chore: repoint URLs at git.swaits.com and bump for release

Rewrites remaining github.com / git.sr.ht references to git.swaits.com,
translating GitHub URL shapes to Gitea's (/blob/<ref>/ -> /src/branch/<ref>/,
/tree/<tag> -> /src/tag/<tag>, /actions/workflows/X -> /actions?workflow=X,
/edit/ -> /_edit/).

Left pointing at GitHub on purpose: pull-request links (PRs were not migrated)
and GitHub-only surfaces Gitea lacks (security advisories, discussions).

.cargo/mutants.toml

@@ -0,0 +1,18 @@
+# Use nextest for faster test execution
+test_tool = "nextest"
+
+# Skip slow proptests during mutation testing — the 60+ unit/interop tests
+# are sufficient to catch mutants. Full proptests run separately via just test-unit.
+additional_cargo_test_args = ["-E", "not test(proptests)"]
+
+# False positives excluded:
+#
+# - `> 0` → `>= 0` or `< 0` on usize: tautological (usize always >= 0)
+# - `< 254` → `<= 254` in max_run: both branches yield 254 at the boundary
+#
+# The legacy-msrv const_copy is excluded via #[mutants::skip] in source.
+exclude_re = [
+    "replace > with < in encode",
+    "replace > with >= in encode",
+    "replace < with <= in encode",
+]

.github/workflows/ci.yml

@@ -0,0 +1,69 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+
+env:
+  CARGO_TERM_COLOR: always
+
+jobs:
+  test:
+    name: Test (${{ matrix.rust }})
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false
+      matrix:
+        rust: [stable, "1.93"]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: dtolnay/rust-toolchain@master
+        with:
+          toolchain: ${{ matrix.rust }}
+      - run: cargo test
+
+  test-legacy-msrv:
+    name: Test (1.83, legacy-msrv)
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: dtolnay/rust-toolchain@master
+        with:
+          toolchain: "1.83"
+      - run: cargo test --features legacy-msrv
+
+  clippy:
+    name: Clippy
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: dtolnay/rust-toolchain@stable
+        with:
+          components: clippy
+      - run: cargo clippy -- -D warnings
+
+  fmt:
+    name: Format
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: dtolnay/rust-toolchain@stable
+        with:
+          components: rustfmt
+      - run: cargo fmt -- --check
+
+  doc:
+    name: Doc
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: dtolnay/rust-toolchain@stable
+      - run: cargo doc --no-deps
+
+  deny:
+    name: Cargo Deny
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: EmbarkStudios/cargo-deny-action@v2

.gitignore

@@ -2,4 +2,8 @@
 /fuzz/target
 /fuzz/corpus
 /fuzz/artifacts
+/mutants.out
+/mutants.out.old
 Cargo.lock
+__pycache__/
+*.pyc

CHANGELOG.md

@@ -0,0 +1,122 @@
+# Changelog
+
+## 0.3.4 — 2026-04-01
+
+### Changed
+
+- **Encode shrunk by 41 bytes** — simplified branch structure in encoder exit
+  path. 388 B encode, 392 B decode, 780 B total (was 821 B). Zero benchmark
+  regressions.
+- **Clarified decode invariant** — added inline comment explaining the subtle
+  trailing-0x01 COBS rule (end-of-data marker vs zero byte).
+- **Clippy cleanup** — `saturating_sub` in proptest (cosmetic, test-only).
+
+### Added
+
+- **GitHub Actions CI** — tests on stable + MSRV 1.93 + 1.83/legacy-msrv,
+  clippy, fmt, doc, and cargo-deny.
+- **`cargo-deny` integration** — `deny.toml` with permissive-only license
+  policy, advisory checks, source auditing. Added `just test-deny` gate.
+- **Encode-only fuzz target** — `fuzz_encode` exercises buffer-too-small
+  paths and validates output properties independently of decode.
+- **Feature coverage** — `just test-unit` and `just test-clippy` now test
+  both default and `legacy-msrv` features. `just test-doc` builds with
+  `--all-features`.
+
+### Fixed
+
+- **README accuracy** — overhead description changed from "exactly" to
+  "at most ... plus 1", added miri to test mentions, updated code size
+  numbers.
+- **CLAUDE.md rewrite** — updated stale API signatures, LOC count, roadmap,
+  dev-deps list, and development commands.
+
+## 0.3.3 — 2026-04-01
+
+### Fixed
+
+- **Fix compilation for downstream users** — `#[mutants::skip]` attribute was
+  not gated behind `cfg(test)`, causing unresolved module error when ucobs is
+  used as a dependency.
+
+## 0.3.2 — 2026-04-01 (yanked)
+
+### Testing
+
+- **Mutation testing: 100% kill rate** — 89 mutants tested, 59 caught, 30
+  unviable (type system rejects them), 0 missed. Uses `cargo-mutants` v27
+  with nextest. Added `mutants` crate for `#[mutants::skip]` annotations.
+- **111 tests** (was 106): added block-boundary interop verification against
+  corncobs, exact-fit and too-small decode buffer sizing for the batch zero
+  path.
+
+## 0.3.1 — 2026-03-31
+
+### Performance
+
+- **Decode: rewritten with shrinking-slice pattern** — decode zeros is 14%
+  faster (25,018 vs 29,114 instructions at 4 KB), decode nonzero flipped from
+  5% behind corncobs to 2% ahead, decode mixed closed from 8% behind to tied.
+  ucobs now wins or ties 16 of 18 benchmarks.
+- **Encode zeros: 6% faster** — zero fast path moved before `split_at` calls,
+  eliminating per-byte overhead. Now tied with cobs (previously 6% behind).
+- **Code size reduced** — 821 B total (was 869 B). Encode 429 B, decode 392 B.
+
+### Changed
+
+- **Replaced criterion with iai-callgrind** — benchmarks now measure
+  deterministic instruction counts via Valgrind instead of wall-clock time.
+  Zero run-to-run variance. Requires `valgrind` installed.
+- **Expanded benchmark coverage** — 9 payload sizes (0–4096 B) × 3 patterns
+  × 3 crates × encode/decode = 162 benchmarks, all running in ~2 minutes.
+- **Code size measurement** added to `just bench` via `just bench-size`.
+- **Mutation testing** added via `cargo-mutants` (`just test-mutants`).
+- Added `mise.toml` for dev tool management.
+
+## 0.3.0 — 2026-03-31
+
+### Performance
+
+- **Encode nonzero/mixed: ~2x faster** — uses const `copy_from_slice` (memcpy at
+  runtime) via Rust 1.93's stabilization of const slice methods. Sub-slice scan
+  pattern helps LLVM eliminate bounds checks.
+- **Encode zeros: ~50% faster** — fast path skips scan loop overhead for leading
+  zero bytes.
+- **Decode zeros: ~4x faster** — batch consecutive `0x01` codes and fill zeros
+  via `fill(0)` (memset) instead of per-byte iteration.
+- Encode now wins or ties corncobs on nonzero/mixed workloads.
+- Decode zeros now leads all competitors by 3x.
+
+### Added
+
+- `legacy-msrv` feature flag — enables compilation on Rust 1.83+ (e.g. Xtensa
+  toolchain) by replacing const `copy_from_slice` with a manual byte loop.
+  All other optimizations (sub-slice scan, zero fast path, decode batch fill)
+  work without this feature on any supported Rust version.
+
+### Changed
+
+- **MSRV bumped from 1.83 to 1.93** — required for const `copy_from_slice`.
+  Use the `legacy-msrv` feature for older toolchains.
+- Dev-dependencies updated: `cobs` 0.3 to 0.5, `criterion` 0.5 to 0.8,
+  `rand` 0.8 to 0.10.
+
+## 0.2.0 — 2026-03-30
+
+### Added
+
+- `const fn` encode — encode data at compile time with zero runtime cost.
+- Comprehensive documentation with 7 usage examples (basic, buffer sizing,
+  framing, stream parsing, compile-time encoding, error handling).
+- Initial encode/decode throughput optimizations.
+
+### Changed
+
+- MSRV set to 1.83.
+
+## 0.1.0 — 2026-03-29
+
+- Initial release.
+- `no_std`, zero-alloc, `unsafe`-free COBS encoder and decoder.
+- 106 tests: canonical vectors, property-based (proptest), fuzz targets,
+  cross-crate interop with `corncobs` and `cobs`.

CLAUDE.md

@@ -10,8 +10,8 @@
 
 1. **Correct** — 100% COBS spec-compliant (Cheshire & Baker 1999). No shortcuts, no deviations.
 2. **Interoperable** — proven byte-for-byte compatibility with `corncobs`, `cobs` crates and Python `cobs` packages.
-3. **Insanely tested** — canonical vectors, property-based tests (proptest), fuzz targets (cargo-fuzz), cross-crate interop, randomized payloads. If you think testing is done, you're 10% done.
-4. **Fast** — benchmark against every competing implementation. Must be at least as fast, ideally fastest. Benchmarks are CI-enforced.
+3. **Insanely tested** — canonical vectors, property-based tests (proptest), fuzz targets (cargo-fuzz), cross-crate interop, miri, mutation testing, randomized payloads. If you think testing is done, you're 10% done.
+4. **Fast** — benchmark against every competing implementation. Must be at least as fast, ideally fastest. Benchmarks run via `just bench`.
 5. **Tiny** — `no_std`, zero-alloc, zero runtime dependencies. ~140 lines of implementation. Runs on 8-bit MCUs and servers alike.
 6. **Trustworthy** — small enough to audit by hand. No unsafe. No panics in release. Fuzz-hardened.
 
@@ -22,7 +22,7 @@ Single-file library: `src/lib.rs`
 ### Public API (3 functions)
 
 ```rust
-pub fn encode(src: &[u8], dest: &mut [u8]) -> Option<usize>
+pub const fn encode(src: &[u8], dest: &mut [u8]) -> Option<usize>
 pub fn decode(src: &[u8], dest: &mut [u8]) -> Option<usize>
 pub const fn max_encoded_len(src_len: usize) -> usize
 ```
@@ -32,26 +32,29 @@ pub const fn max_encoded_len(src_len: usize) -> usize
 ## Development
 
 ```sh
-# Run all tests
-cargo test
+# Run all quality gates (tests, clippy, fmt, doc, fuzz, miri, mutants)
+just test
+
+# Unit + property + interop tests only
+just test-unit
+
+# Benchmarks (iai-callgrind, requires valgrind)
+just bench
 
 # Fuzz (requires nightly + cargo-fuzz)
-cd fuzz
-cargo +nightly fuzz run fuzz_decode -- -max_total_time=60
-cargo +nightly fuzz run fuzz_roundtrip -- -max_total_time=60
+just test-fuzz
 ```
 
 ## Dev Dependencies (test-only)
 
+- `cobs` — cross-validation against the `cobs` 0.5 crate
 - `corncobs` — cross-validation against another COBS implementation
+- `iai-callgrind` — deterministic instruction-count benchmarks
+- `mutants` — mutation testing annotations
 - `proptest` — property-based testing
 - `rand` — randomized test payloads
 
 ## Roadmap
 
-- More interop targets: `cobs` crate, Python `cobs` package
-- Benchmarks (criterion) vs corncobs, cobs, and others
 - quickcheck in addition to proptest
 - Even more fuzz targets and edge-case generators
-- MSRV policy and CI
-- `#[deny(unsafe_code)]` badge

Cargo.toml

@@ -1,12 +1,13 @@
 [package]
 name = "ucobs"
-version = "0.1.0"
+version = "0.3.5"
 edition = "2021"
+rust-version = "1.93"
 license = "MIT"
 description = "μCOBS — Consistent Overhead Byte Stuffing. no_std, zero-alloc, fastest, most-tested COBS implementation."
 authors = ["Stephen Waits <steve@waits.net>"]
-repository = "https://github.com/swaits/ucobs"
-homepage = "https://github.com/swaits/ucobs"
+repository = "https://git.swaits.com/swaits/ucobs"
+homepage = "https://git.swaits.com/swaits/ucobs"
 documentation = "https://docs.rs/ucobs"
 readme = "README.md"
 keywords = ["cobs", "framing", "no-std", "embedded", "codec"]
@@ -14,8 +15,22 @@ categories = ["encoding", "no-std::no-alloc", "embedded"]
 
 [features]
 default = []
+legacy-msrv = []
+
+[lints.rust]
+unsafe_code = "forbid"
+
+[package.metadata.docs.rs]
+all-features = true
 
 [dev-dependencies]
+cobs = "0.5"
 corncobs = "0.1"
+iai-callgrind = "0.16"
+mutants = "0.0.4"
 proptest = "1"
-rand = "0.8"
+rand = "0.10"
+
+[[bench]]
+name = "bench"
+harness = false

README.md

@@ -8,8 +8,8 @@
 
 [Consistent Overhead Byte Stuffing](https://en.wikipedia.org/wiki/Consistent_Overhead_Byte_Stuffing)
 (COBS) encodes arbitrary bytes so that `0x00` never appears in the output,
-allowing `0x00` to serve as an unambiguous frame delimiter. Overhead is exactly
-1 byte per 254 input bytes (worst case).
+allowing `0x00` to serve as an unambiguous frame delimiter. Overhead is at most
+1 byte per 254 input bytes, plus 1.
 
 ## Features
 
@@ -18,46 +18,299 @@ allowing `0x00` to serve as an unambiguous frame delimiter. Overhead is exactly
 - **100% spec-compliant** — Cheshire & Baker, IEEE/ACM Transactions on Networking, 1999
 - **Proven interoperability** — cross-validated against `corncobs`, `cobs`, and Python `cobs`
 - **Insanely tested** — canonical vectors, property-based tests, fuzz targets,
-  cross-crate interop, randomized payloads, and more coming
+  cross-crate interop, miri, mutation testing, randomized payloads
 
-## Usage
+## Performance
+
+Benchmarked against the two other `no_std` COBS crates using
+[iai-callgrind](https://github.com/iai-callgrind/iai-callgrind) (instruction
+counts via Valgrind — deterministic, zero run-to-run variance). Three payload
+patterns: **all zeros** (worst case), **no zeros**, and **mixed** (0x00–0xFF
+cycling). Instruction count per call, lower is better.
+Legend: 🥇 winner, 🥉 last, 🤝 tie (<3%).
+
+### Encode
+
+| Payload | Pattern | μCOBS | `cobs` 0.5 | `corncobs` 0.1 |
+|---------|---------|------:|-----------:|---------------:|
+| 64 B    | zeros   | 🤝 1,171 | 🤝 1,164 |    🥉 4,098   |
+| 64 B    | nonzero | 🤝 512 | 🥉 1,100  |       🤝 514   |
+| 64 B    | mixed   | 🥇 523 |  🥉 1,101 |          570   |
+| 256 B   | zeros   | 🤝 4,243 | 🤝 4,236 |   🥉 15,810   |
+| 256 B   | nonzero | 🤝 1,552 | 🥉 3,992 |   🤝 1,531    |
+| 256 B   | mixed   | 🤝 1,561 | 🥉 3,993 |   🤝 1,585    |
+| 4096 B  | zeros   | 🤝 65,904 | 🤝 65,897 | 🥉 250,271  |
+| 4096 B  | nonzero | 🤝 22,077 | 🥉 61,993 | 🤝 21,711   |
+| 4096 B  | mixed   | 🤝 22,932 | 🥉 62,009 | 🤝 22,656   |
+
+### Decode
+
+| Payload | Pattern | μCOBS | `cobs` 0.5 | `corncobs` 0.1 |
+|---------|---------|------:|-----------:|---------------:|
+| 64 B    | zeros   | 🥇 499 | 🥉 2,354  |        1,507   |
+| 64 B    | nonzero | 🥇 179 | 🥉 1,778  |          185   |
+| 64 B    | mixed   |    226 | 🥉 1,787  |       🥇 206   |
+| 256 B   | zeros   | 🥇 1,473 | 🥉 8,882 |       5,539   |
+| 256 B   | nonzero | 🤝 240 | 🥉 6,607  |       🤝 247   |
+| 256 B   | mixed   |    285 | 🥉 6,613  |       🥇 266   |
+| 4096 B  | zeros   | 🥇 25,018 | 🥉 139,698 |   86,435   |
+| 4096 B  | nonzero | 🤝 1,315 | 🥉 103,253 |  🤝 1,337   |
+| 4096 B  | mixed   | 🤝 2,086 | 🥉 103,394 |  🤝 2,067   |
+
+### Code size
+
+Measured from `.text` section of release-optimized symbols (`encode` + `decode`
+combined). Run `just bench-size` to reproduce.
+
+| Crate          | `encode` | `decode` | Total |
+|----------------|--------:|---------:|------:|
+| μCOBS          |   388 B |    392 B | 780 B |
+| `cobs` 0.5     |   282 B |    494 B | 776 B |
+| `corncobs` 0.1 |   375 B |    262 B | 637 B |
+
+All three crates are under 1 KB. μCOBS and `cobs` are nearly tied;
+`corncobs` is the smallest. μCOBS's safe `const fn` encoding with
+`copy_from_slice`/memcpy requires `split_at` bounds checks that
+generate panic paths — the cost of combining compile-time safety with
+runtime speed. Embedded users targeting size over speed can build with
+`opt-level = "s"` to shrink further.
+
+### Crate properties
+
+| Property       | μCOBS  | `cobs` 0.5       | `corncobs` 0.1 |
+|----------------|:------:|:-----------------:|:--------------:|
+| `no_std`       | always | opt-in[^1]        | always         |
+| Zero-alloc     | yes    | opt-in[^2]        | yes            |
+| `unsafe`-free  | yes    | yes               | yes            |
+| `const fn` encode | yes | no                | no             |
+| Implementation | ~140 LOC | ~790 LOC        | ~645 LOC       |
+
+[^1]: Requires disabling the default `std` feature.
+[^2]: `alloc` is enabled by default via the `std` feature.
+
+> Benchmarked with iai-callgrind (Valgrind instruction counting) for
+> deterministic, reproducible results. Run `just bench` to reproduce.
+> Requires `valgrind` installed.
+
+## When to use μCOBS
+
+| You need… | μCOBS | `cobs` 0.5 | `corncobs` 0.1 |
+|---|:---:|:---:|:---:|
+| Minimal code to audit | **~140 LOC** | ~790 LOC | ~645 LOC |
+| Compile-time (`const fn`) encode | **yes** | no | no |
+| Fewest encode instructions | **ties cobs/corncobs** | zeros only | nonzero only |
+| Fewest decode instructions (zero-heavy) | **yes (3–5×)** | no | 2nd |
+| Fewest decode instructions (nonzero) | **yes/tied** | no | tied |
+| Fewest decode instructions (mixed) | tied at 4 KB | no | **slight edge at small sizes** |
+| Dead-simple API (3 functions) | **yes** | yes | more surface area |
+| In-place / streaming encode | no | no | **yes** |
+| `no_std` + zero-alloc by default | **yes** | opt-in | **yes** |
+| Thorough test suite | **111 tests, fuzz, proptest, miri, mutants** | basic | basic |
+
+**Choose μCOBS** if you want the smallest, most auditable COBS implementation
+with a `const fn` encoder, a dead-simple 3-function API, and leading or tied
+instruction efficiency across most workloads — plus dominant decode performance
+on zero-heavy data (3–5× fewer instructions than alternatives). Wins or ties
+16 of 18 benchmarks.
+
+**Choose `corncobs`** if you need in-place encoding, an iterator-based API,
+or your workload is dominated by small mixed-pattern decoding where it uses
+~8% fewer instructions at 64–256 B.
+
+**Choose `cobs`** if you need `std` convenience features or a streaming state
+machine. Note: `cobs` uses 50–75× more instructions than μCOBS/corncobs for
+decoding non-zero data.
+
+## Examples
+
+### Basic encode and decode
 
 ```rust
-// Encode
+// Encode: [0x11, 0x00, 0x33] → [0x02, 0x11, 0x02, 0x33]
 let mut buf = [0u8; 16];
 let n = ucobs::encode(&[0x11, 0x00, 0x33], &mut buf).unwrap();
 assert_eq!(&buf[..n], &[0x02, 0x11, 0x02, 0x33]);
 
-// Decode
+// Decode reverses it
 let mut out = [0u8; 16];
-let n = ucobs::decode(&[0x02, 0x11, 0x02, 0x33], &mut out).unwrap();
+let m = ucobs::decode(&buf[..n], &mut out).unwrap();
+assert_eq!(&out[..m], &[0x11, 0x00, 0x33]);
+```
+
+### Buffer sizing
+
+Use `max_encoded_len` to determine the required destination buffer size:
+
+```rust
+let data = [0x01, 0x02, 0x03];
+let max = ucobs::max_encoded_len(data.len()); // 4 bytes
+
+let mut buf = [0u8; 4];
+let n = ucobs::encode(&data, &mut buf).unwrap();
+assert_eq!(n, 4); // fits exactly
+```
+
+If the destination is too small, `encode` returns `None` rather than panicking:
+
+```rust
+let mut tiny = [0u8; 1];
+assert_eq!(ucobs::encode(&[0x01, 0x02], &mut tiny), None);
+```
+
+### Framing for transport
+
+Append a `0x00` sentinel after encoding to delimit frames on a wire. Strip
+it before decoding:
+
+```rust
+let data = [0x11, 0x00, 0x33];
+
+// Encode and append sentinel
+let mut frame = [0u8; 16];
+let n = ucobs::encode(&data, &mut frame).unwrap();
+frame[n] = 0x00;
+let wire = &frame[..n + 1];
+
+// Receive: strip sentinel, then decode
+let cobs_data = &wire[..wire.len() - 1];
+let mut out = [0u8; 16];
+let m = ucobs::decode(cobs_data, &mut out).unwrap();
+assert_eq!(&out[..m], &data);
+```
+
+### Parsing a stream of frames
+
+Split a byte stream on `0x00` to extract individual COBS frames:
+
+```rust
+// Two frames separated by 0x00 sentinels
+let stream = [
+    0x02, 0x11, 0x02, 0x33, 0x00,  // frame 1: [0x11, 0x00, 0x33]
+    0x03, 0xAA, 0xBB, 0x00,        // frame 2: [0xAA, 0xBB]
+];
+
+let mut out = [0u8; 16];
+let frames: Vec<&[u8]> = stream.split(|&b| b == 0x00)
+    .filter(|f| !f.is_empty())
+    .collect();
+
+let n = ucobs::decode(frames[0], &mut out).unwrap();
 assert_eq!(&out[..n], &[0x11, 0x00, 0x33]);
 
-// Buffer sizing
-let max = ucobs::max_encoded_len(256); // 258
+let n = ucobs::decode(frames[1], &mut out).unwrap();
+assert_eq!(&out[..n], &[0xAA, 0xBB]);
+```
+
+### Compile-time encoding
+
+`encode` is `const fn`, so you can build COBS-encoded tables at compile
+time with zero runtime cost:
+
+```rust
+const PING: [u8; 2] = {
+    let mut buf = [0u8; 2];
+    match ucobs::encode(&[0x01], &mut buf) {
+        Some(_) => buf,
+        None => panic!("buffer too small"),
+    }
+};
+
+const ACK: [u8; 3] = {
+    let mut buf = [0u8; 3];
+    match ucobs::encode(&[0x06, 0x00], &mut buf) {
+        Some(_) => buf,
+        None => panic!("buffer too small"),
+    }
+};
+
+// Baked into the binary — no runtime encoding needed
+assert_eq!(PING, [0x02, 0x01]);
+assert_eq!(ACK, [0x02, 0x06, 0x01]);
+```
+
+### Error handling
+
+Both `encode` and `decode` return `Option<usize>` — `None` on failure,
+never panic:
+
+```rust
+let mut buf = [0u8; 8];
+
+// Destination too small
+assert_eq!(ucobs::encode(&[1, 2, 3], &mut buf[..1]), None);
+
+// Malformed input (zero byte in encoded stream)
+assert_eq!(ucobs::decode(&[0x00], &mut buf), None);
+
+// Truncated frame (code byte promises more data than exists)
+assert_eq!(ucobs::decode(&[0x05, 0x11], &mut buf), None);
+
+// Empty input is valid
+assert_eq!(ucobs::encode(&[], &mut buf), Some(1)); // encodes to [0x01]
+assert_eq!(ucobs::decode(&[], &mut buf), Some(0)); // decodes to []
 ```
 
 ## API
 
 | Function | Signature | Description |
 |---|---|---|
-| `encode` | `(src: &[u8], dest: &mut [u8]) -> Option<usize>` | COBS-encode `src` into `dest`. Returns byte count or `None` if buffer too small. |
+| `encode` | `const fn(src: &[u8], dest: &mut [u8]) -> Option<usize>` | COBS-encode `src` into `dest`. Returns byte count or `None` if buffer too small. |
 | `decode` | `(src: &[u8], dest: &mut [u8]) -> Option<usize>` | Decode COBS data. Returns byte count or `None` if input is malformed. |
 | `max_encoded_len` | `const fn(src_len: usize) -> usize` | Maximum encoded size for a given input length. |
 
 **Note:** The encoder does *not* append a trailing `0x00` sentinel. Append it
 yourself when framing for transport.
 
+## Minimum Supported Rust Version
+
+The default build requires **Rust 1.93+** (for const `copy_from_slice`).
+
+### `legacy-msrv` feature
+
+If you're targeting a toolchain older than 1.93 (e.g. Xtensa), enable the
+`legacy-msrv` feature to fall back to a manual byte copy in the encoder:
+
+```toml
+[dependencies]
+ucobs = { version = "0.3", features = ["legacy-msrv"] }
+```
+
+All other optimizations (sub-slice scan, zero fast path, decode batch fill)
+are available regardless of this feature. Only the encode copy phase is
+affected — it uses a byte loop instead of `copy_from_slice`/memcpy.
+
 ## Testing
 
+μCOBS has one of the most thorough test suites of any COBS implementation —
+111 tests across 8 categories:
+
+- **Canonical vectors** — every example from Cheshire & Baker 1999 (the original
+  COBS paper), plus the full Wikipedia vector set
+- **External corpora** — vectors drawn from `cobs-c`, `nanocobs`, `cobs2-rs`,
+  Jacques Fortier's C implementation, and the Python `cobs` package
+- **Cross-crate interop** — byte-for-byte validation against both `corncobs` and
+  `cobs` crates on randomized payloads up to 4 KB
+- **Property-based tests** — 6 proptest suites covering round-trip correctness,
+  no-zero-in-output invariants, encoding properties, decoding properties,
+  structural/algebraic properties, and boundary-region behavior
+- **Fuzz targets** — 3 `cargo-fuzz` harnesses (decode, round-trip, small-dest)
+  for continuous coverage of edge cases
+- **Mutation testing** — `cargo-mutants` with 100% kill rate (59/59 viable
+  mutants caught; 30 more rejected by the type system)
+- **Boundary tests** — exhaustive coverage of 254/255-byte block boundaries,
+  multi-block splits, buffer-exact fits, and off-by-one dest sizes
+- **Compile-time verification** — `const` assertions that validate encode
+  correctness at compile time
+
 ```sh
 # Unit + property + interop tests
-cargo test
+just test-unit
 
-# Fuzz testing (requires nightly + cargo-fuzz)
-cd fuzz
-cargo +nightly fuzz run fuzz_decode -- -max_total_time=60
-cargo +nightly fuzz run fuzz_roundtrip -- -max_total_time=60
+# All quality gates (tests, clippy, fmt, doc, fuzz, miri, mutants)
+just test
+
+# Mutation testing only
+just test-mutants
 ```
 
 ## License

bench_python.py

@@ -0,0 +1,67 @@
+#!/usr/bin/env python3
+"""Benchmark Python cobs encode/decode vs ucobs (reference timing only).
+
+Requires: pip install cobs
+"""
+
+import time
+
+
+def payload_zeros(n: int) -> bytes:
+    return b"\x00" * n
+
+
+def payload_nonzero(n: int) -> bytes:
+    return bytes((i % 255 + 1) for i in range(n))
+
+
+def payload_mixed(n: int) -> bytes:
+    return bytes(i % 256 for i in range(n))
+
+
+def bench(func, data, iterations=10_000):
+    start = time.perf_counter()
+    for _ in range(iterations):
+        func(data)
+    elapsed = time.perf_counter() - start
+    return elapsed
+
+
+def main():
+    try:
+        from cobs import cobs as cobs_mod
+    except ImportError:
+        print("ERROR: python cobs package not installed. Run: pip install cobs")
+        return
+
+    sizes = [0, 1, 10, 64, 254, 255, 256, 1024, 4096]
+    payloads = [
+        ("zeros", payload_zeros),
+        ("nonzero", payload_nonzero),
+        ("mixed", payload_mixed),
+    ]
+    iterations = 10_000
+
+    print(f"{'operation':<30} {'size':>6} {'iters':>7} {'total_ms':>10} {'per_op_us':>10} {'MB/s':>10}")
+    print("-" * 80)
+
+    for name, make in payloads:
+        for size in sizes:
+            data = make(size)
+
+            # Encode
+            elapsed = bench(cobs_mod.encode, data, iterations)
+            per_op_us = (elapsed / iterations) * 1_000_000
+            mbps = (size * iterations / elapsed / 1_000_000) if elapsed > 0 and size > 0 else 0
+            print(f"{'python-cobs encode/' + name:<30} {size:>6} {iterations:>7} {elapsed*1000:>10.2f} {per_op_us:>10.3f} {mbps:>10.1f}")
+
+            # Decode
+            encoded = cobs_mod.encode(data)
+            elapsed = bench(cobs_mod.decode, encoded, iterations)
+            per_op_us = (elapsed / iterations) * 1_000_000
+            mbps = (size * iterations / elapsed / 1_000_000) if elapsed > 0 and size > 0 else 0
+            print(f"{'python-cobs decode/' + name:<30} {size:>6} {iterations:>7} {elapsed*1000:>10.2f} {per_op_us:>10.3f} {mbps:>10.1f}")
+
+
+if __name__ == "__main__":
+    main()

benches/bench.rs

@@ -0,0 +1,271 @@
+use iai_callgrind::{library_benchmark, library_benchmark_group, main};
+use std::hint::black_box;
+
+// ── Test payloads ───────────────────────────────────────────────────
+
+fn zeros(n: usize) -> Vec<u8> {
+    vec![0u8; n]
+}
+
+fn nonzero(n: usize) -> Vec<u8> {
+    (0..n).map(|i| (i % 255 + 1) as u8).collect()
+}
+
+fn mixed(n: usize) -> Vec<u8> {
+    (0..n).map(|i| (i % 256) as u8).collect()
+}
+
+// ── Setup helpers ───────────────────────────────────────────────────
+
+fn setup_encode(data: Vec<u8>) -> (Vec<u8>, Vec<u8>) {
+    let buf = vec![0u8; ucobs::max_encoded_len(data.len()) + 1];
+    (data, buf)
+}
+
+fn setup_decode_ucobs(data: Vec<u8>) -> (Vec<u8>, Vec<u8>) {
+    let mut enc = vec![0u8; ucobs::max_encoded_len(data.len()) + 1];
+    let n = ucobs::encode(&data, &mut enc).unwrap();
+    enc.truncate(n);
+    let dec = vec![0u8; data.len() + 1];
+    (enc, dec)
+}
+
+fn setup_encode_corncobs(data: Vec<u8>) -> (Vec<u8>, Vec<u8>) {
+    let buf = vec![0u8; corncobs::max_encoded_len(data.len())];
+    (data, buf)
+}
+
+fn setup_decode_corncobs(data: Vec<u8>) -> (Vec<u8>, Vec<u8>) {
+    let mut enc = vec![0u8; corncobs::max_encoded_len(data.len())];
+    let n = corncobs::encode_buf(&data, &mut enc);
+    enc.truncate(n);
+    let dec = vec![0u8; data.len() + 1];
+    (enc, dec)
+}
+
+fn setup_encode_cobs(data: Vec<u8>) -> (Vec<u8>, Vec<u8>) {
+    let buf = vec![0u8; cobs::max_encoding_length(data.len())];
+    (data, buf)
+}
+
+fn setup_decode_cobs(data: Vec<u8>) -> (Vec<u8>, Vec<u8>) {
+    let mut enc = vec![0u8; cobs::max_encoding_length(data.len())];
+    let n = cobs::encode(&data, &mut enc);
+    enc.truncate(n);
+    let dec = vec![0u8; data.len() + 1];
+    (enc, dec)
+}
+
+// ── Encode benchmarks ──────────────────────────────────────────────
+
+#[library_benchmark]
+#[bench::zeros_0(setup_encode(zeros(0)))]
+#[bench::zeros_1(setup_encode(zeros(1)))]
+#[bench::zeros_10(setup_encode(zeros(10)))]
+#[bench::zeros_64(setup_encode(zeros(64)))]
+#[bench::zeros_254(setup_encode(zeros(254)))]
+#[bench::zeros_255(setup_encode(zeros(255)))]
+#[bench::zeros_256(setup_encode(zeros(256)))]
+#[bench::zeros_1024(setup_encode(zeros(1024)))]
+#[bench::zeros_4096(setup_encode(zeros(4096)))]
+#[bench::nonzero_0(setup_encode(nonzero(0)))]
+#[bench::nonzero_1(setup_encode(nonzero(1)))]
+#[bench::nonzero_10(setup_encode(nonzero(10)))]
+#[bench::nonzero_64(setup_encode(nonzero(64)))]
+#[bench::nonzero_254(setup_encode(nonzero(254)))]
+#[bench::nonzero_255(setup_encode(nonzero(255)))]
+#[bench::nonzero_256(setup_encode(nonzero(256)))]
+#[bench::nonzero_1024(setup_encode(nonzero(1024)))]
+#[bench::nonzero_4096(setup_encode(nonzero(4096)))]
+#[bench::mixed_0(setup_encode(mixed(0)))]
+#[bench::mixed_1(setup_encode(mixed(1)))]
+#[bench::mixed_10(setup_encode(mixed(10)))]
+#[bench::mixed_64(setup_encode(mixed(64)))]
+#[bench::mixed_254(setup_encode(mixed(254)))]
+#[bench::mixed_255(setup_encode(mixed(255)))]
+#[bench::mixed_256(setup_encode(mixed(256)))]
+#[bench::mixed_1024(setup_encode(mixed(1024)))]
+#[bench::mixed_4096(setup_encode(mixed(4096)))]
+fn encode_ucobs((data, mut buf): (Vec<u8>, Vec<u8>)) {
+    black_box(ucobs::encode(black_box(&data), &mut buf));
+}
+
+#[library_benchmark]
+#[bench::zeros_0(setup_encode_corncobs(zeros(0)))]
+#[bench::zeros_1(setup_encode_corncobs(zeros(1)))]
+#[bench::zeros_10(setup_encode_corncobs(zeros(10)))]
+#[bench::zeros_64(setup_encode_corncobs(zeros(64)))]
+#[bench::zeros_254(setup_encode_corncobs(zeros(254)))]
+#[bench::zeros_255(setup_encode_corncobs(zeros(255)))]
+#[bench::zeros_256(setup_encode_corncobs(zeros(256)))]
+#[bench::zeros_1024(setup_encode_corncobs(zeros(1024)))]
+#[bench::zeros_4096(setup_encode_corncobs(zeros(4096)))]
+#[bench::nonzero_0(setup_encode_corncobs(nonzero(0)))]
+#[bench::nonzero_1(setup_encode_corncobs(nonzero(1)))]
+#[bench::nonzero_10(setup_encode_corncobs(nonzero(10)))]
+#[bench::nonzero_64(setup_encode_corncobs(nonzero(64)))]
+#[bench::nonzero_254(setup_encode_corncobs(nonzero(254)))]
+#[bench::nonzero_255(setup_encode_corncobs(nonzero(255)))]
+#[bench::nonzero_256(setup_encode_corncobs(nonzero(256)))]
+#[bench::nonzero_1024(setup_encode_corncobs(nonzero(1024)))]
+#[bench::nonzero_4096(setup_encode_corncobs(nonzero(4096)))]
+#[bench::mixed_0(setup_encode_corncobs(mixed(0)))]
+#[bench::mixed_1(setup_encode_corncobs(mixed(1)))]
+#[bench::mixed_10(setup_encode_corncobs(mixed(10)))]
+#[bench::mixed_64(setup_encode_corncobs(mixed(64)))]
+#[bench::mixed_254(setup_encode_corncobs(mixed(254)))]
+#[bench::mixed_255(setup_encode_corncobs(mixed(255)))]
+#[bench::mixed_256(setup_encode_corncobs(mixed(256)))]
+#[bench::mixed_1024(setup_encode_corncobs(mixed(1024)))]
+#[bench::mixed_4096(setup_encode_corncobs(mixed(4096)))]
+fn encode_corncobs((data, mut buf): (Vec<u8>, Vec<u8>)) {
+    black_box(corncobs::encode_buf(black_box(&data), &mut buf));
+}
+
+#[library_benchmark]
+#[bench::zeros_0(setup_encode_cobs(zeros(0)))]
+#[bench::zeros_1(setup_encode_cobs(zeros(1)))]
+#[bench::zeros_10(setup_encode_cobs(zeros(10)))]
+#[bench::zeros_64(setup_encode_cobs(zeros(64)))]
+#[bench::zeros_254(setup_encode_cobs(zeros(254)))]
+#[bench::zeros_255(setup_encode_cobs(zeros(255)))]
+#[bench::zeros_256(setup_encode_cobs(zeros(256)))]
+#[bench::zeros_1024(setup_encode_cobs(zeros(1024)))]
+#[bench::zeros_4096(setup_encode_cobs(zeros(4096)))]
+#[bench::nonzero_0(setup_encode_cobs(nonzero(0)))]
+#[bench::nonzero_1(setup_encode_cobs(nonzero(1)))]
+#[bench::nonzero_10(setup_encode_cobs(nonzero(10)))]
+#[bench::nonzero_64(setup_encode_cobs(nonzero(64)))]
+#[bench::nonzero_254(setup_encode_cobs(nonzero(254)))]
+#[bench::nonzero_255(setup_encode_cobs(nonzero(255)))]
+#[bench::nonzero_256(setup_encode_cobs(nonzero(256)))]
+#[bench::nonzero_1024(setup_encode_cobs(nonzero(1024)))]
+#[bench::nonzero_4096(setup_encode_cobs(nonzero(4096)))]
+#[bench::mixed_0(setup_encode_cobs(mixed(0)))]
+#[bench::mixed_1(setup_encode_cobs(mixed(1)))]
+#[bench::mixed_10(setup_encode_cobs(mixed(10)))]
+#[bench::mixed_64(setup_encode_cobs(mixed(64)))]
+#[bench::mixed_254(setup_encode_cobs(mixed(254)))]
+#[bench::mixed_255(setup_encode_cobs(mixed(255)))]
+#[bench::mixed_256(setup_encode_cobs(mixed(256)))]
+#[bench::mixed_1024(setup_encode_cobs(mixed(1024)))]
+#[bench::mixed_4096(setup_encode_cobs(mixed(4096)))]
+fn encode_cobs((data, mut buf): (Vec<u8>, Vec<u8>)) {
+    black_box(cobs::encode(black_box(&data), &mut buf));
+}
+
+// ── Decode benchmarks ──────────────────────────────────────────────
+
+#[library_benchmark]
+#[bench::zeros_0(setup_decode_ucobs(zeros(0)))]
+#[bench::zeros_1(setup_decode_ucobs(zeros(1)))]
+#[bench::zeros_10(setup_decode_ucobs(zeros(10)))]
+#[bench::zeros_64(setup_decode_ucobs(zeros(64)))]
+#[bench::zeros_254(setup_decode_ucobs(zeros(254)))]
+#[bench::zeros_255(setup_decode_ucobs(zeros(255)))]
+#[bench::zeros_256(setup_decode_ucobs(zeros(256)))]
+#[bench::zeros_1024(setup_decode_ucobs(zeros(1024)))]
+#[bench::zeros_4096(setup_decode_ucobs(zeros(4096)))]
+#[bench::nonzero_0(setup_decode_ucobs(nonzero(0)))]
+#[bench::nonzero_1(setup_decode_ucobs(nonzero(1)))]
+#[bench::nonzero_10(setup_decode_ucobs(nonzero(10)))]
+#[bench::nonzero_64(setup_decode_ucobs(nonzero(64)))]
+#[bench::nonzero_254(setup_decode_ucobs(nonzero(254)))]
+#[bench::nonzero_255(setup_decode_ucobs(nonzero(255)))]
+#[bench::nonzero_256(setup_decode_ucobs(nonzero(256)))]
+#[bench::nonzero_1024(setup_decode_ucobs(nonzero(1024)))]
+#[bench::nonzero_4096(setup_decode_ucobs(nonzero(4096)))]
+#[bench::mixed_0(setup_decode_ucobs(mixed(0)))]
+#[bench::mixed_1(setup_decode_ucobs(mixed(1)))]
+#[bench::mixed_10(setup_decode_ucobs(mixed(10)))]
+#[bench::mixed_64(setup_decode_ucobs(mixed(64)))]
+#[bench::mixed_254(setup_decode_ucobs(mixed(254)))]
+#[bench::mixed_255(setup_decode_ucobs(mixed(255)))]
+#[bench::mixed_256(setup_decode_ucobs(mixed(256)))]
+#[bench::mixed_1024(setup_decode_ucobs(mixed(1024)))]
+#[bench::mixed_4096(setup_decode_ucobs(mixed(4096)))]
+fn decode_ucobs((enc, mut buf): (Vec<u8>, Vec<u8>)) {
+    black_box(ucobs::decode(black_box(&enc), &mut buf));
+}
+
+#[library_benchmark]
+#[bench::zeros_0(setup_decode_corncobs(zeros(0)))]
+#[bench::zeros_1(setup_decode_corncobs(zeros(1)))]
+#[bench::zeros_10(setup_decode_corncobs(zeros(10)))]
+#[bench::zeros_64(setup_decode_corncobs(zeros(64)))]
+#[bench::zeros_254(setup_decode_corncobs(zeros(254)))]
+#[bench::zeros_255(setup_decode_corncobs(zeros(255)))]
+#[bench::zeros_256(setup_decode_corncobs(zeros(256)))]
+#[bench::zeros_1024(setup_decode_corncobs(zeros(1024)))]
+#[bench::zeros_4096(setup_decode_corncobs(zeros(4096)))]
+#[bench::nonzero_0(setup_decode_corncobs(nonzero(0)))]
+#[bench::nonzero_1(setup_decode_corncobs(nonzero(1)))]
+#[bench::nonzero_10(setup_decode_corncobs(nonzero(10)))]
+#[bench::nonzero_64(setup_decode_corncobs(nonzero(64)))]
+#[bench::nonzero_254(setup_decode_corncobs(nonzero(254)))]
+#[bench::nonzero_255(setup_decode_corncobs(nonzero(255)))]
+#[bench::nonzero_256(setup_decode_corncobs(nonzero(256)))]
+#[bench::nonzero_1024(setup_decode_corncobs(nonzero(1024)))]
+#[bench::nonzero_4096(setup_decode_corncobs(nonzero(4096)))]
+#[bench::mixed_0(setup_decode_corncobs(mixed(0)))]
+#[bench::mixed_1(setup_decode_corncobs(mixed(1)))]
+#[bench::mixed_10(setup_decode_corncobs(mixed(10)))]
+#[bench::mixed_64(setup_decode_corncobs(mixed(64)))]
+#[bench::mixed_254(setup_decode_corncobs(mixed(254)))]
+#[bench::mixed_255(setup_decode_corncobs(mixed(255)))]
+#[bench::mixed_256(setup_decode_corncobs(mixed(256)))]
+#[bench::mixed_1024(setup_decode_corncobs(mixed(1024)))]
+#[bench::mixed_4096(setup_decode_corncobs(mixed(4096)))]
+fn decode_corncobs((enc, mut buf): (Vec<u8>, Vec<u8>)) {
+    let _ = black_box(corncobs::decode_buf(black_box(&enc), &mut buf));
+}
+
+#[library_benchmark]
+#[bench::zeros_0(setup_decode_cobs(zeros(0)))]
+#[bench::zeros_1(setup_decode_cobs(zeros(1)))]
+#[bench::zeros_10(setup_decode_cobs(zeros(10)))]
+#[bench::zeros_64(setup_decode_cobs(zeros(64)))]
+#[bench::zeros_254(setup_decode_cobs(zeros(254)))]
+#[bench::zeros_255(setup_decode_cobs(zeros(255)))]
+#[bench::zeros_256(setup_decode_cobs(zeros(256)))]
+#[bench::zeros_1024(setup_decode_cobs(zeros(1024)))]
+#[bench::zeros_4096(setup_decode_cobs(zeros(4096)))]
+#[bench::nonzero_0(setup_decode_cobs(nonzero(0)))]
+#[bench::nonzero_1(setup_decode_cobs(nonzero(1)))]
+#[bench::nonzero_10(setup_decode_cobs(nonzero(10)))]
+#[bench::nonzero_64(setup_decode_cobs(nonzero(64)))]
+#[bench::nonzero_254(setup_decode_cobs(nonzero(254)))]
+#[bench::nonzero_255(setup_decode_cobs(nonzero(255)))]
+#[bench::nonzero_256(setup_decode_cobs(nonzero(256)))]
+#[bench::nonzero_1024(setup_decode_cobs(nonzero(1024)))]
+#[bench::nonzero_4096(setup_decode_cobs(nonzero(4096)))]
+#[bench::mixed_0(setup_decode_cobs(mixed(0)))]
+#[bench::mixed_1(setup_decode_cobs(mixed(1)))]
+#[bench::mixed_10(setup_decode_cobs(mixed(10)))]
+#[bench::mixed_64(setup_decode_cobs(mixed(64)))]
+#[bench::mixed_254(setup_decode_cobs(mixed(254)))]
+#[bench::mixed_255(setup_decode_cobs(mixed(255)))]
+#[bench::mixed_256(setup_decode_cobs(mixed(256)))]
+#[bench::mixed_1024(setup_decode_cobs(mixed(1024)))]
+#[bench::mixed_4096(setup_decode_cobs(mixed(4096)))]
+fn decode_cobs((enc, mut buf): (Vec<u8>, Vec<u8>)) {
+    black_box(
+        cobs::decode(black_box(&enc), &mut buf)
+            .map(|r| r.parsed_size())
+            .unwrap_or(0),
+    );
+}
+
+// ── Groups ─────────────────────────────────────────────────────────
+
+library_benchmark_group!(
+    name = encode;
+    benchmarks = encode_ucobs, encode_corncobs, encode_cobs
+);
+
+library_benchmark_group!(
+    name = decode;
+    benchmarks = decode_ucobs, decode_corncobs, decode_cobs
+);
+
+main!(library_benchmark_groups = encode, decode);

deny.toml

@@ -0,0 +1,37 @@
+[graph]
+all-features = true
+
+[advisories]
+yanked = "deny"
+ignore = [
+    # bincode 1.3.3 — transitive dev-dep via iai-callgrind (latest 0.16.1).
+    # Unmaintained but stable and complete; no security issue. Dev-only, never
+    # ships in ucobs. Remove this ignore when iai-callgrind drops bincode.
+    "RUSTSEC-2025-0141",
+]
+
+[licenses]
+# Permissive licenses only. All copyleft (GPL, LGPL, AGPL, MPL, etc.) is denied by omission.
+allow = [
+    "MIT",
+    "Apache-2.0",
+    "BSD-2-Clause",
+    "BSD-3-Clause",
+    "ISC",
+    "Zlib",
+    "Unlicense",
+    "CC0-1.0",
+    "Unicode-3.0",
+    "Unicode-DFS-2016",
+]
+unused-allowed-license = "allow"
+
+[bans]
+multiple-versions = "warn"
+wildcards = "deny"
+
+[sources]
+unknown-registry = "deny"
+unknown-git = "deny"
+allow-registry = ["https://github.com/rust-lang/crates.io-index"]
+allow-git = []

fuzz/Cargo.toml

@@ -24,3 +24,15 @@ name = "fuzz_roundtrip"
 path = "fuzz_targets/fuzz_roundtrip.rs"
 test = false
 doc = false
+
+[[bin]]
+name = "fuzz_small_dest"
+path = "fuzz_targets/fuzz_small_dest.rs"
+test = false
+doc = false
+
+[[bin]]
+name = "fuzz_encode"
+path = "fuzz_targets/fuzz_encode.rs"
+test = false
+doc = false

fuzz/fuzz_targets/fuzz_encode.rs

@@ -0,0 +1,33 @@
+//! Fuzz target: encode arbitrary input into buffers of varying sizes.
+//! Exercises encode error paths (buffer too small) and validates output properties.
+#![no_main]
+use libfuzzer_sys::fuzz_target;
+
+fuzz_target!(|data: &[u8]| {
+    if data.len() > 1024 {
+        return;
+    }
+
+    let max = ucobs::max_encoded_len(data.len());
+
+    // Encode into a correctly-sized buffer — must always succeed.
+    let mut buf = vec![0u8; max];
+    let n = ucobs::encode(data, &mut buf)
+        .expect("encode failed with max_encoded_len buffer");
+    let encoded = &buf[..n];
+
+    // Output must never contain 0x00.
+    assert!(!encoded.contains(&0x00), "encoded output contains 0x00");
+
+    // Output length must be bounded by max_encoded_len.
+    assert!(n <= max, "encoded length exceeds max_encoded_len");
+
+    // Output must be at least 1 byte (even empty input encodes to [0x01]).
+    assert!(n >= 1, "encoded length is zero");
+
+    // Encode into every possible undersized buffer — must return None, never panic.
+    for size in 0..n {
+        let mut small = vec![0u8; size];
+        assert_eq!(ucobs::encode(data, &mut small), None);
+    }
+});

fuzz/fuzz_targets/fuzz_small_dest.rs

@@ -0,0 +1,36 @@
+//! Fuzz target: encode and decode with undersized destination buffers.
+//! Must never panic — should return None when dest is too small.
+#![no_main]
+use libfuzzer_sys::fuzz_target;
+
+fuzz_target!(|data: &[u8]| {
+    if data.is_empty() {
+        return;
+    }
+
+    // Split fuzz input: first byte selects dest size, rest is payload.
+    let dest_size = data[0] as usize;
+    let payload = &data[1..];
+
+    // Encode with a potentially too-small buffer.
+    let mut enc_buf = vec![0u8; dest_size];
+    let enc_result = ucobs::encode(payload, &mut enc_buf);
+
+    if let Some(enc_len) = enc_result {
+        let encoded = &enc_buf[..enc_len];
+
+        // Encoded output must never contain 0x00.
+        assert!(!encoded.contains(&0x00), "encoded contains sentinel");
+
+        // Decode with a potentially too-small buffer.
+        let mut dec_buf = vec![0u8; dest_size];
+        let dec_result = ucobs::decode(encoded, &mut dec_buf);
+
+        if let Some(dec_len) = dec_result {
+            assert_eq!(
+                &dec_buf[..dec_len], payload,
+                "round-trip mismatch with small buffers"
+            );
+        }
+    }
+});

justfile

@@ -0,0 +1,87 @@
+# μCOBS development tasks
+
+# default: list all recipes
+default:
+    @just --list
+
+# run ALL quality gates (tests, clippy, fmt check, doc, deny, fuzz, miri, mutants)
+test: test-unit test-clippy test-fmt test-doc test-deny test-fuzz test-miri test-mutants
+
+# quick compile check (no tests)
+check:
+    cargo check
+
+# unit + integration + proptest tests (via nextest for parallelism)
+test-unit:
+    cargo nextest run
+    cargo nextest run --features legacy-msrv
+    cargo test --doc
+
+# clippy with all warnings as errors
+test-clippy:
+    cargo clippy -- -D warnings
+    cargo clippy --features legacy-msrv -- -D warnings
+
+# check formatting
+test-fmt:
+    cargo fmt -- --check
+
+# doc build (catches broken doc links)
+test-doc:
+    cargo doc --no-deps --all-features
+
+# license and advisory audit (requires cargo-deny)
+test-deny:
+    cargo deny check
+
+# fuzz all targets (60s each, requires nightly + cargo-fuzz)
+test-fuzz:
+    cd fuzz && cargo +nightly fuzz run fuzz_decode -- -max_total_time=60
+    cd fuzz && cargo +nightly fuzz run fuzz_encode -- -max_total_time=60
+    cd fuzz && cargo +nightly fuzz run fuzz_roundtrip -- -max_total_time=60
+    cd fuzz && cargo +nightly fuzz run fuzz_small_dest -- -max_total_time=60
+
+# run tests under miri (catches UB, requires nightly + miri component)
+# proptest is excluded — its file-persistence layer calls getcwd which miri does not support
+test-miri:
+    cargo +nightly miri test -- --skip proptests
+
+# mutation testing (requires cargo-mutants)
+test-mutants:
+    cargo mutants --no-shuffle
+
+# run iai-callgrind benchmarks (deterministic instruction counts, requires valgrind)
+bench: bench-iai bench-size
+
+# run iai-callgrind benchmarks (ucobs vs cobs vs corncobs)
+bench-iai:
+    cargo bench
+
+# measure compiled code size (encode + decode .text bytes)
+bench-size:
+    @cargo build --release 2>/dev/null
+    @echo ""
+    @echo "Code size (release, encode + decode .text):"
+    @nm -S target/release/libucobs.rlib 2>/dev/null \
+        | grep 'T _ZN5ucobs' \
+        | awk '{size=strtonum("0x"$2); name=$4; sub(/.*ucobs[0-9]*/, "", name); sub(/17h.*/, "", name); printf "  ucobs %-7s %4d B\n", name, size}'
+
+# run python cobs benchmarks (requires: pip install cobs)
+bench-python:
+    python3 bench_python.py
+
+# run all benchmarks (rust + python)
+bench-all: bench bench-python
+
+# format code
+fmt:
+    cargo fmt
+
+# build in release mode
+build:
+    cargo build --release
+
+# clean all build artifacts
+clean:
+    cargo clean
+    rm -rf fuzz/target fuzz/corpus fuzz/artifacts

mise.toml

@@ -0,0 +1,7 @@
+[tools]
+rust = "1.93"
+"cargo:cargo-deny" = "latest"
+"cargo:cargo-fuzz" = "latest"
+"cargo:cargo-mutants" = "latest"
+"cargo:cargo-nextest" = "latest"
+"cargo:iai-callgrind-runner" = "latest"

proptest-regressions/lib.txt

@@ -0,0 +1,10 @@
+# Seeds for failure cases proptest has generated in the past. It is
+# automatically read and these particular cases re-run before any
+# novel cases are generated.
+#
+# It is recommended to check this file in to source control so that
+# everyone who runs the test benefits from these saved cases.
+cc 00411472de4d6cdb296c04bc52fd533354bc25ba1765807da8b7994868fcb987 # shrinks to len = 1
+cc 81ebf688f1dd5ea2a735756a7fb703b92c2a2033057ae91886577c6a63763cf5 # shrinks to data = [0]
+cc 32d0d2d26e2be6ec66b432f1e9baee378777b1a00b959922cff7a5784292ec4b # shrinks to data = [0]
+cc fe447e0e58f4c0b21f6e1b38a90bd33578c968967c956964a06013a4df977d68 # shrinks to data = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0]