fix: persist WAL tail truncation per design protocol (C5+H8)

Recovery tail-truncation had two compounding bugs in wal/recover.go:

C5: truncateSegment only called os.Truncate. Missing per design §3.2
    line 787-794:
      - Step 2: fsync the truncated segment
      - Step 3: delete empty trailing segments
      - Step 4: fsync WAL directory
    And all errors were swallowed into result.TruncateError with recovery
    still returning success, violating design line 799: "若 ftruncate、
    segment fsync、空 segment 删除或 WAL directory fsync 任一步失败，
    recovery 必须报错，DB 不得进入可写状态".

H8: findValidOffset only checked physical record CRCs, ignoring the
    FragmentCollector state machine. For a tail of First + Middle*
    without Last, it returned the offset AFTER the last Middle fragment
    instead of the last COMPLETE batch end. Result: residual half-batch
    fragments caused repeated tail-corruption reports on every restart.

Changes:
- wal/recover.go:
  - Add findLastCompleteBatchEnd: batch-aware offset finder using
    FragmentCollector state machine. Handles block-boundary padding
    correctly (continue across full-block padding, return on short-block).
  - Add truncateAndPersist: 4-step protocol (ftruncate + fsync segment +
    delete empty trailing + fsync dir). Any step failure is fatal.
  - Add segmentFsyncFn (package-level var for test injection, same
    pattern as C6's dirFsyncFn).
  - Refactor Recover failure path: use new functions, hard-error on
    truncation persist failure (was: swallow to TruncateError).
  - TruncateError field semantics: informational only ("tail corruption
    was detected and repair attempted"). Persist failures return error.
  - Delete findValidOffset and truncateSegment (replaced).
- wal/recover_offset_test.go (new): 8 unit tests for
  findLastCompleteBatchEnd covering clean/partial-tail/no-batch/
  physical-corruption/partial-only/block-boundary-padding/non-zero-tail/
  zero-tail cases. 5 unit tests for truncateAndPersist covering success/
  ftruncate-fail/dir-fsync-fail/segment-fsync-fail/retry-after-failure.
- wal/recover_test.go: add TestRecoverPartialFragmentTailIdempotent
  (H8 e2e regression: truncation point must be at last complete batch),
  TestRecoverTruncationFailureFailsRecovery (C5 e2e regression: any
  step failure fails Recover), TestRecoverInvalidBatchNotTruncatable
  (design line 778-781: invalid batch content hard-fails, NOT truncatable).

Injection note: segmentFsyncFn and dirFsyncFn (from C6) are package-level
vars; tests that override either must not use t.Parallel().

Verified: each new test fails on pre-fix code by logical analysis and
passes after the fix. Full suite green including go test -race ./... .

Phase 1 simplification: emptyTrailingSegments is always nil in Phase 1
(truncated segment is always segments[last]). The parameter is kept in
truncateAndPersist's signature for forward compatibility with the C4 fix.

Audit context: docs/audit-3.2.md C5 and H8 (H8 Oracle-verified bg_ef425776).

.omo/plans/fix-c5-h8-truncation-persist.md

@@ -0,0 +1,592 @@
+# C5 + H8 修复方案：truncation 持久化 + batch-aware 截断点
+
+## TL;DR
+
+> **目标**：让 WAL 尾部截断真正按设计 §3.2 line 787-800 执行 4 步协议（ftruncate + fsync segment + 删空 segment + fsync dir），并让截断点基于完整 batch 边界而不是物理 record 边界。两条 bug 在同一段代码（truncation 路径），捆绑修最经济。
+>
+> **交付**：
+> - 新函数 `findLastCompleteBatchEnd` —— batch-aware 的截断点计算（替换 `findValidOffset`）
+> - 新函数 `truncateAndPersist` —— 4 步串行执行（替换 `truncateSegment`）
+> - `Recover` 调用新函数，**失败硬错误**（不再吞）
+> - 复用 C6 的 `dirFsyncFn` 做 step 4
+> - 新增 ~6 个测试覆盖各场景
+> - 单次 commit
+>
+> **预估工时**：3.5-5 小时（Oracle 修订后加了 block-boundary、padding、retry、invalid-batch 等测试）
+> **风险**：中。改动局限在 `wal/recover.go` 一个文件 + 一处 record_parser 钩子，但 truncation 涉及多步 fsync，要小心顺序
+
+---
+
+## Context
+
+### Bug 摘要
+
+#### C5（truncation 4 步缺失）
+
+`wal/recover.go:62-68`：
+
+```go
+validOffset, truncErr := findValidOffset(lastSeg.FilePath)
+if truncErr != nil {
+    result.TruncateError = fmt.Errorf("%w (find valid offset: %v)", err, truncErr)  // ← 吞
+} else if truncErr := truncateSegment(lastSeg.FilePath, validOffset); truncErr != nil {
+    result.TruncateError = fmt.Errorf("%w (truncate: %v)", err, truncErr)          // ← 吞
+}
+```
+
+`truncateSegment` 只调 `os.Truncate`，**缺**：
+- Step 2: fsync 被截断的 segment
+- Step 3: 删除空的后续 segment
+- Step 4: fsync WAL directory
+
+错误被吞到 `result.TruncateError`，recovery 仍返回成功。DB 进入可写状态。
+
+#### H8（截断点不跟踪 batch 边界）
+
+`findValidOffset` 只用 `DecodePhysicalRecord` 校验物理记录，**不跟踪 fragment 状态机**。尾部 `First + Middle*` 没 `Last` 时返回错误截断点（残留半截 fragment），导致下次启动再次报 tail corruption，**反复 repair**。
+
+### 设计依据
+
+`docs/design.md` §3.2 line 786-800：
+
+> 截断目标始终是**最后一个完整 WAL Batch 的结束位置** `lastCompleteBatchEnd`
+
+```
+1. ftruncate 当前 active segment 到 lastCompleteBatchEnd
+2. fsync 被截断的 segment
+3. 删除 startSequence == expectedSequence 且不含任何 complete batch 的后续空 segment
+4. fsync WAL directory
+```
+
+> 若 ftruncate、segment fsync、空 segment 删除或 WAL directory fsync **任一步失败，recovery 必须报错，DB 不得进入可写状态**
+
+### 协同：复用 C6 的 `dirFsyncFn`
+
+C6 已经抽出 `dirFsyncFn`（package-level 变量支持测试注入）。本次 fix 的 Step 4 直接复用，无需再抽一个。
+
+### 错误传播
+
+`Recover` 失败 → `DB.Open` 失败 → 用户看到 error，DB 没进可写状态 ✓
+
+---
+
+## 执行计划
+
+### Phase A：代码改动（90-120 分钟）
+
+#### A.1 新增 `findLastCompleteBatchEnd`（替换 `findValidOffset`）
+
+文件：`wal/recover.go`
+
+> **Oracle 修订（bg_f1e4db4f）BLOCKING**：原版看到 zero padding 立即 return，但 WAL 格式允许 full block 末尾 padding + 下一个 block 继续写 record。必须区分"full block 内 padding（continue 下一块）"和"short block 尾部 padding（return）"。
+
+```go
+// findLastCompleteBatchEnd walks the segment file, runs physical records
+// through the FragmentCollector state machine, and returns the byte offset
+// of the END of the last complete WAL Batch.
+//
+// This is the correct truncation target per design §3.2 line 786. The
+// previous findValidOffset only checked physical record CRCs, missing
+// the case where a First + Middle* fragment chain has no Last (H8 bug):
+// physical CRCs pass but no complete batch exists at that offset.
+//
+// Block-boundary handling: WAL format allows a full block to end with
+// zero padding when the next record doesn't fit (see BlockWriter.paddingNeeded).
+// This function must CONTINUE to the next block on padding in a full block,
+// and only RETURN on padding in a short (final) block or actual corruption.
+func findLastCompleteBatchEnd(filePath string) (int64, error) {
+    f, err := os.Open(filePath)
+    if err != nil {
+        return 0, fmt.Errorf("open %s: %w", filePath, err)
+    }
+    defer f.Close()
+
+    if _, err := f.Seek(WalFileHeaderSize, 0); err != nil {
+        return 0, fmt.Errorf("seek past header: %w", err)
+    }
+
+    collector := NewFragmentCollector()
+    lastCompleteEnd := int64(WalFileHeaderSize)
+    blockStartOffset := int64(WalFileHeaderSize)
+    buf := make([]byte, WalBlockSize)
+
+    for {
+        n, readErr := f.Read(buf)
+        if n > 0 {
+            blockData := buf[:n]
+            isFullBlock := n == WalBlockSize && readErr == nil
+            pos := 0
+            for pos < len(blockData) {
+                remaining := len(blockData) - pos
+
+                if remaining < PhysicalRecordHeaderSize {
+                    if isFullBlock {
+                        break // padding in full block, continue to next block
+                    }
+                    return lastCompleteEnd, nil // tail padding in short block
+                }
+
+                if isAllZeros(blockData[pos : pos+PhysicalRecordHeaderSize]) {
+                    if isFullBlock {
+                        break // zero-led padding in full block, continue
+                    }
+                    return lastCompleteEnd, nil // tail padding
+                }
+
+                rec, consumed, err := DecodePhysicalRecord(blockData[pos:])
+                if err != nil {
+                    return lastCompleteEnd, nil // physical corruption
+                }
+                if err := collector.Append(rec.Type, rec.Payload); err != nil {
+                    return lastCompleteEnd, nil // fragment state machine rejected
+                }
+
+                pos += consumed
+                recordEndAbsolute := blockStartOffset + int64(pos)
+
+                if collector.IsComplete() {
+                    lastCompleteEnd = recordEndAbsolute
+                    collector.Reset()
+                }
+            }
+            blockStartOffset += int64(n)
+        }
+        if readErr != nil || n < WalBlockSize {
+            break
+        }
+    }
+
+    return lastCompleteEnd, nil
+}
+```
+
+**关键不变量**：
+- `lastCompleteEnd` 只在 collector 回到 Idle（完整 batch 形成）时推进
+- 跨 block padding：`isFullBlock` 判断决定 continue vs return
+- 任何中断（short block padding / 物理损坏 / fragment 顺序错）→ 返回当前 `lastCompleteEnd`
+- EOF 时如果 collector 在 Collecting 状态 → 不推进（半截 batch 不算）
+
+#### A.2 新增 `truncateAndPersist`（替换 `truncateSegment`）
+
+文件：`wal/recover.go`
+
+```go
+// segmentFsyncFn is the package-level indirection for fsyncing a truncated
+// segment file. Tests that override this must not use t.Parallel().
+var segmentFsyncFn = segmentFsync
+
+func segmentFsync(filePath string) error {
+    f, err := os.OpenFile(filePath, os.O_WRONLY, 0o644)
+    if err != nil {
+        return fmt.Errorf("open for fsync: %w", err)
+    }
+    defer f.Close()
+    if err := f.Sync(); err != nil {
+        return fmt.Errorf("fsync: %w", err)
+    }
+    return nil
+}
+
+// truncateAndPersist executes the 4-step tail-truncation protocol per
+// design §3.2 line 787-794. Any step failure is fatal: per line 799,
+// DB must NOT enter writable state if truncation cannot be persisted.
+//
+// Steps:
+//   1. ftruncate segment to lastCompleteBatchEnd
+//   2. fsync the truncated segment
+//   3. delete trailing empty segments (startSequence == expectedSequence
+//      and no complete batch)
+//   4. fsync WAL directory (reuses dirFsyncFn from C6)
+func truncateAndPersist(
+    filePath string,
+    lastCompleteBatchEnd int64,
+    dir string,
+    emptyTrailingSegments []string,
+) error {
+    // Step 1: ftruncate
+    if err := os.Truncate(filePath, lastCompleteBatchEnd); err != nil {
+        return fmt.Errorf("ftruncate %s to %d: %w", filePath, lastCompleteBatchEnd, err)
+    }
+
+    // Step 2: fsync the truncated segment
+    if err := segmentFsyncFn(filePath); err != nil {
+        return fmt.Errorf("fsync truncated segment: %w", err)
+    }
+
+    // Step 3: delete empty trailing segments
+    for _, segPath := range emptyTrailingSegments {
+        if err := os.Remove(segPath); err != nil {
+            return fmt.Errorf("remove empty segment %s: %w", segPath, err)
+        }
+    }
+
+    // Step 4: fsync WAL directory (reuses C6's dirFsyncFn)
+    if err := dirFsyncFn(dir); err != nil {
+        return fmt.Errorf("fsync WAL dir after truncation: %w", err)
+    }
+    return nil
+}
+```
+
+> **关于 hook 触发的注释**：docstring 引用设计行号 + 列出 4 步，是必要的回归防护。`segmentFsyncFn` 的 not-parallel-safe 注释和 C6 的 `dirFsyncFn` 一致。
+
+#### A.3 重构 `Recover` 失败路径
+
+文件：`wal/recover.go`，替换 line 59-69 的 truncation 块：
+
+```go
+// 改前（C5 bug：吞错误）：
+if len(segments) > 0 {
+    lastSeg := segments[len(segments)-1]
+    validOffset, truncErr := findValidOffset(lastSeg.FilePath)
+    if truncErr != nil {
+        result.TruncateError = fmt.Errorf("%w (find valid offset: %v)", err, truncErr)
+    } else if truncErr := truncateSegment(lastSeg.FilePath, validOffset); truncErr != nil {
+        result.TruncateError = fmt.Errorf("%w (truncate: %v)", err, truncErr)
+    }
+}
+
+// 改后（按设计 line 787-800 执行 4 步，失败硬错误）：
+if len(segments) > 0 {
+    lastSeg := segments[len(segments)-1]
+    lastCompleteBatchEnd, findErr := findLastCompleteBatchEnd(lastSeg.FilePath)
+    if findErr != nil {
+        return nil, fmt.Errorf("wal: recover: find truncation offset: %w", findErr)
+    }
+
+    // Phase 1: truncated segment is always segments[last], no trailing empty
+    // segments to clean up. C4 fix will need to identify trailing empties
+    // based on the actually-corrupted segment's index.
+    var emptyTrailing []string
+
+    if err := truncateAndPersist(lastSeg.FilePath, lastCompleteBatchEnd, dir, emptyTrailing); err != nil {
+        // Per design §3.2 line 799: DB must NOT enter writable state.
+        return nil, fmt.Errorf("wal: recover: persist tail truncation: %w", err)
+    }
+}
+```
+
+> **TruncateError 字段语义澄清（Oracle 修订）**：`result.TruncateError` 在 `Recover` line 42-46 创建 result 时被设置为 `err`（the original tail corruption error）。这个字段保留为 **informational**："tail corruption was found and repaired"。现有测试 `TestRecoverWithTailCorruption` 和 `TestRecoverIdempotentAfterTruncation` 依赖这个非 nil 检查。
+>
+> **不要再**用 `TruncateError` 报告 truncation 持久化失败 —— 那种失败现在硬错误返回。两种语义不混。
+
+#### A.4 `emptyTrailingSegments` Phase 1 简化（Oracle 修订）
+
+> **Oracle 修订（bg_f1e4db4f）**：原计划的 `identifyEmptyTrailingSegments` 函数在 Phase 1 永远返回 nil，且其签名不足以支持 C4 修复（C4 需要 corrupted segment 的 index/path，不只是 `segments` 和 `expectedSequence`）。**删除函数**，改用 inline 注释 + `nil`。C4 修复时再补正确的实现。
+
+`Recover` 内部直接：
+
+```go
+// Phase 1: truncated segment is always segments[last], no trailing empty
+// segments to clean up. C4 fix will need to identify trailing empties
+// based on the actually-corrupted segment's index.
+var emptyTrailing []string // always nil in Phase 1
+
+if err := truncateAndPersist(lastSeg.FilePath, lastCompleteBatchEnd, dir, emptyTrailing); err != nil {
+    return nil, fmt.Errorf("wal: recover: persist tail truncation: %w", err)
+}
+```
+
+`truncateAndPersist` 的签名保留 `emptyTrailingSegments []string` 参数（forward compat），Phase 1 永远传 nil。`hasCompleteBatch` helper 也不需要了。
+
+#### A.5 删除老函数
+
+`wal/recover.go` 中删除：
+- `truncateSegment`（被 `truncateAndPersist` 替换）
+- `findValidOffset`（被 `findLastCompleteBatchEnd` 替换）
+
+### Phase B：测试（60-90 分钟）
+
+#### B.1 `findLastCompleteBatchEnd` 单元测试
+
+新增 `wal/recover_offset_test.go`：
+
+```go
+// Regression guards for H8: findLastCompleteBatchEnd must return the
+// offset of the last COMPLETE batch, not the last physical record.
+
+func TestFindLastCompleteBatchEnd_CleanSegment(t *testing.T) {
+    // 2 complete batches, no partial tail.
+    // Expect: offset = end of batch 2.
+}
+
+func TestFindLastCompleteBatchEnd_PartialTailFragment(t *testing.T) {
+    // 1 complete batch + First + Middle* (no Last).
+    // Expect: offset = end of batch 1 (NOT after the Middle fragment).
+    // This is the H8 regression case.
+}
+
+func TestFindLastCompleteBatchEnd_NoBatches(t *testing.T) {
+    // Empty segment (only header).
+    // Expect: offset = WalFileHeaderSize.
+}
+
+func TestFindLastCompleteBatchEnd_PhysicalCorruption(t *testing.T) {
+    // 1 complete batch + corrupted bytes (bad CRC).
+    // Expect: offset = end of batch 1.
+}
+
+func TestFindLastCompleteBatchEnd_PartialBatchOnly(t *testing.T) {
+    // Only First + Middle* at the start (no complete batch ever).
+    // Expect: offset = WalFileHeaderSize.
+}
+
+// Oracle BLOCKING test: must distinguish "padding in full block (continue
+// to next block)" from "padding in short block (return)".
+func TestFindLastCompleteBatchEnd_BlockBoundaryPadding(t *testing.T) {
+    // Construct: [Batch A in block 1 (filling)] [padding to end of block 1]
+    //             [Batch B in block 2]
+    // Expect: offset = end of Batch B in block 2 (NOT end of Batch A).
+    // This catches the bug where the original plan returned on first padding.
+}
+
+func TestFindLastCompleteBatchEnd_NonZeroTailPadding(t *testing.T) {
+    // 1 complete batch + 3 non-zero bytes (< PhysicalRecordHeaderSize).
+    // Expect: offset = end of batch 1.
+}
+
+func TestFindLastCompleteBatchEnd_ZeroTailPadding(t *testing.T) {
+    // 1 complete batch + 3 zero bytes.
+    // Expect: offset = end of batch 1.
+}
+```
+
+#### B.2 `truncateAndPersist` 测试
+
+新增到 `wal/recover_offset_test.go`：
+
+```go
+// Regression guards for C5: 4-step protocol must execute all steps.
+
+func TestTruncateAndPersist_Success(t *testing.T) {
+    // Create segment, write some bytes past lastCompleteBatchEnd.
+    // Call truncateAndPersist.
+    // Verify:
+    //   - File size == lastCompleteBatchEnd (Step 1)
+    //   - File fsync'd (hard to verify directly; trust the call)
+    //   - Empty trailing segments deleted (Step 3)
+    //   - Dir fsync'd (Step 4)
+}
+
+func TestTruncateAndPersist_FtruncateFailure(t *testing.T) {
+    // Pass non-existent file path.
+    // Expect: error mentioning "ftruncate".
+}
+
+func TestTruncateAndPersist_DirFsyncFailure(t *testing.T) {
+    // Inject dirFsyncFn failure (reuse C6 injection).
+    // Expect: error mentioning "fsync WAL dir".
+    // Note: file IS truncated (step 1 succeeded), but step 4 failed.
+    // Per design, Recover must return error → DB.Open fails.
+}
+
+func TestTruncateAndPersist_SegmentFsyncFailure(t *testing.T) {
+    // Inject segmentFsyncFn failure.
+    // Expect: error mentioning "fsync truncated segment".
+}
+
+// Oracle nice-to-have: verify retry after dir-fsync failure.
+func TestTruncateAndPersist_RetryAfterDirFsyncFailure(t *testing.T) {
+    // First call: inject dirFsyncFn failure → error.
+    // Second call: restore dirFsyncFn, call again → success.
+    // Verifies state stays recoverable across failures.
+}
+```
+
+#### B.3 `Recover` 集成测试
+
+修改或新增到 `wal/recover_test.go`：
+
+```go
+// Regression guard for C5+H8: end-to-end recovery with partial fragment
+// tail must persist truncation correctly and be idempotent.
+func TestRecoverPartialFragmentTailIdempotent(t *testing.T) {
+    dir := t.TempDir()
+    // Build segment with: [Batch A] [Batch B] [First + Middle* no Last]
+    // ...
+    
+    replayer1 := &mockReplayer{}
+    result1, err := Recover(dir, replayer1)
+    if err != nil { t.Fatalf("1st Recover: %v", err) }
+    if !result1.Truncated { t.Fatal("Truncated = false, want true") }
+    
+    // Verify file size == end of Batch B (H8 fix)
+    fi, _ := os.Stat(segmentPath)
+    if fi.Size() != expectedBatchBEnd {
+        t.Errorf("file size = %d, want %d (last complete batch end)",
+            fi.Size(), expectedBatchBEnd)
+    }
+    
+    // Second recovery should not see corruption (file is clean now)
+    replayer2 := &mockReplayer{}
+    result2, err := Recover(dir, replayer2)
+    if err != nil { t.Fatalf("2nd Recover: %v", err) }
+    if result2.Truncated { t.Error("2nd Recover should not see corruption") }
+    if result1.NextSequence != result2.NextSequence {
+        t.Errorf("NextSequence differs: %d vs %d", result1.NextSequence, result2.NextSequence)
+    }
+}
+
+// Regression guard for C5: any truncation step failure must fail DB.Open.
+func TestRecoverTruncationFailureFailsRecovery(t *testing.T) {
+    dir := t.TempDir()
+    // Build segment with tail corruption
+    // ...
+    
+    // Inject dirFsyncFn failure
+    orig := dirFsyncFn
+    dirFsyncFn = func(string) error { return errors.New("simulated") }
+    t.Cleanup(func() { dirFsyncFn = orig })
+    
+    _, err := Recover(dir, &mockReplayer{})
+    if err == nil { t.Fatal("expected Recover to fail when truncation persist fails") }
+    if !strings.Contains(err.Error(), "persist tail truncation") {
+        t.Errorf("error should mention 'persist tail truncation', got: %v", err)
+    }
+}
+
+// Oracle nice-to-have: CRC-valid but batch-content-invalid must hard-fail
+// through DecodeWalBatch, NOT enter truncation path. Per design line 778-781.
+func TestRecoverInvalidBatchNotTruncatable(t *testing.T) {
+    dir := t.TempDir()
+    // Construct segment with: physical records CRC-valid, but assembled
+    // batch has invalid header (e.g., entryCount=0).
+    // Expect: Recover returns error (NOT tail corruption, NOT success).
+    // Expect: file is NOT truncated.
+}
+```
+
+> **注意**：现有的 `TestRecoverIdempotentAfterTruncation` 测试（C2+C3 加的）已经覆盖了"截断后第二次 Recover 干净"的场景。本次新增的 `TestRecoverPartialFragmentTailIdempotent` 专门覆盖 **H8**（partial fragment tail），是更严格的回归测试。
+>
+> **Oracle 提醒**：检查现有 `TestRecoverIdempotentAfterTruncation` 的 `TruncateError` 期望是否需要更新（原期望非 nil，新行为下仍非 nil 因为 `result.TruncateError` 在 result 创建时就设了 original err，仍保留 informational 语义）。
+
+### Phase C：验证（15 分钟）
+
+```bash
+# 1. 编译
+go build ./...
+
+# 2. 重点测试
+go test ./wal -run 'TestFindLastCompleteBatchEnd|TestTruncateAndPersist|TestRecoverPartialFragmentTail|TestRecoverTruncationFailure|TestRecoverIdempotent' -count=1 -v
+
+# 3. wal 包全量
+go test ./wal/... -count=1
+
+# 4. 全仓
+go test ./... -count=1
+
+# 5. race 全仓
+go test -race ./... -count=1
+
+# 6. vet
+go vet ./...
+
+# 7. lint（如果装了）
+if command -v golangci-lint >/dev/null 2>&1; then
+    golangci-lint run ./wal/...
+fi
+```
+
+### Phase D：Commit message draft
+
+```
+fix: persist WAL tail truncation per design protocol (C5+H8)
+
+Recovery tail-truncation had two compounding bugs in wal/recover.go:
+
+C5: truncateSegment only called os.Truncate. Missing per design §3.2
+    line 787-794:
+      - Step 2: fsync the truncated segment
+      - Step 3: delete empty trailing segments
+      - Step 4: fsync WAL directory
+    And all errors were swallowed into result.TruncateError with recovery
+    still returning success, violating design line 799: "若 ftruncate、
+    segment fsync、空 segment 删除或 WAL directory fsync 任一步失败，
+    recovery 必须报错，DB 不得进入可写状态".
+
+H8: findValidOffset only checked physical record CRCs, ignoring the
+    FragmentCollector state machine. For a tail of First + Middle*
+    without Last, it returned the offset AFTER the last Middle fragment
+    instead of the last COMPLETE batch end. Result: residual half-batch
+    fragments caused repeated tail-corruption reports on every restart.
+
+Changes:
+- wal/recover.go:
+  - Add findLastCompleteBatchEnd: batch-aware offset finder using
+    FragmentCollector state machine. Returns true last batch boundary.
+  - Add truncateAndPersist: 4-step protocol (ftruncate + fsync segment +
+    delete empty trailing + fsync dir). Any step failure is fatal.
+  - Add segmentFsyncFn (package-level var for test injection, same
+    pattern as C6's dirFsyncFn).
+  - Refactor Recover failure path: use new functions, hard-error on
+    truncation persist failure (was: swallow to TruncateError).
+  - Delete findValidOffset and truncateSegment (replaced).
+- wal/recover_offset_test.go (new): 5 unit tests for
+  findLastCompleteBatchEnd covering clean/partial-tail/no-batch/
+  physical-corruption/partial-only cases. 4 unit tests for
+  truncateAndPersist covering success/ftruncate-fail/dir-fsync-fail/
+  segment-fsync-fail.
+- wal/recover_test.go: add TestRecoverPartialFragmentTailIdempotent
+  (H8 e2e regression) and TestRecoverTruncationFailureFailsRecovery
+  (C5 e2e regression).
+
+Injection note: segmentFsyncFn and dirFsyncFn (from C6) are package-level
+vars; tests that override either must not use t.Parallel().
+
+Verified: each new test fails on pre-fix code and passes after the fix.
+Full suite green including go test -race ./... .
+
+Phase 1 simplification: emptyTrailingSegments is always nil in Phase 1
+(truncated segment is always segments[last]). The parameter is kept in
+truncateAndPersist's signature for forward compatibility with the C4 fix.
+
+Audit context: docs/audit-3.2.md C5 and H8 (H8 Oracle-verified bg_ef425776).
+```
+
+---
+
+## 验收清单
+
+- [ ] Phase A.1：`findLastCompleteBatchEnd` 存在，使用 FragmentCollector
+- [ ] Phase A.2：`truncateAndPersist` 存在，4 步串行执行
+- [ ] Phase A.2：`segmentFsyncFn` package-level 变量存在
+- [ ] Phase A.3：`Recover` 失败路径调用新函数，失败硬错误
+- [ ] Phase A.4：`identifyEmptyTrailingSegments` + `hasCompleteBatch` helper 存在
+- [ ] Phase A.5：`truncateSegment` 和 `findValidOffset` 已删除
+- [ ] Phase B.1：5 个 `findLastCompleteBatchEnd` 单元测试存在
+- [ ] Phase B.2：4 个 `truncateAndPersist` 单元测试存在
+- [ ] Phase B.3：2 个 Recover 集成测试存在
+- [ ] `go test ./wal/... -count=1` 全绿
+- [ ] `go test ./... -count=1` 全绿
+- [ ] `go test -race ./... -count=1` 全绿
+- [ ] `go vet ./...` 无新增警告
+- [ ] 单次 commit，message 引用 audit C5+H8
+
+---
+
+## 不在本次范围内（后续 issue）
+
+| 编号 | 为什么不放进来 |
+|------|---------------|
+| C4 | 同一文件但不耦合。C4 修好后 `identifyEmptyTrailingSegments` 才会返回非 nil，但本 patch 已经写了前向兼容代码 |
+| C8 | segment_manager.go startSequence bug，独立 |
+| C1 | CRC 多项式错，独立 |
+| C7 | Put/Close 竞态，独立 |
+| H1-H7 | 其他 High，独立 |
+
+---
+
+## 修订记录
+
+- **v1（原始）**：C5+H8 修复方案初稿，送 Momus 审
+- **v1.0（Momus 审核 bg_c8823195）**：[OKAY]，无 blocking。思考过程提到 batch-content validity 担忧，Oracle Q2 确认非阻塞（upstream ReplaySegmentFile 已经处理）
+- **v1.1（Oracle 修订 bg_f1e4db4f）**：
+  - **BLOCKING**：`findLastCompleteBatchEnd` 跨 block padding bug — full block 末尾 padding 应 continue 到下一块，不是 return。加 `isFullBlock` 判断
+  - **BLOCKING**：缺跨 block 测试，加 `TestFindLastCompleteBatchEnd_BlockBoundaryPadding`
+  - **NEW**：加非零 padding / 零 padding 测试（Q1 f/g）
+  - **NEW**：加 `TestTruncateAndPersist_RetryAfterDirFsyncFailure`
+  - **NEW**：加 `TestRecoverInvalidBatchNotTruncatable`（CRC-valid 但 batch 内容无效，按 design line 778-781 应硬错误不截断）
+  - **CHANGE**：删除 `identifyEmptyTrailingSegments` + `hasCompleteBatch`（dead code，签名不足以支持 C4）。改 inline `var emptyTrailing []string` + 注释
+  - **CLARIFY**：`TruncateError` 字段保留 informational 语义（"tail corruption was found and repaired"），不再用于报告 truncation 持久化失败
+  - **FIX**：commit message 去掉 "C5 Oracle-verified bg_ef425776"（C5 实际未经过 Oracle 修订，只有 C2/C3/C4/H8 有）
+  - **BUMP**：估时 2.5-3.5h → 3.5-5h

wal/recover.go

@@ -13,12 +13,15 @@ type RecoveryResult struct {
 	NextSegmentID   uint64
 	ReplayedEntries int
 	Truncated       bool
-	TruncateError   error // non-nil if tail corruption was found
+	// TruncateError is informational only: non-nil means "tail corruption
+	// was found and repair was attempted". It does NOT report persistence
+	// failures — those cause Recover to return an error instead.
+	TruncateError error
 }
 
 // Recover performs a full WAL recovery: reads the recovery checkpoint from
-// MANIFEST (or CURRENT), scans segments, replays entries, and handles tail
+// MANIFEST, scans segments, replays entries, and persists tail truncation
-// truncation. On success the MANIFEST is updated with the new recovery state.
+// per design §3.2 line 787-800.
 func Recover(dir string, replayer BatchReplayer) (*RecoveryResult, error) {
 	if replayer == nil {
 		return nil, fmt.Errorf("wal: recover: replayer is nil")
@@ -37,15 +40,14 @@ func Recover(dir string, replayer BatchReplayer) (*RecoveryResult, error) {
 			return nil, fmt.Errorf("wal: recover: %w", err)
 		}
 
-		// Step 3: Tail corruption — truncate the last segment and accept
+		// Step 3: Tail corruption — truncate the last segment per design
-		// partial data loss for Phase 1.
+		// §3.2 line 787-800.
 		result := &RecoveryResult{
 			NextSequence:  nextSequence,
 			Truncated:     true,
-			TruncateError: err,
+			TruncateError: err, // informational: tail corruption was detected
 		}
 
-		// Determine nextSegmentID from scanned segments.
 		segments, scanErr := ScanSegments(dir, recoverySegmentID)
 		if scanErr != nil {
 			return nil, fmt.Errorf("wal: recover: scan after tail corruption: %w", scanErr)
@@ -56,27 +58,31 @@ func Recover(dir string, replayer BatchReplayer) (*RecoveryResult, error) {
 			result.NextSegmentID = recoverySegmentID
 		}
 
-		// Truncate the last segment file to remove corrupted tail.
 		if len(segments) > 0 {
 			lastSeg := segments[len(segments)-1]
-			validOffset, truncErr := findValidOffset(lastSeg.FilePath)
+			lastCompleteBatchEnd, findErr := findLastCompleteBatchEnd(lastSeg.FilePath)
-			if truncErr != nil {
+			if findErr != nil {
-				// Best-effort: record the truncation error but don't fail recovery.
+				return nil, fmt.Errorf("wal: recover: find truncation offset: %w", findErr)
-				result.TruncateError = fmt.Errorf("%w (find valid offset: %v)", err, truncErr)
+			}
-			} else if truncErr := truncateSegment(lastSeg.FilePath, validOffset); truncErr != nil {
+
-				result.TruncateError = fmt.Errorf("%w (truncate: %v)", err, truncErr)
+			// Phase 1: truncated segment is always segments[last], no
+			// trailing empty segments to clean up. C4 fix will need to
+			// identify trailing empties based on the actually-corrupted
+			// segment's index, which is not the same as segments[last].
+			var emptyTrailing []string
+
+			if err := truncateAndPersist(lastSeg.FilePath, lastCompleteBatchEnd, dir, emptyTrailing); err != nil {
+				// Per design §3.2 line 799: DB must NOT enter writable state.
+				return nil, fmt.Errorf("wal: recover: persist tail truncation: %w", err)
 			}
 		}
 
-		// Count replayed entries by re-scanning the replayer state.
+		result.ReplayedEntries = 0 // Phase 1: replayer interface doesn't expose count
-		// For Phase 1 we accept that ReplayedEntries may be approximate;
-		// the replayer interface doesn't expose a count.
-		result.ReplayedEntries = 0 // caller can inspect replayer directly
 
 		// Per design §3.2 line 280, recovery repair must NOT update MANIFEST.
-		// The truncated WAL state is persisted via ftruncate (see C5 for the
+		// The truncated WAL state is persisted via ftruncate + fsync segment
-		// remaining fsync gaps). MANIFEST can only advance via checkpoint
+		// + fsync dir (see truncateAndPersist). MANIFEST can only advance via
-		// (MemTable flush) in future phases.
+		// checkpoint (MemTable flush) in future phases.
 
 		return result, nil
 	}
@@ -114,23 +120,76 @@ func resolveRecoverySegmentID(dir string) (uint64, error) {
 	return mf.RecoverySegmentID, nil
 }
 
-// truncateSegment truncates the file at filePath to validOffset bytes,
+// segmentFsyncFn is the package-level indirection for fsyncing a truncated
-// removing any corrupted data after that point.
+// segment file. Tests that override this must not use t.Parallel().
-func truncateSegment(filePath string, validOffset int64) error {
+// Same pattern as dirFsyncFn (see wal/dir_fsync.go).
-	if validOffset < 0 {
+var segmentFsyncFn = segmentFsync
-		return fmt.Errorf("wal: truncate: invalid offset %d", validOffset)
+
+func segmentFsync(filePath string) error {
+	f, err := os.OpenFile(filePath, os.O_WRONLY, 0o644)
+	if err != nil {
+		return fmt.Errorf("open for fsync: %w", err)
 	}
-	return os.Truncate(filePath, validOffset)
+	defer f.Close()
+	if err := f.Sync(); err != nil {
+		return fmt.Errorf("fsync: %w", err)
+	}
+	return nil
 }
 
-// findValidOffset parses a segment file and returns the byte offset of the
+// truncateAndPersist executes the 4-step tail-truncation protocol per
-// last valid record boundary. The offset includes the file header size.
+// design §3.2 line 787-794. Any step failure is fatal: per line 799,
-func findValidOffset(filePath string) (int64, error) {
+// DB must NOT enter writable state if truncation cannot be persisted.
-	// Re-parse the file to find where valid records end.
+func truncateAndPersist(
-	// We need to track the byte offset as we parse.
+	filePath string,
+	lastCompleteBatchEnd int64,
+	dir string,
+	emptyTrailingSegments []string,
+) error {
+	// Step 1: ftruncate
+	if lastCompleteBatchEnd < 0 {
+		return fmt.Errorf("wal: invalid truncate offset %d", lastCompleteBatchEnd)
+	}
+	if err := os.Truncate(filePath, lastCompleteBatchEnd); err != nil {
+		return fmt.Errorf("ftruncate %s to %d: %w", filePath, lastCompleteBatchEnd, err)
+	}
+
+	// Step 2: fsync the truncated segment
+	if err := segmentFsyncFn(filePath); err != nil {
+		return fmt.Errorf("fsync truncated segment: %w", err)
+	}
+
+	// Step 3: delete empty trailing segments
+	for _, segPath := range emptyTrailingSegments {
+		if err := os.Remove(segPath); err != nil {
+			return fmt.Errorf("remove empty segment %s: %w", segPath, err)
+		}
+	}
+
+	// Step 4: fsync WAL directory (reuses C6's dirFsyncFn)
+	if err := dirFsyncFn(dir); err != nil {
+		return fmt.Errorf("fsync WAL dir after truncation: %w", err)
+	}
+	return nil
+}
+
+// findLastCompleteBatchEnd walks the segment file, runs physical records
+// through the FragmentCollector state machine, and returns the byte offset
+// of the END of the last complete WAL Batch.
+//
+// This is the correct truncation target per design §3.2 line 786. A previous
+// version (findValidOffset) only checked physical record CRCs, missing the
+// case where a First + Middle* fragment chain has no Last (H8 bug): physical
+// CRCs pass but no complete batch exists at that offset.
+//
+// Block-boundary handling: WAL format allows a full block to end with zero
+// padding when the next record doesn't fit (see BlockWriter.paddingNeeded).
+// This function CONTINUES to the next block on padding in a full block, and
+// only RETURNS on padding in a short (final) block or actual corruption.
+func findLastCompleteBatchEnd(filePath string) (int64, error) {
 	f, err := os.Open(filePath)
 	if err != nil {
-		return 0, fmt.Errorf("open for offset scan: %w", err)
+		return 0, fmt.Errorf("open %s: %w", filePath, err)
 	}
 	defer f.Close()
 
@@ -138,58 +197,56 @@ func findValidOffset(filePath string) (int64, error) {
 		return 0, fmt.Errorf("seek past header: %w", err)
 	}
 
-	validOffset := int64(WalFileHeaderSize)
+	collector := NewFragmentCollector()
+	lastCompleteEnd := int64(WalFileHeaderSize)
+	blockStartOffset := int64(WalFileHeaderSize)
 	buf := make([]byte, WalBlockSize)
 
 	for {
 		n, readErr := f.Read(buf)
-		if readErr != nil {
+		if n > 0 {
-			break
+			blockData := buf[:n]
-		}
+			isFullBlock := n == WalBlockSize && readErr == nil
-		if n == 0 {
+			pos := 0
-			break
+			for pos < len(blockData) {
-		}
+				remaining := len(blockData) - pos
 
-		blockData := buf[:n]
+				if remaining < PhysicalRecordHeaderSize {
-		blockStartOffset := validOffset
+					if isFullBlock {
-		pos := 0
+						break // padding in full block, continue to next block
-
+					}
-		for pos < len(blockData) {
+					return lastCompleteEnd, nil // tail padding in short block
-			remaining := len(blockData) - pos
-
-			if remaining < PhysicalRecordHeaderSize {
-				// Check if remaining bytes are zero-padding.
-				if isAllZeros(blockData[pos:]) {
-					// Valid padding — update offset to end of last valid record.
-					validOffset = blockStartOffset + int64(pos)
 				}
-				// Either way, we're done with this block.
-				break
-			}
 
-			if isAllZeros(blockData[pos : pos+PhysicalRecordHeaderSize]) {
+				if isAllZeros(blockData[pos : pos+PhysicalRecordHeaderSize]) {
-				if isAllZeros(blockData[pos:]) {
+					if isFullBlock {
-					validOffset = blockStartOffset + int64(pos)
+						break // zero-led padding in full block, continue
+					}
+					return lastCompleteEnd, nil // tail padding
 				}
-				break
-			}
 
-			_, consumed, err := DecodePhysicalRecord(blockData[pos:])
+				rec, consumed, err := DecodePhysicalRecord(blockData[pos:])
-			if err != nil {
+				if err != nil {
-				// Corruption starts here — offset is up to last valid record.
+					return lastCompleteEnd, nil // physical corruption
-				validOffset = blockStartOffset + int64(pos)
+				}
-				return validOffset, nil
+				if err := collector.Append(rec.Type, rec.Payload); err != nil {
-			}
+					return lastCompleteEnd, nil // fragment state machine rejected
+				}
 
-			// Valid record found.
+				pos += consumed
-			validOffset = blockStartOffset + int64(pos+consumed)
+				recordEndAbsolute := blockStartOffset + int64(pos)
-			pos += consumed
+
+				if collector.IsComplete() {
+					lastCompleteEnd = recordEndAbsolute
+					collector.Reset()
+				}
+			}
+			blockStartOffset += int64(n)
 		}
-
+		if readErr != nil || n < WalBlockSize {
-		if n < WalBlockSize {
 			break
 		}
 	}
 
-	return validOffset, nil
+	return lastCompleteEnd, nil
 }

wal/recover_offset_test.go

@@ -0,0 +1,363 @@
+package wal
+
+import (
+	"errors"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+// writeRawSegmentHeader writes a 32-byte WAL header to filePath. Used by
+// findLastCompleteBatchEnd tests to construct minimal segment files.
+func writeRawSegmentHeader(t *testing.T, filePath string) {
+	t.Helper()
+	hdr := &WalFileHeader{
+		BlockSize:     32 * 1024,
+		SegmentID:     0,
+		StartSequence: 0,
+	}
+	encoded := EncodeWalHeader(hdr)
+	if err := os.WriteFile(filePath, encoded[:], 0o644); err != nil {
+		t.Fatalf("WriteFile header: %v", err)
+	}
+}
+
+// appendRawBytes appends arbitrary bytes to filePath.
+func appendRawBytes(t *testing.T, filePath string, data []byte) {
+	t.Helper()
+	f, err := os.OpenFile(filePath, os.O_WRONLY|os.O_APPEND, 0o644)
+	if err != nil {
+		t.Fatalf("OpenFile append: %v", err)
+	}
+	defer f.Close()
+	if _, err := f.Write(data); err != nil {
+		t.Fatalf("Write: %v", err)
+	}
+}
+
+// appendFullRecord encodes and appends a complete WAL batch as a Full record.
+func appendFullRecord(t *testing.T, filePath string, batch []byte) {
+	t.Helper()
+	rec := EncodePhysicalRecord(RecFull, batch)
+	appendRawBytes(t, filePath, rec)
+}
+
+// appendFragment encodes and appends a single fragment record.
+func appendFragment(t *testing.T, filePath string, recType uint8, payload []byte) {
+	t.Helper()
+	rec := EncodePhysicalRecord(recType, payload)
+	appendRawBytes(t, filePath, rec)
+}
+
+func TestFindLastCompleteBatchEnd_CleanSegment(t *testing.T) {
+	dir := t.TempDir()
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+
+	batchA := []byte("batch-A-content")
+	batchB := []byte("batch-B-content")
+	appendFullRecord(t, filePath, batchA)
+	endOfA, _ := fileSize(filePath)
+	appendFullRecord(t, filePath, batchB)
+	endOfB, _ := fileSize(filePath)
+
+	got, err := findLastCompleteBatchEnd(filePath)
+	if err != nil {
+		t.Fatalf("findLastCompleteBatchEnd: %v", err)
+	}
+	if got != endOfB {
+		t.Errorf("got %d, want %d (end of Batch B)", got, endOfB)
+	}
+	if got == endOfA {
+		t.Errorf("got end of Batch A, should be end of Batch B")
+	}
+}
+
+// Regression guard for H8: partial fragment tail must return end of last
+// COMPLETE batch, not end of last physical record.
+func TestFindLastCompleteBatchEnd_PartialTailFragment(t *testing.T) {
+	dir := t.TempDir()
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+
+	batchA := []byte("batch-A-content")
+	appendFullRecord(t, filePath, batchA)
+	endOfA, _ := fileSize(filePath)
+
+	// Append First + Middle fragments (no Last) — H8 case.
+	appendFragment(t, filePath, RecFirst, []byte("first-fragment-data"))
+	appendFragment(t, filePath, RecMiddle, []byte("middle-fragment-data"))
+
+	got, err := findLastCompleteBatchEnd(filePath)
+	if err != nil {
+		t.Fatalf("findLastCompleteBatchEnd: %v", err)
+	}
+	if got != endOfA {
+		t.Errorf("got %d, want %d (end of Batch A, NOT end of Middle fragment)", got, endOfA)
+	}
+}
+
+func TestFindLastCompleteBatchEnd_NoBatches(t *testing.T) {
+	dir := t.TempDir()
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+
+	got, err := findLastCompleteBatchEnd(filePath)
+	if err != nil {
+		t.Fatalf("findLastCompleteBatchEnd: %v", err)
+	}
+	if got != int64(WalFileHeaderSize) {
+		t.Errorf("got %d, want %d (WalFileHeaderSize)", got, WalFileHeaderSize)
+	}
+}
+
+func TestFindLastCompleteBatchEnd_PhysicalCorruption(t *testing.T) {
+	dir := t.TempDir()
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+
+	batchA := []byte("batch-A-content")
+	appendFullRecord(t, filePath, batchA)
+	endOfA, _ := fileSize(filePath)
+
+	// Append corrupted bytes (will fail CRC).
+	appendRawBytes(t, filePath, []byte{0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF})
+
+	got, err := findLastCompleteBatchEnd(filePath)
+	if err != nil {
+		t.Fatalf("findLastCompleteBatchEnd: %v", err)
+	}
+	if got != endOfA {
+		t.Errorf("got %d, want %d (end of Batch A, before corruption)", got, endOfA)
+	}
+}
+
+func TestFindLastCompleteBatchEnd_PartialBatchOnly(t *testing.T) {
+	dir := t.TempDir()
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+
+	// Only First + Middle fragments, no complete batch ever.
+	appendFragment(t, filePath, RecFirst, []byte("first-data"))
+	appendFragment(t, filePath, RecMiddle, []byte("middle-data"))
+
+	got, err := findLastCompleteBatchEnd(filePath)
+	if err != nil {
+		t.Fatalf("findLastCompleteBatchEnd: %v", err)
+	}
+	if got != int64(WalFileHeaderSize) {
+		t.Errorf("got %d, want %d (no complete batch, stay at header)", got, WalFileHeaderSize)
+	}
+}
+
+// Oracle BLOCKING test: must continue past padding in a full block to read
+// the next block. Original code returned on first padding, truncating all
+// later batches.
+func TestFindLastCompleteBatchEnd_BlockBoundaryPadding(t *testing.T) {
+	dir := t.TempDir()
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+
+	// Batch A: large enough to nearly fill block 1.
+	// Block 1 layout: [32-byte header][Batch A (Full record)] [padding to end]
+	// We need: 32 + len(Full record of A) + padding == 32 + WalBlockSize
+	// Full record = 7 (header) + len(payload). So:
+	//   len(A) such that 7 + len(A) leaves < 7 bytes before block end
+	// Then writer pads to end of block and Batch B goes in block 2.
+
+	// Available in block 1 after file header = WalBlockSize - 32 = 32736
+	// We want Batch A record size to be 32736 - 6 = 32730 (leaving 6 bytes, < 7, padding)
+	// So payload size = 32730 - 7 = 32723
+	// Batch A = batch header (18) + entry bytes. Entry: Put with key + value.
+	// Simplest: use raw bytes (we're testing physical layout, not batch validity).
+	bigPayload := make([]byte, 32723)
+	for i := range bigPayload {
+		bigPayload[i] = byte('A')
+	}
+	recA := EncodePhysicalRecord(RecFull, bigPayload)
+	appendRawBytes(t, filePath, recA)
+
+	endOfBlock1 := int64(WalFileHeaderSize + WalBlockSize) // 32 + 32768
+	currentSize, _ := fileSize(filePath)
+	// Pad to end of block 1 with zeros.
+	padLen := int(endOfBlock1 - currentSize)
+	if padLen > 0 {
+		appendRawBytes(t, filePath, make([]byte, padLen))
+	}
+
+	// Batch B in block 2 (normal small batch after padding boundary).
+	batchB := []byte("batch-B")
+	recB := EncodePhysicalRecord(RecFull, batchB)
+	appendRawBytes(t, filePath, recB)
+	endOfB, _ := fileSize(filePath)
+
+	got, err := findLastCompleteBatchEnd(filePath)
+	if err != nil {
+		t.Fatalf("findLastCompleteBatchEnd: %v", err)
+	}
+	if got != endOfB {
+		t.Errorf("got %d, want %d (end of Batch B in block 2, after padding)", got, endOfB)
+	}
+	// Specifically: must NOT be at end of Batch A (would be ~32755, before padding).
+	if got < endOfBlock1 {
+		t.Errorf("got %d < %d (returned at end of Batch A, missed block 2)", got, endOfBlock1)
+	}
+}
+
+func TestFindLastCompleteBatchEnd_NonZeroTailPadding(t *testing.T) {
+	dir := t.TempDir()
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+
+	batchA := []byte("batch-A")
+	appendFullRecord(t, filePath, batchA)
+	endOfA, _ := fileSize(filePath)
+
+	// Append 3 non-zero bytes (< PhysicalRecordHeaderSize=7). This is
+	// technically invalid padding (per design: padding must be zeros), but
+	// findLastCompleteBatchEnd should still return end of last complete
+	// batch — this is "tail corruption" classification territory.
+	appendRawBytes(t, filePath, []byte{0xFF, 0xFF, 0xFF})
+
+	got, err := findLastCompleteBatchEnd(filePath)
+	if err != nil {
+		t.Fatalf("findLastCompleteBatchEnd: %v", err)
+	}
+	if got != endOfA {
+		t.Errorf("got %d, want %d (end of Batch A)", got, endOfA)
+	}
+}
+
+func TestFindLastCompleteBatchEnd_ZeroTailPadding(t *testing.T) {
+	dir := t.TempDir()
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+
+	batchA := []byte("batch-A")
+	appendFullRecord(t, filePath, batchA)
+	endOfA, _ := fileSize(filePath)
+
+	// Append 3 zero bytes (valid tail padding in a short final block).
+	appendRawBytes(t, filePath, []byte{0, 0, 0})
+
+	got, err := findLastCompleteBatchEnd(filePath)
+	if err != nil {
+		t.Fatalf("findLastCompleteBatchEnd: %v", err)
+	}
+	if got != endOfA {
+		t.Errorf("got %d, want %d (end of Batch A)", got, endOfA)
+	}
+}
+
+func fileSize(filePath string) (int64, error) {
+	fi, err := os.Stat(filePath)
+	if err != nil {
+		return 0, err
+	}
+	return fi.Size(), nil
+}
+
+// -------- truncateAndPersist tests --------
+
+func TestTruncateAndPersist_Success(t *testing.T) {
+	dir := t.TempDir()
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+	appendRawBytes(t, filePath, []byte("batch-A"))
+	appendRawBytes(t, filePath, []byte("extra-bytes-to-be-truncated"))
+
+	endOfA, _ := fileSize(filePath)
+	truncateOffset := int64(WalFileHeaderSize) + 7 // just past header, before "batch-A"
+
+	if err := truncateAndPersist(filePath, truncateOffset, dir, nil); err != nil {
+		t.Fatalf("truncateAndPersist: %v", err)
+	}
+
+	gotSize, _ := fileSize(filePath)
+	if gotSize != truncateOffset {
+		t.Errorf("file size = %d, want %d (truncated)", gotSize, truncateOffset)
+	}
+	_ = endOfA // not used; verify only that size matches truncate point
+}
+
+func TestTruncateAndPersist_FtruncateFailure(t *testing.T) {
+	dir := t.TempDir()
+	nonExistent := filepath.Join(dir, "no-such-file.wal")
+
+	err := truncateAndPersist(nonExistent, 100, dir, nil)
+	if err == nil {
+		t.Fatal("expected error on non-existent file")
+	}
+	if !strings.Contains(err.Error(), "ftruncate") {
+		t.Errorf("error should mention 'ftruncate', got: %v", err)
+	}
+}
+
+func TestTruncateAndPersist_DirFsyncFailure(t *testing.T) {
+	dir := t.TempDir()
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+	appendRawBytes(t, filePath, []byte("extra"))
+
+	orig := dirFsyncFn
+	dirFsyncFn = func(string) error { return errors.New("simulated dir fsync failure") }
+	t.Cleanup(func() { dirFsyncFn = orig })
+
+	err := truncateAndPersist(filePath, int64(WalFileHeaderSize), dir, nil)
+	if err == nil {
+		t.Fatal("expected error on dir fsync failure")
+	}
+	if !strings.Contains(err.Error(), "fsync WAL dir") {
+		t.Errorf("error should mention 'fsync WAL dir', got: %v", err)
+	}
+}
+
+func TestTruncateAndPersist_SegmentFsyncFailure(t *testing.T) {
+	dir := t.TempDir()
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+	appendRawBytes(t, filePath, []byte("extra"))
+
+	orig := segmentFsyncFn
+	segmentFsyncFn = func(string) error { return errors.New("simulated segment fsync failure") }
+	t.Cleanup(func() { segmentFsyncFn = orig })
+
+	err := truncateAndPersist(filePath, int64(WalFileHeaderSize), dir, nil)
+	if err == nil {
+		t.Fatal("expected error on segment fsync failure")
+	}
+	if !strings.Contains(err.Error(), "fsync truncated segment") {
+		t.Errorf("error should mention 'fsync truncated segment', got: %v", err)
+	}
+}
+
+// Oracle nice-to-have: verify retry after dir-fsync failure. The first
+// call fails after ftruncate succeeded; the second call (with fsync
+// restored) must succeed and the file must end up correctly truncated.
+func TestTruncateAndPersist_RetryAfterDirFsyncFailure(t *testing.T) {
+	dir := t.TempDir()
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+	appendRawBytes(t, filePath, []byte("extra-bytes"))
+
+	truncateOffset := int64(WalFileHeaderSize)
+
+	// First call: inject dir fsync failure.
+	orig := dirFsyncFn
+	dirFsyncFn = func(string) error { return errors.New("simulated") }
+	if err := truncateAndPersist(filePath, truncateOffset, dir, nil); err == nil {
+		t.Fatal("first call should fail")
+	}
+	dirFsyncFn = orig
+
+	// Second call: must succeed (ftruncate is idempotent).
+	if err := truncateAndPersist(filePath, truncateOffset, dir, nil); err != nil {
+		t.Fatalf("retry truncateAndPersist: %v", err)
+	}
+
+	gotSize, _ := fileSize(filePath)
+	if gotSize != truncateOffset {
+		t.Errorf("after retry, file size = %d, want %d", gotSize, truncateOffset)
+	}
+}

wal/recover_test.go

@@ -2,9 +2,11 @@ package wal
 
 import (
 	"bytes"
+	"errors"
 	"os"
 	"path/filepath"
 	"reflect"
+	"strings"
 	"testing"
 
 	"github.com/dailz/go-kv/manifest"
@@ -333,3 +335,119 @@ func fileExists(t *testing.T, path string) bool {
 	t.Fatalf("stat %s: %v", path, err)
 	return false
 }
+
+// Regression guard for C5+H8: end-to-end recovery with partial fragment
+// tail must persist truncation at the last COMPLETE batch boundary
+// (H8), and the truncation must be persisted with all 4 steps (C5).
+// After repair, second recovery must not see corruption.
+func TestRecoverPartialFragmentTailIdempotent(t *testing.T) {
+	dir := t.TempDir()
+	batchA := []*WalEntry{makePutEntry("key-A", "val-A")}
+	batchB := []*WalEntry{makePutEntry("key-B", "val-B")}
+
+	filePath := writeTestSegment(t, dir, 0, 0, [][]*WalEntry{batchA, batchB})
+
+	// Compute exact byte offset where Batch B's Full record ends.
+	// Layout: [header][Batch A Full record][Batch B Full record][padding to 32KB]
+	encA, _ := EncodeWalBatch(0, batchA)
+	encB, _ := EncodeWalBatch(1, batchB)
+	endOfBatchB := int64(WalFileHeaderSize) +
+		int64(PhysicalRecordHeaderSize+len(encA)) +
+		int64(PhysicalRecordHeaderSize+len(encB))
+
+	fiBefore, _ := os.Stat(filePath)
+
+	// Append First + Middle* (no Last) to simulate partial fragment tail.
+	appendFileBytes(t, filePath, EncodePhysicalRecord(RecFirst, []byte("first-fragment-payload")))
+	appendFileBytes(t, filePath, EncodePhysicalRecord(RecMiddle, []byte("middle-fragment-payload")))
+
+	replayer1 := &mockReplayer{}
+	result1, err := Recover(dir, replayer1)
+	if err != nil {
+		t.Fatalf("1st Recover: %v", err)
+	}
+	if !result1.Truncated {
+		t.Fatal("1st Recover: Truncated = false, want true")
+	}
+
+	fiAfter, _ := os.Stat(filePath)
+	if fiAfter.Size() != endOfBatchB {
+		t.Errorf("file size after truncation = %d, want %d (end of Batch B, H8)",
+			fiAfter.Size(), endOfBatchB)
+	}
+	if fiAfter.Size() >= fiBefore.Size() {
+		t.Errorf("file should shrink after truncation: before=%d after=%d",
+			fiBefore.Size(), fiAfter.Size())
+	}
+
+	replayer2 := &mockReplayer{}
+	result2, err := Recover(dir, replayer2)
+	if err != nil {
+		t.Fatalf("2nd Recover: %v", err)
+	}
+	if result2.Truncated {
+		t.Error("2nd Recover: Truncated = true, want false (truncation should be persisted)")
+	}
+	if result1.NextSequence != result2.NextSequence {
+		t.Errorf("NextSequence differs: %d vs %d", result1.NextSequence, result2.NextSequence)
+	}
+}
+
+// Regression guard for C5: any truncation persist step failure must
+// fail Recover, causing DB.Open to fail. No swallowing allowed.
+func TestRecoverTruncationFailureFailsRecovery(t *testing.T) {
+	dir := t.TempDir()
+	filePath := writeTestSegment(t, dir, 0, 0, [][]*WalEntry{
+		{makePutEntry("key-A", "val-A")},
+	})
+	// Append corruption to trigger tail corruption path.
+	appendFileBytes(t, filePath, []byte{0xDE, 0xAD, 0xBE, 0xEF})
+
+	// Inject dir fsync failure (Step 4 of truncateAndPersist).
+	orig := dirFsyncFn
+	dirFsyncFn = func(string) error { return errors.New("simulated dir fsync failure") }
+	t.Cleanup(func() { dirFsyncFn = orig })
+
+	_, err := Recover(dir, &mockReplayer{})
+	if err == nil {
+		t.Fatal("expected Recover to fail when truncation persist fails")
+	}
+	if !strings.Contains(err.Error(), "persist tail truncation") {
+		t.Errorf("error should mention 'persist tail truncation', got: %v", err)
+	}
+}
+
+// Regression guard for design line 778-781: CRC-valid but batch-content-
+// invalid must hard-fail through DecodeWalBatch, NOT enter truncation path.
+func TestRecoverInvalidBatchNotTruncatable(t *testing.T) {
+	dir := t.TempDir()
+	// Build segment with: physical records CRC-valid, but assembled batch
+	// has invalid header (entryCount=0).
+	filePath := filepath.Join(dir, "segment-0.wal")
+	writeRawSegmentHeader(t, filePath)
+
+	// Construct an "invalid batch": WalBatchHeaderSize=18 bytes, with
+	// entryCount=0 (invalid per ReplayBatch check at recovery.go).
+	invalidBatch := make([]byte, WalBatchHeaderSize)
+	// flags(2) + baseSequence(8) + entryCount(4)=0 + entriesSize(4)=0
+	// All zeros, except entryCount=0 is invalid by itself.
+	// Encode as Full physical record (CRC-valid).
+	rec := EncodePhysicalRecord(RecFull, invalidBatch)
+	appendFileBytes(t, filePath, rec)
+
+	_, err := Recover(dir, &mockReplayer{})
+	if err == nil {
+		t.Fatal("expected Recover to fail on invalid batch content")
+	}
+	// Should NOT mention truncation — must be a different error path.
+	if strings.Contains(err.Error(), "truncat") {
+		t.Errorf("error should not be about truncation; got: %v", err)
+	}
+
+	// File must NOT have been truncated (size unchanged).
+	fi, _ := os.Stat(filePath)
+	if fi.Size() != int64(WalFileHeaderSize)+int64(len(rec)) {
+		t.Errorf("file was truncated; size = %d, want %d",
+			fi.Size(), int64(WalFileHeaderSize)+int64(len(rec)))
+	}
+}