intents before checkout/lock effects and before fetching. After establishing that the original process stopped, supply an explicit intent digest and original-result presence/digest to:
.chrono-harness/bin/chrono-worktree recover-interrupted --host-root . --config .chrono-harness/worktree.json --plan .chrono-harness/state/interrupted-checkout.json
.chrono-harness/bin/chrono-worktree cleanup-fetch-interrupted --host-root . --config .chrono-harness/worktree.json --plan .chrono-harness/state/interrupted-fetch.jsonCheckout recovery also requires the reconciled HEAD/index tree. Fetch cleanup requires the expected current OID and a local branch retaining it; an already absent ref needs an explicit retry plan. Both preserve the original bytes and keep the original outcome unknown; terminal reports use ordinary maintenance. See the pinned interruption contracts. These commands do not reconstruct a lost index, make concurrent writers atomic, or certify full governance or deterministic parity.
The pinned release also provides optional explicitly bound Git readers. This host keeps its registered scoped profile; adopting new binaries does not implicitly change the Git-binding or governance policy. See the pinned Git facts contract.
The pinned binaries also provide explicit release-workflow generation and full-context CI transport. This host retains its registered scoped profile; installing binaries does not activate full governance. See the pinned release CI and full context CI contracts.
chrono-worktree cleanup-remote consumes an explicit state plan with the expected remote URL, work-branch OID and retained target commit. It performs exact leased deletion, verifies absence and preserves original failures; this is not an atomic remote transaction or a PR/merge verdict. See remote retirement.
The pinned binaries compare registered Git-tree/index identities with physical bytes, file types and owner executable bits. Git configuration cannot silently normalize changed inputs for this comparison; checkouts must materialize the registered exact bytes.
chrono-worktree inspect-rebind and rebind use an explicit branch, HEAD, chosen index tree, backup and donor to repair missing or damaged linked-checkout metadata while preserving observed work. The released chrono-worktree resume-rebind command continues an interrupted or failed rebind only from its retained intent and original plan; it does not infer a new index, backup, donor or ownership choice. These commands do not infer the lost historical index or original outcome. See metadata rebind and literal identity.
The pinned chrono-ci migrate command supports explicit owned workflow transitions and preserves host configuration. For ordinary binary upgrades the existing provider, SDK profiles and generated workflows stay unchanged; use verify and the same unit checks. See host customization and updates.