Cluster: parse_duration()
  - test_parse_seconds          → parse_duration("5s")  == Duration::from_secs(5)
  - test_parse_minutes          → parse_duration("3m")  == Duration::from_secs(180)
  - test_parse_hours            → parse_duration("2h")  == Duration::from_secs(7200)
  - test_parse_zero             → parse_duration("0s")  == Duration::ZERO
  - test_parse_large            → parse_duration("999h") == Duration::from_secs(3596400)
  - test_parse_invalid_unit     → parse_duration("5x")  == Err(...)
  - test_parse_empty            → parse_duration("")     == Err(...)
  - test_parse_negative         → parse_duration("-1s")  == Err(...)

// BEFORE: 6 unit tests
#[test] fn encode_ascii()    { assert_eq!(encode("hello"), "aGVsbG8="); }
#[test] fn encode_empty()    { assert_eq!(encode(""), ""); }
#[test] fn encode_unicode()  { assert_eq!(encode("café"), "Y2Fmw6k="); }
#[test] fn decode_ascii()    { assert_eq!(decode("aGVsbG8="), "hello"); }
#[test] fn decode_empty()    { assert_eq!(decode(""), ""); }
#[test] fn decode_unicode()  { assert_eq!(decode("Y2Fmw6k="), "café"); }

// AFTER: 1 property test (covers infinite inputs)
proptest! {
    #[test]
    fn roundtrip(input in "\\PC*") {
        let encoded = encode(&input);
        let decoded = decode(&encoded).unwrap();
        prop_assert_eq!(input, decoded);
    }
}
// KEEP: encode_empty as anchor (documents empty-input behavior)

// BEFORE: 7 unit tests
#[test] fn status_200() { assert_eq!(status_text(200), "OK"); }
#[test] fn status_201() { assert_eq!(status_text(201), "Created"); }
#[test] fn status_400() { assert_eq!(status_text(400), "Bad Request"); }
#[test] fn status_404() { assert_eq!(status_text(404), "Not Found"); }
#[test] fn status_500() { assert_eq!(status_text(500), "Internal Server Error"); }
#[test] fn status_unknown() { assert_eq!(status_text(999), "Unknown"); }
#[test] fn status_zero() { assert_eq!(status_text(0), "Unknown"); }

// AFTER: 1 parameterized test
#[rstest]
#[case(200, "OK")]
#[case(201, "Created")]
#[case(400, "Bad Request")]
#[case(404, "Not Found")]
#[case(500, "Internal Server Error")]
#[case(999, "Unknown")]
#[case(0, "Unknown")]
fn status_text_mapping(#[case] code: u16, #[case] expected: &str) {
    assert_eq!(status_text(code), expected);
}

// BEFORE: 5 individual tests with complex setup
#[test] fn gate_short_window() {
    let spec = CharClassSpec { max_lower_pct: 95, min_window_len: 32 };
    assert!(gate_passes(&[b'a'; 31], spec));
}
#[test] fn gate_all_lowercase() {
    let spec = CharClassSpec { max_lower_pct: 95, min_window_len: 32 };
    assert!(!gate_passes(&[b'a'; 40], spec));
}
#[test] fn gate_mixed_case() {
    let spec = CharClassSpec { max_lower_pct: 95, min_window_len: 32 };
    let mut window = vec![b'a'; 38];
    window.extend_from_slice(b"AB");
    assert!(gate_passes(&window, spec));
}
// ... 2 more ...

// AFTER: 1 rstest with descriptive case names
#[rstest]
#[case::short_window_passes(&[b'a'; 31], 95, 32, true)]
#[case::all_lowercase_rejected(&[b'a'; 40], 95, 32, false)]
#[case::mixed_case_passes(&[b'a'; 35, b'A', b'B', b'C', b'D', b'E'], 95, 32, true)]
#[case::at_threshold_rejected(&[b'a'; 39, b'A'], 95, 32, false)]
#[case::high_threshold_passes(&[b'a'; 40], 100, 32, true)]
fn char_class_gate(
    #[case] window: &[u8],
    #[case] max_lower_pct: u8,
    #[case] min_window_len: usize,
    #[case] expected: bool,
) {
    let spec = CharClassSpec { max_lower_pct, min_window_len };
    assert_eq!(gate_passes(window, spec), expected);
}

// ANTI-PATTERN — do NOT write this:
#[test]
fn status_text_cases() {
    let cases = [
        (200, "OK"),
        (201, "Created"),
        (400, "Bad Request"),
    ];
    for (code, expected) in cases {
        assert_eq!(status_text(code), expected, "code={code}");
    }
}

// CORRECT — use rstest instead:
#[rstest]
#[case(200, "OK")]
#[case(201, "Created")]
#[case(400, "Bad Request")]
fn status_text_mapping(#[case] code: u16, #[case] expected: &str) {
    assert_eq!(status_text(code), expected);
}

// BEFORE: 12 unit tests trying to crash the parser
#[test] fn parse_null_bytes()   { let _ = parse(b"\x00\x00"); }
#[test] fn parse_huge_input()   { let _ = parse(&vec![0xFF; 10_000]); }
#[test] fn parse_truncated()    { let _ = parse(b"\x01\x02"); }
// ... 9 more ...

// AFTER: 1 fuzz target (explores billions of inputs)
fuzz_target!(|data: &[u8]| {
    let _ = parse(data);
});
// KEEP: 1-2 specific regression tests if they document known bugs

## Test Consolidation Report: [module/file]

### Cluster Analysis

#### Cluster 1: `function_name()` — N tests, M lines

**Tests in cluster:**
| # | Test | Input | Assertion |
|---|------|-------|-----------|
| 1 | `test_foo_basic` | "hello" | returns "HELLO" |
| 2 | `test_foo_empty` | "" | returns "" |
| 3 | `test_foo_unicode` | "café" | returns "CAFÉ" |
| ... | ... | ... | ... |

**Pattern detected:** All tests call `foo(input)` and assert exact output.
Inputs vary, assertion structure is identical.

**Recommended strategy:** Property-based test
**Rationale:** The invariant `foo(x).to_lowercase() == x.to_lowercase()` holds
for all inputs. A single proptest replaces all 8 unit tests.

**Consolidation:**
- REPLACE tests #1-#6 with proptest `prop_foo_case_invariant`
- KEEP `test_foo_empty` as anchor (documents empty-input behavior)
- KEEP `test_foo_regression_gh_42` (regression test, bug reference)

**Proposed code:**
```rust
proptest! {
    #[test]
    fn prop_foo_case_invariant(input in "\\PC{0,100}") {
        prop_assert_eq!(foo(&input).to_lowercase(), input.to_lowercase());
    }
}

## Decision Heuristics

### When to prefer proptest over rstest
- You can state a property about the output without knowing the exact value
- The input space is continuous or very large
- Roundtrip properties exist (encode/decode, serialize/deserialize)
- Order/sorting/containment invariants exist

### When to prefer rstest over proptest
- Each case has a specific expected output that must be exact
- The set of important cases is finite and known
- The mapping is arbitrary (no mathematical relationship)
- You want each case to appear as a named sub-test in output

### When to prefer rstest over table-driven (almost always)
- **Default to rstest for ALL finite case sets.** rstest gives you:
  - Per-case test names in `cargo test` output (e.g., `status_text_mapping::case_3`)
  - Independent execution — one failing case doesn't hide others
  - `#[ignore]` on individual cases
  - Native `cargo test --test-threads` parallelism per case
- Table-driven is only acceptable for dynamically generated cases or >50 entries

### When to prefer fuzz over all else
- The function processes untrusted/external input
- The goal is "never panic" rather than "correct output"
- You've been writing tests that try to "trick" the parser

### When NOT to consolidate
- Each test has genuinely different setup logic (not just different values)
- Tests verify different error paths with different error types
- Tests are regression tests for specific bugs (keep the history)
- The "consolidated" version would be harder to understand than the originals
- There are only 2-3 tests — consolidation overhead isn't worth it

## Judgment Calls

- **Threshold**: Don't consolidate clusters of fewer than 4 tests unless the
  consolidation is obviously cleaner (e.g., perfect roundtrip property).
- **Mixed clusters**: If 6 of 8 tests consolidate but 2 are genuinely unique,
  consolidate the 6 and keep the 2. Don't force everything into one construct.
- **Error tests**: Tests for error cases often belong in a separate rstest or
  table, not mixed into the happy-path property test. Group by success/failure.
- **Naming**: Consolidated tests should have clear names describing the
  invariant or case set, not generic names like `test_all_cases`.

## Related Skills

- `/test-dedup` — Remove tests subsumed by higher-level tests (complementary)
- `/test-strategy` — Decide what kind of test to write for new code