diff --git a/README.md b/README.md
index c90e5b8..f25bd7e 100644
--- a/README.md
+++ b/README.md
@@ -62,7 +62,7 @@ uv run controltrace export     # 重新测试并导出 CSV、可核验底稿 ZIP
 uv run controltrace verify     # 校验 ZIP 文件哈希，并从包内源数据重放自动测试
 ```
 
-界面可以按系统、期间、控制类型和风险等级筛选线索；点开一条线索可查看事件时间线、关联记录、规则依据和证据 ID。复核者填写姓名、处理状态、结论与备注后，导出的 CSV 和 ZIP 会包含当前复核记录。ZIP 同时包含合成源表的原值 JSON、适合表格查看的 CSV、机器可读的复核历史 `reviews.json`、规则元数据、全部文件的 SHA-256 清单和逐条 Markdown 底稿。`verify` 会核对文件哈希、CSV 与 JSON 的一致性，并用包内原值重放规则，逐字节比较所选发现清单与底稿。格式 2/3 继续按其历史规则核验，更早格式请使用原版本。
+界面可以按系统、期间、控制类型和风险等级筛选线索；点开一条线索可查看事件时间线、关联记录、规则依据和证据 ID。复核者填写姓名、处理状态、结论与备注后，导出的 CSV 和 ZIP 会包含当前复核记录。ZIP 同时包含合成源表的原值 JSON、适合表格查看的 CSV、机器可读的复核历史 `reviews.json`、规则元数据、全部文件的 SHA-256 清单和逐条 Markdown 底稿。`verify` 会核对文件哈希、CSV 与 JSON 的一致性，并用包内原值重放规则，逐字节比较所选发现清单与底稿。格式 2–4 继续按其历史规则核验，更早格式请使用原版本。
 
 底稿 JSON 会先经过结构校验：清单必须是对象、源表必须是对象行数组，重复键与 `NaN`/`Infinity` 等非有限数值会被拒绝。非法输入由 `verify` 返回错误说明和退出码 1，避免把解析歧义带入后续重放。
 
@@ -70,7 +70,7 @@ uv run controltrace verify     # 校验 ZIP 文件哈希，并从包内源数据
 
 **复核不能只跟着发现 ID 走。** 0.2.8 将每次新复核绑定到自动发现的完整证据指纹。相同证据重跑保留当前结论；同一 ID 的关联来源记录或判断口径变化后，原结论保留在历史中，当前状态返回待复核。界面保存时核对首次展示的编辑版本；其他人追加复核或证据发生变化后，须显式重新加载，防止旧表单覆盖新证据。版本检查、复核及指纹关联在一个事务内提交，并对同一发现的并发写入进行冲突检查。[复核完整性设计](docs/REVIEW_INTEGRITY.md)解释业务场景、Python API、迁移和可复现实验。
 
-新 ZIP 使用格式 4，延续格式 3 的 `review_evidence.json`，`verify` 根据重放的当前证据选择可适用的复核。0.2.9 加强 CSV 字符串和表头的公式防护：全角触发符、前置空白与控制字符使用可见 `text: ` 前缀，JSON 原值不变。格式 3 / 0.2.8 保留旧 CSV 规则，格式 2 / 0.2.7 也仍可核验，后者明确说明没有证据版本关联。旧数据库的未知历史仍须重新评估，不补造证据关联；新包没有归档历史源快照，也不提供身份认证或签名。[CSV 策略与版本兼容表](docs/CSV_EXPORTS.md)。
+新 ZIP 使用格式 5，延续 `review_evidence.json` 和可见 `text: ` CSV 策略，按解析后的 UTC 时刻展示时间线，保留原始时间字符串。0.2.10 修复小数秒、混合时区被字符串顺序颠倒的问题。旧格式 2–4 继续按其原顺序重放，校验器明确提示不保证 UTC 时间顺序；格式 2 还明确说明没有证据版本关联。排序修正会改变发现的完整指纹，旧复核保留为历史，当前结论须按新指纹重新评估。新包不补造未知历史关联，不归档历史源快照，也不提供身份认证或签名。[时间线与复核迁移](docs/TIMELINE_ORDER.md) · [CSV 兼容表](docs/CSV_EXPORTS.md)。
 
 ## 数据字典与案例
 
@@ -91,7 +91,7 @@ python -m ruff check src tests
 python -m pytest -q
 ```
 
-**本机验证（2026-10-24，Linux / Python 3.12.14）：** 完整 pytest 为 **108 passed**，Ruff 与 git diff --check 通过。新增 18 个 CSV/兼容回归，覆盖全角及控制字符、表头、前缀碰撞、数值、多行记录边界、新格式原值保留，以及由未修改的 0.2.8 实际导出的旧包重放。该旧包含 12 条预置线索中的一条导出、28 个源文件和一条程序生成的合成复核；不是实际人员的审计意见。既有用例覆盖证据变更后重新待复核、并发复核、事务回滚、迁移、非法 JSON 和重哈希后的伪造结论。[GitHub Actions CI](https://github.com/dev-belly/ControlTrace/actions/workflows/ci.yml) 在 Ubuntu / Python 3.11、3.12 和 Windows / Python 3.12 运行 Ruff 与完整 pytest。当前包版本为 0.2.9，新导出采用格式 4；格式 2/3 保持其历史重放策略。
+**本机验证（2026-10-24，Linux / Python 3.12.14）：** 完整 pytest 为 **113 passed**，Ruff、git diff --check 与锁文件检查通过。18 个 CSV/兼容回归覆盖全角和控制字符、表头、前缀碰撞、数值、多行记录边界及原值保留；另有 5 个时间线/历史绑定回归，覆盖小数秒、混合时区、无效展示时间、旧顺序提示和旧结论不能继承。保留未修改的 0.2.8、0.2.9 代码实际生成的合成底稿，不以当前序列化器伪造旧包。既有用例覆盖并发复核、事务回滚、迁移、非法 JSON 和重哈希后的伪造结论。[GitHub Actions CI](https://github.com/dev-belly/ControlTrace/actions/workflows/ci.yml) 在 Ubuntu / Python 3.11、3.12 和 Windows / Python 3.12 运行 Ruff 与完整 pytest。当前包版本为 0.2.10，新导出采用格式 5；格式 2–4 保持其原有 CSV 与时间线重放策略。
 
 ## 已知局限
 
diff --git a/docs/CSV_EXPORTS.md b/docs/CSV_EXPORTS.md
index 2418791..b5915e6 100644
--- a/docs/CSV_EXPORTS.md
+++ b/docs/CSV_EXPORTS.md
@@ -21,9 +21,11 @@ CSV 是便于表格查看的副本。`source_tables/*.json` 与 `reviews.json` 
 | 2 / 0.2.7 | 原单引号策略 | 无，明确返回 `review_evidence_bound=false` |
 | 3 / 0.2.8 | 原单引号策略 | 有 |
 | 4 / 0.2.9 | 可见 `text: ` 策略，包含表头 | 有 |
+| 5 / 0.2.10 | 同格式 4，时间线改用实际 UTC 顺序 | 有 |
 
 格式号与应用版本必须匹配；校验器不能把旧 CSV 当成新策略处理。格式 2/3
-只在校验旧底稿时使用历史序列化规则，新导出始终使用格式 4。
+只在校验旧底稿时使用历史序列化规则，新导出始终使用格式 5。
+格式 2–4 仍按原字符串时间线重放，并明确报告旧排序；见[时间线与复核迁移](TIMELINE_ORDER.md)。
 保留旧包原文件，再用当前源快照重新导出，可得到采用新 CSV 策略的副本；
 不要手改旧包的清单、版本号或哈希。新导出不能补造未知的历史证据版本。
 
diff --git a/docs/REVIEW_INTEGRITY.md b/docs/REVIEW_INTEGRITY.md
index ad6b2d2..ac50c7b 100644
--- a/docs/REVIEW_INTEGRITY.md
+++ b/docs/REVIEW_INTEGRITY.md
@@ -42,7 +42,7 @@ save_review(
 
 从格式 3 起，底稿保留原 `reviews.json` 字段，同时增加 `review_evidence.json`：按发现和复核 ID 保存关联指纹，未知历史为 `null`。CSV 标明当前证据指纹和不适用历史数量，Markdown 明确显示当前结论与全部历史。校验器从包内源记录重放自动发现，重算指纹并独立选择当前结论；即使 CSV、底稿和文件哈希被一致修改，旧证据结论冒充当前结论仍会被拒绝。0.2.9 新导出采用格式 4，加强 CSV 字符串和表头的展示策略，JSON 原值与证据关联规则不变；[CSV 版本表](CSV_EXPORTS.md)列出细节。
 
-格式 2、版本 0.2.7 的旧底稿继续支持原始重放；返回 `review_evidence_bound=false`，命令行也说明旧格式没有证据版本关联。格式 3 / 0.2.8 保留原 CSV 策略及证据关联重放。更早版本仍需其原版本校验器。这种兼容不会把旧底稿升级成拥有新保证的证据。
+格式 2、版本 0.2.7 的旧底稿继续支持原始重放；返回 `review_evidence_bound=false`，命令行也说明旧格式没有证据版本关联。格式 3 / 0.2.8 保留原 CSV 策略及证据关联重放，格式 4 / 0.2.9 保留其可见文本 CSV。0.2.10 新导出使用格式 5，按实际 UTC 时刻排列时间线；格式 2–4 继续按原字符串顺序重放，并返回 `legacy_timeline_order=true`。更早版本仍需其原版本校验器。这种兼容不会把旧底稿升级成拥有新保证的证据。[时间线迁移](TIMELINE_ORDER.md)说明为什么排序修正也可能要求重新复核。
 
 指纹和包内清单不能证明复核者身份、源系统真实性或防御同时重写所有内容的攻击者。新包只带当前源快照，保留的历史指纹并不等于保存了旧源快照。认证、签名和历史源版本归档仍是实际业务接入前需要单独解决的工作。
 
diff --git a/docs/TIMELINE_ORDER.md b/docs/TIMELINE_ORDER.md
new file mode 100644
index 0000000..260847f
--- /dev/null
+++ b/docs/TIMELINE_ORDER.md
@@ -0,0 +1,31 @@
+# 审计时间线按实际 UTC 时刻排序
+
+ISO 时间字符串的字典顺序不一定是事件顺序：`11:00:00+01:00` 是 UTC
+`10:00:00`，而 `06:00:02-04:00` 是 UTC `10:00:02`。旧展示会把后者排在
+前面。整秒的 `...00Z` 与小数秒的 `...00.050000Z` 也不能直接按字符排序。
+
+0.2.10 根据解析后的 UTC 时刻排序，以证据 ID 和字段名打破同时刻的并列。
+时间线上仍显示原始来源字符串；不能解析或为空的时间不加入时间线，但原记录
+继续保留在关联证据中。控制谓词未改变，未知或非法业务输入仍须单独核实。
+
+## 为什么需要格式 5
+
+发现指纹包含完整时间线。若修正顺序，旧指纹就不再表示当前自动发现，之前的
+复核保留为历史，当前状态返回 `pending`；不能因为来源 ID 没变就继承旧结论。
+这里可能只改变了证据呈现的顺序，并不意味着来源时间被改写。
+
+| 底稿格式 / 版本 | 时间线重放 | CSV 策略 | 证据关联 |
+| --- | --- | --- | --- |
+| 2 / 0.2.7 | 原字符串顺序 | 原单引号方式 | 无 |
+| 3 / 0.2.8 | 原字符串顺序 | 原单引号方式 | 有 |
+| 4 / 0.2.9 | 原字符串顺序 | 可见 `text: ` | 有 |
+| 5 / 0.2.10 | 实际 UTC 时刻 | 可见 `text: ` | 有 |
+
+格式 2–4 校验时重建原顺序，以保留原底稿与复核指纹的一致性，返回
+`legacy_timeline_order=true`，命令行明确提醒它不保证 UTC 时间顺序。
+新导出始终使用格式 5，返回 `legacy_timeline_order=false`。不要直接修改旧包
+的格式号或清单来冒充升级；保留原包，再用当前源快照重跑并按新指纹重新复核。
+
+[0.2.9 实际导出包](../tests/fixtures/README.md)保留混合时区的旧顺序和一条
+程序生成的合成复核。回归同时验证旧包可核验，以及当前排序不会继承旧指纹
+的关闭结论。这些是合成工程测试，不是实际审计意见或来源真实性证明。
diff --git a/pyproject.toml b/pyproject.toml
index c071b9c..d93cc1c 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "controltrace"
-version = "0.2.9"
+version = "0.2.10"
 description = "Synthetic IT audit evidence and internal control testing workbench"
 readme = "README.md"
 requires-python = ">=3.11"
diff --git a/src/controltrace/__init__.py b/src/controltrace/__init__.py
index f209780..f3759d1 100644
--- a/src/controltrace/__init__.py
+++ b/src/controltrace/__init__.py
@@ -1,3 +1,3 @@
 """ControlTrace synthetic IT audit demonstration."""
 
-__version__ = "0.2.9"
+__version__ = "0.2.10"
diff --git a/src/controltrace/cli.py b/src/controltrace/cli.py
index d429eaa..928a93e 100644
--- a/src/controltrace/cli.py
+++ b/src/controltrace/cli.py
@@ -45,6 +45,8 @@ def main(argv: list[str] | None = None) -> int:
         )
         if not result["review_evidence_bound"]:
             print("Legacy bundle: review history is valid but was not bound to an evidence snapshot.")
+        if result["legacy_timeline_order"]:
+            print("Legacy timeline: original string order is replayed; UTC chronology is not guaranteed.")
         return 0
     from controltrace.rules import run_tests
     from controltrace.store import initialize_db
diff --git a/src/controltrace/exports.py b/src/controltrace/exports.py
index 1077593..de89726 100644
--- a/src/controltrace/exports.py
+++ b/src/controltrace/exports.py
@@ -217,7 +217,7 @@ def workpaper_zip(db_path: str | Path, finding_ids: set[str] | None = None) -> b
     })
     manifest = {
         "notice": "Synthetic case study; no real enterprise audit data",
-        "bundle_format_version": 4,
+        "bundle_format_version": 5,
         "audit_cutoff": DEMO_CUTOFF,
         "generator_seed": int(meta.get("seed", DEFAULT_SEED)),
         "controltrace_version": __version__,
diff --git a/src/controltrace/rules.py b/src/controltrace/rules.py
index 4f4135a..fecc550 100644
--- a/src/controltrace/rules.py
+++ b/src/controltrace/rules.py
@@ -109,6 +109,32 @@ def _index(rows: list[dict], field: str) -> dict[str, dict]:
     return {row[field]: row for row in rows}
 
 
+def evidence_timeline(related: list[dict], *, legacy: bool = False) -> list[dict]:
+    """Preserve source timestamps while ordering the current view by UTC."""
+    timeline = []
+    for item in related:
+        record = item["record"]
+        for field, label, kind in _TIMELINE_FIELDS.get(item["table"], []):
+            value = record.get(field)
+            if legacy:
+                include = bool(value)
+            else:
+                try:
+                    include = parse_utc(value) is not None
+                except (ValueError, TypeError, AttributeError, OverflowError):
+                    include = False
+            if include:
+                timeline.append({
+                    "at": value, "event": label, "kind": kind,
+                    "field": field, "evidence_id": item["id"],
+                    "source_system": record["source_system"],
+                })
+    timeline.sort(key=lambda item: (
+        item["at"] if legacy else parse_utc(item["at"]), item["evidence_id"], item["field"],
+    ))
+    return timeline
+
+
 def _finding(
     control_id: str,
     issue_code: str,
@@ -132,17 +158,7 @@ def _finding(
         {"table": table, "id": str(record[PRIMARY_KEYS[table]]), "record": record.copy()}
         for table, record in unique.values()
     ]
-    timeline = []
-    for item in related:
-        record = item["record"]
-        for field, label, kind in _TIMELINE_FIELDS.get(item["table"], []):
-            if record.get(field):
-                timeline.append({
-                    "at": record[field], "event": label, "kind": kind,
-                    "field": field, "evidence_id": item["id"],
-                    "source_system": record["source_system"],
-                })
-    timeline.sort(key=lambda item: (item["at"], item["evidence_id"], item["field"]))
+    timeline = evidence_timeline(related)
     digest = hashlib.sha256(f"{control_id}|{issue_code}|{entity_id}".encode()).hexdigest()[:12]
     return {
         "finding_id": f"F-{digest.upper()}",
diff --git a/src/controltrace/verify.py b/src/controltrace/verify.py
index bef0b02..8576e44 100644
--- a/src/controltrace/verify.py
+++ b/src/controltrace/verify.py
@@ -16,7 +16,7 @@
 from controltrace.data import DEMO_CUTOFF, SOURCE_TABLES
 from controltrace.exports import _csv_bytes, _json, exception_csv, finding_workpaper
 from controltrace.reviews import REVIEW_FIELDS, bind_review_history, validate_review_decision
-from controltrace.rules import RULES, evaluate
+from controltrace.rules import RULES, evaluate, evidence_timeline
 
 
 class BundleVerificationError(ValueError):
@@ -61,10 +61,11 @@ def verify_bundle(path: str | Path) -> dict[str, Any]:
         manifest = _read_json(archive, "manifest.json")
         _require(isinstance(manifest, dict), "Manifest must be a JSON object")
         bundle_format = manifest.get("bundle_format_version")
-        _require(bundle_format in (2, 3, 4),
+        _require(bundle_format in (2, 3, 4, 5),
                  "Unsupported bundle format; export again with this ControlTrace version")
         evidence_bound = bundle_format >= 3
         legacy_csv = bundle_format < 4
+        legacy_timeline = bundle_format < 5
         expected_files = manifest.get("files_sha256")
         _require(isinstance(expected_files, dict), "Manifest has no complete file hash list")
         _require(set(expected_files) == set(names) - {"manifest.json"},
@@ -88,7 +89,7 @@ def verify_bundle(path: str | Path) -> dict[str, Any]:
         _require(manifest.get("audit_cutoff") == DEMO_CUTOFF,
                  "Bundle cutoff does not match the current control rules")
         bundle_version = manifest.get("controltrace_version")
-        _require(bundle_version == {2: "0.2.7", 3: "0.2.8", 4: __version__}[bundle_format],
+        _require(bundle_version == {2: "0.2.7", 3: "0.2.8", 4: "0.2.9", 5: __version__}[bundle_format],
                  "Bundle was created by a different ControlTrace version")
         _require(_read_json(archive, "rules.json") == RULES,
                  "Rule catalog differs from the installed ControlTrace version")
@@ -105,6 +106,11 @@ def verify_bundle(path: str | Path) -> dict[str, Any]:
             source_rows[table] = rows
         data = {table: rows for table, rows in source_rows.items() if table != "expected_results"}
         replayed = evaluate(data)
+        if legacy_timeline:
+            # Old evidence fingerprints include the original timeline order.
+            # Replay those retained bytes without granting UTC-order guarantees.
+            for finding in replayed:
+                finding["timeline"] = evidence_timeline(finding["related_records"], legacy=True)
         all_ids = [finding["finding_id"] for finding in replayed]
         _require(len(all_ids) == len(set(all_ids)), "Replay produced duplicate finding IDs")
         _require(manifest.get("all_finding_ids") == all_ids,
@@ -191,4 +197,5 @@ def verify_bundle(path: str | Path) -> dict[str, Any]:
         "exported_findings": len(selected_ids),
         "scope": scope,
         "review_evidence_bound": evidence_bound,
+        "legacy_timeline_order": legacy_timeline,
     }
diff --git a/tests/fixtures/README.md b/tests/fixtures/README.md
index a3f7135..c1f39c2 100644
--- a/tests/fixtures/README.md
+++ b/tests/fixtures/README.md
@@ -9,3 +9,12 @@ created test review, not a real person's audit decision. The review note contain
 an ASCII formula prefix, embedded CR/LF and fullwidth text. Its CSV uses the
 historical apostrophe policy. The regression must replay these retained bytes
 under the format-3 policy, while new format-4 exports use visible `text: ` prefixes.
+
+`controltrace-0.2.9-v4-timeline.zip` was exported by the unmodified 0.2.9 code at
+`20c0bcfe33e23a95a3c0f95975d86474800fcd26` before changing timeline ordering.
+SHA-256: `85401d33aee4c20b81469d58b505d07658a1f3c8c0b57c27a184b6906ea069c9`.
+Three synthetic CH002/D002 timestamps use mixed UTC offsets, and one test review
+is programmatically recorded. Original lexical order places deployment before
+approval. Format-4 replay must preserve its historical bytes and review binding,
+with an explicit legacy-order notice. A current UTC-order replay must require
+a new review for the corrected finding representation.
diff --git a/tests/fixtures/controltrace-0.2.9-v4-timeline.zip b/tests/fixtures/controltrace-0.2.9-v4-timeline.zip
new file mode 100644
index 0000000..a8f532a
Binary files /dev/null and b/tests/fixtures/controltrace-0.2.9-v4-timeline.zip differ
diff --git a/tests/test_csv_policy.py b/tests/test_csv_policy.py
index 7ccb0ad..0f44576 100644
--- a/tests/test_csv_policy.py
+++ b/tests/test_csv_policy.py
@@ -43,7 +43,7 @@ def test_visible_prefix_does_not_merge_distinct_strings():
     ]
 
 
-def test_v4_review_export_keeps_original_json_and_replays(tmp_path):
+def test_current_review_export_keeps_original_json_and_replays(tmp_path):
     db = tmp_path / "case.duckdb"
     initialize_db(db)
     finding_id = run_tests(db)[0]["finding_id"]
@@ -58,8 +58,8 @@ def test_v4_review_export_keeps_original_json_and_replays(tmp_path):
     assert verify_bundle(path)["review_evidence_bound"] is True
     with zipfile.ZipFile(io.BytesIO(bundle)) as archive:
         manifest = json.loads(archive.read("manifest.json"))
-        assert manifest["bundle_format_version"] == 4
-        assert manifest["controltrace_version"] == "0.2.9"
+        assert manifest["bundle_format_version"] == 5
+        assert manifest["controltrace_version"] == "0.2.10"
         assert json.loads(archive.read("reviews.json"))[finding_id][0]["notes"] == notes
 
 
diff --git a/tests/test_legacy_timeline.py b/tests/test_legacy_timeline.py
new file mode 100644
index 0000000..d2443ae
--- /dev/null
+++ b/tests/test_legacy_timeline.py
@@ -0,0 +1,38 @@
+"""A timeline fix must not reinterpret historical review fingerprints."""
+
+import json
+import zipfile
+from pathlib import Path
+
+from controltrace.cli import main
+from controltrace.data import SOURCE_TABLES, parse_utc
+from controltrace.reviews import bind_review_history
+from controltrace.rules import evaluate
+from controltrace.verify import verify_bundle
+
+FIXTURE = Path(__file__).parent / "fixtures" / "controltrace-0.2.9-v4-timeline.zip"
+
+
+def test_actual_v029_archive_replays_with_an_explicit_legacy_timeline_notice(capsys):
+    result = verify_bundle(FIXTURE)
+    assert result["legacy_timeline_order"] is True
+    assert result["review_evidence_bound"] is True
+    assert result["exported_findings"] == 1
+    assert main(["verify", "--bundle", str(FIXTURE)]) == 0
+    assert "UTC chronology is not guaranteed" in capsys.readouterr().out
+
+
+def test_corrected_timeline_does_not_inherit_an_old_order_review():
+    with zipfile.ZipFile(FIXTURE) as archive:
+        data = {table: json.loads(archive.read(f"source_tables/{table}.json")) for table in SOURCE_TABLES}
+        manifest = json.loads(archive.read("manifest.json"))
+        finding_id = manifest["finding_ids"][0]
+        history = json.loads(archive.read("reviews.json"))[finding_id]
+        bindings = json.loads(archive.read("review_evidence.json"))[finding_id]
+    finding = next(item for item in evaluate(data) if item["finding_id"] == finding_id)
+    instants = [parse_utc(item["at"]) for item in finding["timeline"]]
+    assert instants == sorted(instants)
+    rebound = bind_review_history(finding, history, bindings)
+    assert rebound["review_status"] == "pending"
+    assert rebound["latest_review"] is None
+    assert rebound["review_stale_count"] == 1
diff --git a/tests/test_rules.py b/tests/test_rules.py
index c9a5fb9..ba825a1 100644
--- a/tests/test_rules.py
+++ b/tests/test_rules.py
@@ -7,7 +7,7 @@
 import pytest
 
 from controltrace.data import generate_demo_data, parse_utc
-from controltrace.rules import evaluate
+from controltrace.rules import evaluate, evidence_timeline
 
 
 def _record(data, table, key, value):
@@ -43,6 +43,36 @@ def test_fixture_manifest_matches_actual_rule_outcomes():
         assert finding["rule"]["control_goal"]
 
 
+@pytest.mark.parametrize(
+    ("approved_at", "tested_at", "deployed_at"),
+    [
+        ("2025-06-10T10:00:00Z", "2025-06-10T10:00:00.050000Z", "2025-06-10T10:00:00.100000Z"),
+        ("2025-06-10T11:00:00+01:00", "2025-06-10T10:00:01Z", "2025-06-10T06:00:02-04:00"),
+    ],
+)
+def test_evidence_timeline_orders_actual_instants_and_preserves_source_times(
+    approved_at, tested_at, deployed_at
+):
+    data = deepcopy(generate_demo_data())
+    deployment = _record(data, "deployments", "deployment_id", "D002")
+    deployment["deployed_at"] = deployed_at
+    ticket = _record(data, "change_tickets", "ticket_id", deployment["ticket_id"])
+    ticket.update(approved_at=approved_at, tested_at=tested_at)
+    finding = next(item for item in evaluate(data) if item["entity_id"] == "D002")
+    timeline = finding["timeline"]
+    instants = [parse_utc(item["at"]) for item in timeline]
+    assert instants == sorted(instants)
+    assert [item["at"] for item in timeline[:3]] == [approved_at, tested_at, deployed_at]
+
+
+def test_timeline_omits_unparseable_view_entries_without_rewriting_source():
+    record = {"submitted_at": "", "approved_at": "not-a-time", "tested_at": "2025-06-10T10:00:00Z",
+              "source_system": "synthetic"}
+    related = [{"table": "change_tickets", "id": "test", "record": record}]
+    assert [item["field"] for item in evidence_timeline(related)] == ["tested_at"]
+    assert record["approved_at"] == "not-a-time"
+
+
 def test_exact_deadline_is_on_time_for_each_time_based_control():
     data = deepcopy(generate_demo_data())
     _record(data, "accounts", "account_id", "A001")["disabled_at"] = "2025-06-12T09:00:00Z"
diff --git a/uv.lock b/uv.lock
index 7bb6070..139822d 100644
--- a/uv.lock
+++ b/uv.lock
@@ -220,7 +220,7 @@ wheels = [
 
 [[package]]
 name = "controltrace"
-version = "0.2.9"
+version = "0.2.10"
 source = { editable = "." }
 dependencies = [
     { name = "duckdb" },