changelog: enrich 1.1.0-beta.1 and rc.5.2 entries with concrete mechanisms (version label SoT, per-extent AACS anchor, E7017/E7022 no-key split, 3-sector unit grid)

Use 'Semantics / reference' as the last-column header across all spec
tables, and break enum/option/note cells onto separate lines with <br>
for readability. Formatting only; no semantic changes.

CHANGELOG.md

@@ -1,6 +1,6 @@
 # Changelog
 
-## [1.0.0-rc.6] — UNRELEASED
+## [1.1.0-beta.1] — UNRELEASED
 
 ### Added
 
@@ -11,15 +11,46 @@
   all seven languages, and a Codes-page entry. Messages are source-agnostic
   ("key source", never a specific database).
 
+### Changed
+
+- keydb download/save moved out of the library into freemkv-keysources;
+  libfreemkv no longer has any keydb I/O (it already held no keys).
+
 ### Fixed
 
-- **Multi-extent AACS alignment in `dir://` extraction.** The decrypted
-  file-tree extractor anchored the AACS unit grid to the first extent only, so a
-  file spanning multiple clip extents could mis-align and corrupt. Each extent
-  now anchors on its own encrypted-region start (the same class as the rc.5.2
-  clip-anchor fix). Decryption math is unchanged.
-- **Distinct "no key" reasons.** A disc whose key needs a Volume ID that
-  couldn't be obtained now reports a distinct error from "no key at all".
+- **DVD rips now start on the movie, not the disc menu.** A VTS title VOB's
+  start sector was read from the IFO as a VTS-relative pointer but used as an
+  absolute disc address, so a DVD title's read extents began `ifo_lba` sectors
+  too early — the rip opened on the disc's menu / VMGI region and only drifted
+  into the feature minutes later (Silence of the Lambs, for example, showed
+  several minutes of the main menu before the movie). The title VOB is now
+  rebased to its absolute on-disc location, so the rip begins at the first frame
+  of the feature. Aspect ratio and chapter timing were already correct; only the
+  starting sector was wrong. (Covered by a new absolute-placement regression
+  test.)
+- **Container metadata correctness.** Unknown colorimetry now emits the CICP
+  "unspecified" code point (2) consistently across the MKV track and the FVI
+  sidecar (previously 0); PGS subtitle wipes use the NORMAL composition state
+  rather than a full epoch reset; and FVI source-byte offsets are written
+  within-sector per the format spec.
+- **Multi-extent AACS alignment in `dir://` extraction.** AACS encrypts in
+  aligned units of 3 sectors (6 KiB), and the decrypt-on-read gate accepts a read
+  only when its LBA is unit-aligned against a base. The `dir://` file-tree
+  extractor set that base once, to the file's first extent. A fragmented file
+  (Long-AD / continuation-ICB allocation) has later extents starting at arbitrary
+  LBAs whose distance from the first extent is generally not a multiple of 3
+  sectors, so the first read of every later extent failed the gate, returned a
+  decrypt error, and the whole extent was written as a zero-filled hole — even
+  though the sectors were readable. The unit base is now re-anchored per extent
+  (matching the mux read paths), so each extent gates on its own unit grid.
+  Decryption math is unchanged. Same class as the rc.5.2 clip-anchor fix.
+- **Distinct "no key" reasons.** When AACS key resolution has usable material
+  (device or processing keys) but cannot obtain the disc's Volume ID — needed to
+  derive the unit key — freemkv now reports a distinct "AACS Volume ID
+  unavailable" error (E7017) instead of collapsing it into the generic "no key"
+  error (E7022), which is now reserved for a genuine absence of any key material.
+  No key derivation or descramble logic changed — only the reason reported on a
+  resolution failure.
 - **autorip keydb writes go to the right path.** Auto-download, daily refresh,
   the "Update KEYDB" button, and the startup existence-check now resolve to the
   service's config path (matching where reads look); they previously used the
@@ -28,6 +59,22 @@
   files and parent directories around rename).
 - **Windows-reserved filenames** (`CON`, `NUL`, `COM1`…) inside a disc's file
   tree are safely renamed on extraction instead of aborting the walk.
+- **`--version` now matches the build stamped into MKVs.** The CLI's `--version`
+  string and the `MuxingApp` / `WritingApp` fields written into every MKV now
+  derive from a single libfreemkv constant — the package version plus the git
+  short hash (e.g. `freemkv 1.1.0-beta.1 (g835cc99)`). The muxer previously kept
+  its own copy of that string, so the two could drift; a binary and the files it
+  produces can no longer report different versions.
+- **DTS-HD Master Audio: a false core-sync inside the lossless extension no
+  longer splits an audio frame.** A byte pattern in the extension substream that
+  resembled the `0x7FFE8001` core sync word could truncate the lossless
+  extension and produce decode errors on the affected frames. The extension
+  substream is now sized exactly from its header, so that pattern is skipped as
+  data.
+- **TrueHD: decode timestamps no longer step backward.** In a case where the
+  source PES timing lagged the audio access-unit cadence, the muxed decode
+  timestamp could regress (non-monotonic-DTS warnings to the muxer); the running
+  timestamp is now clamped so it never goes backward.
 
 ### Tests
 
@@ -59,21 +106,16 @@
 
 ### Fixed
 
-- **Windows Explorer now reports the full 25 fps for interlaced SD-DVD.**
-  rc.5.1 added a `DefaultDecodedFieldDuration` (20 ms field) element to the
-  576i/480i track header on the theory that Windows derives fps from it. The
-  captured Silence-of-the-Lambs evidence proved the opposite: with
-  `FlagInterlaced=1` + `DefaultDuration=40 ms` + `DefaultDecodedFieldDuration=20 ms`,
-  Windows Explorer reported 12.5 fps (half) and MediaInfo flipped the track to
-  "Frame rate mode: Variable". MakeMKV's correct rip of the same disc OMITS
-  `DefaultDecodedFieldDuration`, keeps `FlagInterlaced=1` + `FieldOrder=TFF` +
-  full-frame `DefaultDuration` (40 ms), and Explorer shows the full 25 fps with
-  MediaInfo "Constant". The element is no longer written (`MkvTrack::video` now
-  passes `field_duration_ns == 0`); the only frame-rate signal tools trust,
-  `1/DefaultDuration` = 25 fps, is the full-frame value. Interlace signalling
-  (`FlagInterlaced=1`, `FieldOrder=TFF`) is retained, and MediaInfo still
-  reports "Interlaced / Top Field First" because it reads scan type from the
-  MPEG-2 elementary stream's picture coding extension, not the container flag.
+- **Reverted the rc.5.1 `DefaultDecodedFieldDuration` experiment for interlaced
+  SD-DVD.** rc.5.1 added a 20 ms `DefaultDecodedFieldDuration` field element to
+  the 576i/480i track header on the theory that Windows derives fps from it.
+  Captured evidence showed that element made Windows Explorer report 12.5 fps
+  (half) and MediaInfo flip the track to "Frame rate mode: Variable", while
+  MakeMKV's rip of the same disc omits it. The element is therefore no longer
+  written (`MkvTrack::video` now passes `field_duration_ns == 0`); the track
+  keeps `FlagInterlaced=1` + `FieldOrder=TFF` and the full-frame 40 ms
+  `DefaultDuration` (`1/DefaultDuration` = 25 fps), matching MakeMKV. How a given
+  player or shell handler chooses to display interlaced fps is not guaranteed.
 - **Correct AC-3 audio track selected on DVDs with non-standard sub-stream
   ordering.** freemkv assigned each declared audio stream a physical sub-stream
   by ordinal (`0x80+n`), assuming the IFO's first stream lives at `0x80`. On
@@ -83,11 +125,12 @@
   the disc — scanning every AC-3 frame and taking the maximum, so a brief 2.0
   logo bed at the feature head can't mask the real 5.1 — and routes each declared
   stream onto the sub-stream that genuinely matches.
-- **"Decryption failed" on large AACS Blu-ray titles fixed.** The unit-alignment
-  gate measured `lba % 3` against absolute disc LBA 0, but AACS aligned units are
-  anchored at each clip's encrypted-region start. A clip whose start is not
-  3-aligned had its readable units wrongly rejected — failing the feature/large
-  titles of some discs while short clips passed. The gate is now clip-anchored.
+- **"Decryption failed" on large AACS Blu-ray titles fixed.** AACS encrypts in
+  aligned units of 3 sectors (6 KiB); the unit-alignment gate measured `lba % 3`
+  against absolute disc LBA 0, but the unit grid is actually anchored at each
+  clip's encrypted-region start. A clip whose start is not 3-sector-aligned had
+  its readable units wrongly rejected — failing the feature/large titles of some
+  discs while short clips passed. The gate is now clip-anchored.
 - **Single-pass disc→MKV recovers marginal/transient sectors before failing.**
   The direct-to-MKV path now gives the drive its full ECC recovery budget on a
   bad sector (matching the multipass rip) instead of reporting a read failure a

Cargo.toml

@@ -22,7 +22,6 @@ sha1 = "0.10"
 sha2 = "0.10"
 aes = "0.8"
 cbc = "0.1"
-flate2 = "1"
 num-bigint = "0.4"
 num-traits = "0.2"
 num-integer = "0.1"

build.rs

@@ -1,4 +1,6 @@
 fn main() {
+    emit_git_suffix();
+
     let target = std::env::var("CARGO_CFG_TARGET_OS").unwrap_or_default();
     if target == "macos" {
         println!("cargo:rustc-link-lib=framework=IOKit");
@@ -48,3 +50,48 @@ fn main() {
         println!("cargo:rerun-if-changed=src/scsi/macos_shim.c");
     }
 }
+
+/// Bake the git short hash into the build as `GIT_SUFFIX` so any muxed MKV or
+/// FVI index is traceable to the exact source revision (e.g. ` (g835cc99)`).
+/// Empty when git or the repo is unavailable (e.g. a crates.io tarball build),
+/// leaving just the package version. Always emitted so `env!("GIT_SUFFIX")`
+/// resolves on every target.
+fn emit_git_suffix() {
+    // Version label for the muxing-app / FVI generator tag. `FREEMKV_BUILD_LABEL`
+    // overrides the Cargo package version when set (non-empty) — used to stamp a
+    // pre-release/test build without bumping Cargo.toml and disturbing the
+    // tag-pinned [patch] version matching. Unset → the package version.
+    let version = std::env::var("FREEMKV_BUILD_LABEL")
+        .ok()
+        .filter(|s| !s.trim().is_empty())
+        .or_else(|| std::env::var("CARGO_PKG_VERSION").ok())
+        .unwrap_or_default();
+    println!("cargo:rustc-env=FREEMKV_VERSION={version}");
+    println!("cargo:rerun-if-env-changed=FREEMKV_BUILD_LABEL");
+
+    let suffix = git_short_hash()
+        .map(|h| format!(" (g{h})"))
+        .unwrap_or_default();
+    println!("cargo:rustc-env=GIT_SUFFIX={suffix}");
+
+    // Re-run when HEAD (or the branch it points at) moves so the stamp stays
+    // current without a clean rebuild.
+    println!("cargo:rerun-if-changed=.git/HEAD");
+    if let Ok(head) = std::fs::read_to_string(".git/HEAD") {
+        if let Some(ref_path) = head.strip_prefix("ref: ") {
+            println!("cargo:rerun-if-changed=.git/{}", ref_path.trim());
+        }
+    }
+}
+
+fn git_short_hash() -> Option<String> {
+    let out = std::process::Command::new("git")
+        .args(["rev-parse", "--short=7", "HEAD"])
+        .output()
+        .ok()?;
+    if !out.status.success() {
+        return None;
+    }
+    let h = String::from_utf8(out.stdout).ok()?.trim().to_string();
+    if h.is_empty() { None } else { Some(h) }
+}

docs/FVI_FORMAT.md

@@ -1,7 +1,8 @@
 # FVI — Freemkv Video Index Format
 
-**Specification version:** 1.0 (DRAFT)
-**File extension:** `.fvi`  **Media type:** `application/vnd.freemkv.fvi+jsonl`
+**Specification version:** 1.0 (DRAFT)\
+**File extension:** `.fvi`\
+**Media type:** `application/vnd.freemkv.fvi+jsonl`\
 **Status:** Draft for review. This document is the normative reference for the FVI
 format; implementations and downstream tools cite it by section.
 
@@ -25,13 +26,6 @@ FVI is **not** a container, a codec, or a copy of the bitstream. It indexes; it
 never stores coded samples. It is the serialized form of an indexer's per-picture
 truth — carried from the demuxer, **never reconstructed** (§9).
 
-### 1.1 Relationship to prior art
-
-Legacy MPEG-only project-index formats from the AviSynth frameserving ecosystem
-solve a narrow version of (1) and (2) for MPEG-1/2 only, in a bespoke,
-single-tool text encoding. FVI generalizes that idea: codec-agnostic, JSON-based,
-provenance-native, and openly specified so any tool may read or write it.
-
 ## 2. Conformance
 
 The key words **MUST**, **MUST NOT**, **REQUIRED**, **SHALL**, **SHALL NOT**,
@@ -105,15 +99,15 @@ start of a new stream section.
 | `width`,`height` | integer | MUST | Coded luma dimensions in pixels. |
 | `dar` | `[int,int]` | SHOULD | Display aspect ratio as `[num,den]`. |
 | `frame_rate` | `[int,int]` | SHOULD | Nominal rate as exact rational `[num,den]` (e.g. `[24000,1001]`). |
-| `scan` | string | MUST | `"progressive"` \| `"interlaced"` \| `"mbaff"`. |
-| `colour` | object | SHOULD | CICP per ITU-T H.273: `primaries`,`transfer`,`matrix` (integer CICP codes or registered names), `range` (`"limited"`\|`"full"`). HDR: `mastering_display`, `max_cll`, `max_fall` per ITU-T H.273 / SMPTE ST 2086. |
+| `scan` | string | MUST | `"progressive"`<br>`"interlaced"`<br>`"mbaff"` |
+| `colour` | object | SHOULD | CICP per ITU-T H.273: `primaries`, `transfer`, `matrix` (integer CICP codes or registered names)<br>`range`: `"limited"` \| `"full"`<br>HDR: `mastering_display`, `max_cll`, `max_fall` per ITU-T H.273 / SMPTE ST 2086. |
 | `language` | string | MAY | BCP 47 tag, if known. |
 
 ### 6.2 `source` object
 
-| Member | JSON type | Req | Semantics |
+| Member | JSON type | Req | Semantics / reference |
 |---|---|---|---|
-| `medium` | string | MUST | `"disc"` \| `"iso"` \| `"file"` \| `"stream"`. |
+| `medium` | string | MUST | `"disc"`<br>`"iso"`<br>`"file"`<br>`"stream"` |
 | `path` | string | MAY | Source path/label. |
 | `title` | integer | MAY | Title/program number. |
 | `playlist` | string | MAY | Playlist/PGC identifier. |
@@ -128,13 +122,13 @@ One JSON object per coded picture, in coded order.
 |---|---|---|---|
 | `n` | integer | MUST | Coded-order index, 0-based, contiguous. |
 | `src` | object | MUST | Provenance: `{ "file": int?, "sector": uint, "byte": uint }` — the offset of this AU's **first byte** in the source (§9). MUST be carried from demux, never reconstructed. |
-| `type` | string | MUST | Coding type: `"I"` \| `"P"` \| `"B"` (ISO/IEC 13818-2 §6.3.9; H.264/H.265 slice types collapsed to frame type). |
-| `key` | boolean | MUST | `true` iff this picture is an intra (I) picture / parser-flagged decode-restart point (IDR / IRAP / I-picture). MPEG-2 open-GOP clean-RAP precision (`closed_gop`) is not currently distinguished — see note below. |
-| `gop` | boolean | SHOULD | `true` iff this picture begins a GOP / coded video sequence. Omitted when the implementation does not carry a distinct GOP-boundary signal. |
+| `type` | string | MUST | Coding type:<br>`"I"`<br>`"P"`<br>`"B"`<br>_ISO/IEC 13818-2 §6.3.9; H.264/H.265 slice types collapsed to frame type._ |
+| `key` | boolean | MUST | `true` iff this picture is an intra (I) picture / parser-flagged decode-restart point (IDR / IRAP / I-picture).<br>_MPEG-2 open-GOP clean-RAP precision (`closed_gop`) is not currently distinguished — see note below._ |
+| `gop` | boolean | SHOULD | `true` iff this picture begins a GOP / coded video sequence.<br>_Omitted when the implementation does not carry a distinct GOP-boundary signal._ |
 | `pts` | integer\|null | SHOULD | Presentation timestamp in `timescale` ticks; `null` if unknown. |
 | `dts` | integer\|null | MAY | Decode timestamp in `timescale` ticks. |
 | `size` | integer | MAY | AU length in bytes; enables byte-range extraction with `src`. |
-| `recovered` | boolean | MAY | `true` iff any byte of this AU came from a retried/marginal read (§9.1). Default `false`. |
+| `recovered` | boolean | MAY | `true` iff any byte of this AU came from a retried/marginal read (§9.1).<br>_Default `false`._ |
 | codec ext | object | MAY | Codec-specific members under the codec's namespace (§8). |
 
 The `type` and `key` members are **codec-agnostic** and MUST be populated for
@@ -162,11 +156,11 @@ per-picture coding accessors (MPEG-2: ISO/IEC 13818-2 §6.3.10). Emitted as
 top-level members of the record, and ONLY when the codec actually measured the
 signal — an OPTIONAL member that is omitted (not defaulted) when unknown:
 
-| Member | JSON type | Req | Semantics |
+| Member | JSON type | Req | Semantics / reference |
 |---|---|---|---|
-| `field_order` | string | MAY | Display field order: `"tff"` (top field first) \| `"bff"` (bottom field first) \| `"progressive"` (no field order applies). Omitted when the codec did not signal it. |
-| `progressive` | boolean | MAY | `true` iff the picture is progressive. Omitted when the codec did not signal it. |
-| `nb_fields` | integer | MAY | Number of displayed field periods this picture occupies (the soft-telecine / 2:3 pulldown basis): `1` for a single field picture, `2` for a normal frame, `3`/`4`/`6` for `repeat_first_field` pulldown per §6.3.10. |
+| `field_order` | string | MAY | Display field order:<br>`"tff"` — top field first<br>`"bff"` — bottom field first<br>`"progressive"` — no field order applies<br>_Omitted when the codec did not signal it._ |
+| `progressive` | boolean | MAY | `true` iff the picture is progressive.<br>_Omitted when the codec did not signal it._ |
+| `nb_fields` | integer | MAY | Number of displayed field periods this picture occupies (the soft-telecine / 2:3 pulldown basis):<br>`1` for a single field picture<br>`2` for a normal frame<br>`3`/`4`/`6` for `repeat_first_field` pulldown per §6.3.10 |
 
 Codecs that carry only a coding type (e.g. H.264 / HEVC / VC-1 through this
 pipeline) omit `field_order` and `progressive` rather than guessing a default.
@@ -179,8 +173,21 @@ registered set per codec profile (e.g. MPEG-2 §7.1), or (b) placed under an
 `ext` object keyed by codec id for richer/optional data:
 
 ```json
-{"n":42,"type":"P","key":false,"src":{"sector":17,"byte":924},
- "ext":{"hevc":{"temporal_id":0,"nal_type":1}}}
+{
+  "n": 42,
+  "type": "P",
+  "key": false,
+  "src": {
+    "sector": 17,
+    "byte": 924
+  },
+  "ext": {
+    "hevc": {
+      "temporal_id": 0,
+      "nal_type": 1
+    }
+  }
+}
 ```
 
 New codecs and members are added through Appendix B (codec registry) without a
@@ -238,13 +245,13 @@ Header:
 {
   "$schema": "https://json-schema.org/draft/2020-12/schema",
   "type": "object",
-  "required": ["format","fvi_version","stream","source","timescale"],
+  "required": ["format", "fvi_version", "stream", "source", "timescale"],
   "properties": {
-    "format": {"const": "freemkv/video-index"},
-    "fvi_version": {"type": "integer", "minimum": 1},
-    "timescale": {"type": "integer", "minimum": 1},
-    "stream": {"type": "object", "required": ["codec","width","height","scan"]},
-    "source": {"type": "object", "required": ["medium"]}
+    "format": { "const": "freemkv/video-index" },
+    "fvi_version": { "type": "integer", "minimum": 1 },
+    "timescale": { "type": "integer", "minimum": 1 },
+    "stream": { "type": "object", "required": ["codec", "width", "height", "scan"] },
+    "source": { "type": "object", "required": ["medium"] }
   }
 }
 ```
@@ -255,15 +262,20 @@ Record:
 {
   "$schema": "https://json-schema.org/draft/2020-12/schema",
   "type": "object",
-  "required": ["n","src","type","key"],
+  "required": ["n", "src", "type", "key"],
   "properties": {
-    "n": {"type": "integer", "minimum": 0},
-    "type": {"enum": ["I","P","B"]},
-    "key": {"type": "boolean"},
-    "src": {"type":"object","required":["sector","byte"],
-            "properties":{"file":{"type":"integer"},
-                          "sector":{"type":"integer","minimum":0},
-                          "byte":{"type":"integer","minimum":0}}}
+    "n": { "type": "integer", "minimum": 0 },
+    "type": { "enum": ["I", "P", "B"] },
+    "key": { "type": "boolean" },
+    "src": {
+      "type": "object",
+      "required": ["sector", "byte"],
+      "properties": {
+        "file": { "type": "integer" },
+        "sector": { "type": "integer", "minimum": 0 },
+        "byte": { "type": "integer", "minimum": 0 }
+      }
+    }
   }
 }
 ```

src/aacs/boil.rs

@@ -15,11 +15,17 @@
 //!
 //! ```text
 //!   mk_from_dk(device_keys, mkb, vid)  →  MediaKey   (Km)
+//!   mk_from_pk(processing_keys, mkb)   →  MediaKey   (Km)
 //!   vuk_from_mk(MediaKey, Vid)         →  Vuk        (= AES-G(Km, VID))
 //!   uk_from_vuk(Vuk, enc_title_keys)   →  [UnitKey]  (decrypt_unit_key each)
 //! ```
+//!
+//! `mk_from_dk` and `mk_from_pk` are two entry points to the SAME Media Key:
+//! the device-key path walks the MKB's Media-Key-Variant chain, the
+//! processing-key path walks the MKB's Subset-Difference cvalue tables. Neither
+//! needs a VID (the VID enters at `vuk_from_mk`).
 
-use super::keys::{decrypt_unit_key, derive_vuk};
+use super::keys::{decrypt_unit_key, derive_media_key_from_pk, derive_vuk};
 use super::types::DeviceKey;
 use super::variants::{KEY_CORRECTION_DATA_PLACEHOLDER, derive_media_key_variant, walk_mkb};
 
@@ -111,6 +117,31 @@ pub fn mk_from_dk(
     }
 }
 
+/// Derive the Media Key (Km) from one or more Processing Keys and the disc MKB.
+///
+/// Wraps [`derive_media_key_from_pk`] — the Subset-Difference PK→MK walk: each
+/// processing key is validated (and tree-walked) against the MKB's cvalue tables
+/// (records `0x04`/`0x05`) until one yields the Media Key whose verify record
+/// (`0x81`/`0x86`) matches. Unlike [`mk_from_dk`] this path is reachable for
+/// real discs — a leaked/precomputed AACS Processing Key in the keydb resolves
+/// the Media Key directly. No VID is involved at this step; the VID enters at
+/// [`vuk_from_mk`].
+///
+/// Returns [`Error::AacsMkUnavailable`] (E7018) when no processing key resolves
+/// the MKB — the same terminal error as [`mk_from_dk`]; no numeric distinction
+/// is load-bearing at this boundary.
+///
+/// [`Error::AacsMkUnavailable`]: crate::error::Error::AacsMkUnavailable
+pub fn mk_from_pk(
+    processing_keys: &[[u8; 16]],
+    mkb: &[u8],
+) -> Result<MediaKey, crate::error::Error> {
+    match derive_media_key_from_pk(mkb, processing_keys) {
+        Some(km) => Ok(MediaKey(km)),
+        None => Err(crate::error::Error::AacsMkUnavailable),
+    }
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -197,4 +228,81 @@ mod tests {
         let e2 = mk_from_dk(&[dk], &mkb, Vid([0x09; 16]));
         assert!(matches!(e2, Err(crate::error::Error::AacsMkUnavailable)));
     }
+
+    /// Build a 4-byte MKB record header (type + 3-byte big-endian total length,
+    /// header included) and append `body`. Mirrors the MKB record framing the
+    /// parser expects; no crypto.
+    fn mkb_record(rec_type: u8, body: &[u8]) -> Vec<u8> {
+        let total = 4 + body.len();
+        let mut rec = Vec::with_capacity(total);
+        rec.push(rec_type);
+        rec.push(((total >> 16) & 0xFF) as u8);
+        rec.push(((total >> 8) & 0xFF) as u8);
+        rec.push((total & 0xFF) as u8);
+        rec.extend_from_slice(body);
+        rec
+    }
+
+    /// `mk_from_pk` resolves a planted Processing Key against a synthetic MKB and
+    /// drives the FULL boil chain PK → MK → VUK → UK. The MKB is built with the
+    /// same (pk, cv, mk_dv, uv) construction the production SD walk validates, so
+    /// this proves a PK entry yields real Unit Keys — not just an `Ok`.
+    #[test]
+    fn mk_from_pk_drives_full_chain_to_uks() {
+        let pk: [u8; 16] = [
+            0x11, 0x22, 0x33, 0x44, 0x55, 0x66, 0x77, 0x88, 0x99, 0xAA, 0xBB, 0xCC, 0xDD, 0xEE,
+            0xFF, 0x00,
+        ];
+        let mk: [u8; 16] = [
+            0xA0, 0xA1, 0xA2, 0xA3, 0xA4, 0xA5, 0xA6, 0xA7, 0xA8, 0xA9, 0xAA, 0xAB, 0xAC, 0xAD,
+            0xAE, 0xAF,
+        ];
+        let uv: [u8; 4] = [0x00, 0x00, 0x04, 0x00];
+
+        // cv = AES-E(pk, mk_raw), where mk_raw is mk with the last-4-bytes-uv XOR
+        // pre-undone, so the validate step XORs uv back in and recovers mk.
+        let mut mk_raw = mk;
+        for a in 0..4 {
+            mk_raw[12 + a] ^= uv[a];
+        }
+        let cv = aes_ecb_encrypt(&pk, &mk_raw);
+
+        // mk_dv = AES-E(mk, magic||pad): AES-D(mk, mk_dv) starts with the AACS
+        // verify sentinel.
+        let mut vd = [0x11u8; 16];
+        vd[..8].copy_from_slice(&[0x01, 0x23, 0x45, 0x67, 0x89, 0xAB, 0xCD, 0xEF]);
+        let mk_dv = aes_ecb_encrypt(&mk, &vd);
+
+        // Synthetic MKB: type/version (0x10), verify record (0x86 = mk_dv),
+        // one-entry SD index (0x04 = [u_mask_shift=0][uv]), one-entry cvalue
+        // table (0x05 = cv).
+        let mut sd = vec![0u8];
+        sd.extend_from_slice(&uv);
+        let mut mkb = Vec::new();
+        mkb.extend_from_slice(&mkb_record(0x10, &[0, 0, 0, 0x20, 0, 0, 0, 0x52]));
+        mkb.extend_from_slice(&mkb_record(0x86, &mk_dv));
+        mkb.extend_from_slice(&mkb_record(0x04, &sd));
+        mkb.extend_from_slice(&mkb_record(0x05, &cv));
+
+        // PK → MK.
+        let got_mk = mk_from_pk(std::slice::from_ref(&pk), &mkb).expect("planted PK resolves MK");
+        assert_eq!(got_mk, MediaKey(mk), "mk_from_pk recovers the planted MK");
+
+        // MK → VUK → UK over an encrypted title key.
+        let vid = Vid([0x42u8; 16]);
+        let plain_uk = [0x7Eu8; 16];
+        let vuk = vuk_from_mk(got_mk, vid);
+        let enc = aes_ecb_encrypt(&vuk.0, &plain_uk);
+        let uks = uk_from_vuk(vuk, std::slice::from_ref(&enc));
+        assert_eq!(uks.len(), 1);
+        assert_eq!(uks[0].key, plain_uk, "PK chain recovers the title key");
+
+        // A corrupt PK resolves nothing.
+        let mut bad = pk;
+        bad[0] ^= 0xFF;
+        assert!(matches!(
+            mk_from_pk(std::slice::from_ref(&bad), &mkb),
+            Err(crate::error::Error::AacsMkUnavailable)
+        ));
+    }
 }

src/aacs/mod.rs

@@ -24,7 +24,7 @@ pub mod types;
 pub mod variants;
 
 // Boil-down derivation primitives (thin newtypes + wrappers over the crypto).
-pub use boil::{MediaKey, UnitKey, Vid, Vuk, mk_from_dk, uk_from_vuk, vuk_from_mk};
+pub use boil::{MediaKey, UnitKey, Vid, Vuk, mk_from_dk, mk_from_pk, uk_from_vuk, vuk_from_mk};
 // Structured, English-free resolution trace.
 pub use trace::{KeyNode, KeyOutcome, KeyStep, ResolutionTrace, UnlockOutcome, UnlockStep};
 
@@ -35,6 +35,10 @@ pub use decrypt::{
     decrypt_unit_full, decrypt_unit_try_keys, is_aacs_scrambled, is_unit_aligned, ts_packet_total,
     ts_sync_count, unit_key_validates,
 };
+// `probe` is a reproduction-harness helper (see keys.rs), not part of the
+// documented 1.0 surface; keep it reachable but off the rendered docs so we
+// don't commit semver stability to test primitives.
+#[doc(hidden)]
 pub use keys::probe;
 pub use keys::{
     AacsVersion, ContentCert, MKB_20_CATEGORY_C, MKB_21_CATEGORY_C, MKB_TYPE_3_RECORDABLE,
@@ -97,5 +101,6 @@ mod tests {
         let _ = mkb_content_len(&[]);
         let _ = is_variant_mkb(&walk_mkb(&[]));
         let _ = disc_hash_hex(&disc_hash(b"x"));
+        let _ = mk_from_pk(&[[0u8; 16]], &[]);
     }
 }

src/clpi.rs

@@ -6,7 +6,7 @@
 //!
 //! Reference: https://github.com/lw/BluRay/wiki/CLPI
 
-use crate::consts::{BD_SOURCE_PACKET_BYTES, SECTOR_BYTES};
+use crate::consts::{BD_SOURCE_PACKET_BYTES, SECTOR_BYTES_U64};
 use crate::disc::Extent;
 use crate::error::{Error, Result};
 
@@ -175,8 +175,8 @@ impl ClipInfo {
         // a sub-sector-aligned range still spans every sector it touches.
         let start_byte = start_spn as u64 * BD_SOURCE_PACKET_BYTES as u64;
         let end_byte = end_spn as u64 * BD_SOURCE_PACKET_BYTES as u64;
-        let start_sector = (start_byte / SECTOR_BYTES as u64) as u32;
-        let end_sector = end_byte.div_ceil(SECTOR_BYTES as u64) as u32;
+        let start_sector = (start_byte / SECTOR_BYTES_U64) as u32;
+        let end_sector = end_byte.div_ceil(SECTOR_BYTES_U64) as u32;
 
         vec![Extent {
             start_lba: start_sector, // relative to m2ts file start
@@ -261,6 +261,7 @@ pub fn parse(data: &[u8]) -> Result<ClipInfo> {
 /// errors because the EP map is the primary CLPI output, and a corrupt
 /// program_info shouldn't break sector-range lookups.
 fn parse_program_info(data: &[u8]) -> Vec<ClpiStream> {
+    use crate::consts::coding_type as c;
     let mut out = Vec::new();
     if data.len() < 6 {
         return out;
@@ -299,16 +300,15 @@ fn parse_program_info(data: &[u8]) -> Vec<ClpiStream> {
             let mut language = String::new();
 
             match coding_type {
-                // Video — MPEG-2 (0x02), H.264 (0x1B), HEVC (0x24)
-                0x02 | 0x1B | 0x24 => {
+                // Video — MPEG-2, H.264, HEVC
+                c::MPEG2_VIDEO | c::H264 | c::HEVC => {
                     if sci.len() >= 2 {
                         video_format = (sci[1] >> 4) & 0x0F;
                         video_rate = sci[1] & 0x0F;
                     }
                 }
-                // Primary audio — LPCM(0x80), AC-3(0x81), DTS(0x82),
-                // TrueHD(0x83), AC-3+(0x84), DTS-HD(0x85), DTS-HD MA(0x86)
-                0x80..=0x86 => {
+                // Primary audio — LPCM, AC-3, DTS, TrueHD, AC-3+, DTS-HD HR, DTS-HD MA
+                c::LPCM..=c::DTS_HD_MA => {
                     if sci.len() >= 2 {
                         audio_format = (sci[1] >> 4) & 0x0F;
                         audio_rate = sci[1] & 0x0F;
@@ -317,8 +317,8 @@ fn parse_program_info(data: &[u8]) -> Vec<ClpiStream> {
                         language = String::from_utf8_lossy(&sci[2..5]).to_string();
                     }
                 }
-                // Secondary audio (0xA1 AC-3+, 0xA2 DTS-HD)
-                0xA1 | 0xA2 => {
+                // Secondary audio (AC-3+ secondary, DTS-HD secondary)
+                c::AC3_PLUS_SECONDARY | c::DTS_HD_SECONDARY => {
                     if sci.len() >= 2 {
                         audio_format = (sci[1] >> 4) & 0x0F;
                         audio_rate = sci[1] & 0x0F;
@@ -327,8 +327,8 @@ fn parse_program_info(data: &[u8]) -> Vec<ClpiStream> {
                         language = String::from_utf8_lossy(&sci[2..5]).to_string();
                     }
                 }
-                // PG (0x90), IG (0x91): coding_type + 3-byte language [+ char_code for PG]
-                0x90 | 0x91 => {
+                // PG, IG: coding_type + 3-byte language [+ char_code for PG]
+                c::PG | c::IG => {
                     if sci.len() >= 4 {
                         language = String::from_utf8_lossy(&sci[1..4]).to_string();
                     }
@@ -1125,8 +1125,8 @@ mod tests {
         let end_spn = big_spn as u64;
         let start_byte = start_spn * BD_SOURCE_PACKET_BYTES as u64;
         let end_byte = end_spn * BD_SOURCE_PACKET_BYTES as u64;
-        let start_sector = (start_byte / SECTOR_BYTES as u64) as u32;
-        let end_sector = end_byte.div_ceil(SECTOR_BYTES as u64) as u32;
+        let start_sector = (start_byte / SECTOR_BYTES_U64) as u32;
+        let end_sector = end_byte.div_ceil(SECTOR_BYTES_U64) as u32;
         assert_eq!(extents[0].start_lba, start_sector);
         assert_eq!(extents[0].sector_count, end_sector - start_sector);
         // Concretely: 0x20000 × 192 / 2048 = 12288 sectors.

src/consts.rs

@@ -8,8 +8,18 @@
 
 /// Bytes per logical sector on every optical medium freemkv reads
 /// (Blu-ray, DVD-Video, CD-ROM Mode 1). Universal — hence unprefixed.
+///
+/// `usize` because its dominant use is buffer sizing and slice indexing, where
+/// Rust *requires* `usize` (`vec![0u8; SECTOR_BYTES]`, `buf.len() < SECTOR_BYTES`).
+/// For byte-offset / capacity arithmetic — which is `u64` because a disc can
+/// exceed 4 GiB — use [`SECTOR_BYTES_U64`] instead of casting at each site.
 pub const SECTOR_BYTES: usize = 2048;
 
+/// [`SECTOR_BYTES`] as `u64`, for byte-offset and capacity arithmetic. The
+/// single `usize → u64` boundary cast lives here, once, so offset math across
+/// the workspace reads as `sectors * SECTOR_BYTES_U64` with no per-site cast.
+pub const SECTOR_BYTES_U64: u64 = SECTOR_BYTES as u64;
+
 /// Bytes per MPEG-2 transport-stream packet. Common to all MPEG-TS, not just
 /// Blu-ray — prefixed by the format, not a disc type.
 pub const TS_PACKET_BYTES: usize = 188;
@@ -31,3 +41,81 @@ pub const TS_PAYLOAD_BYTES: usize = TS_PACKET_BYTES - TS_HEADER_BYTES;
 /// prefixed with the [`BD_TIMESTAMP_PREFIX_BYTES`] arrival-timestamp header.
 /// A BDAV/M2TS construct only — DVD VOBs have no source packets — hence `BD_`.
 pub const BD_SOURCE_PACKET_BYTES: usize = TS_PACKET_BYTES + BD_TIMESTAMP_PREFIX_BYTES;
+
+/// Elementary-stream coding-type codes — the single source of truth for the
+/// byte that identifies a stream's codec.
+///
+/// This is one registry used in two places that share the same value space:
+/// the MPEG-TS PMT `stream_type` (ISO/IEC 13818-1 Table 2-34) and the Blu-ray
+/// STN/CLPI `stream_coding_type` (BD-ROM Part 3). The standardized video codes
+/// (`0x02`, `0x1B`, `0x24`) are ISO assignments (ISO/IEC 13818-1 Table 2-34);
+/// `0xEA` (VC-1) is a BD-ROM convention in the ISO user-private range. The
+/// `0x80..=0xA2` audio/graphics codes also sit in the user-private range and follow the
+/// Blu-ray Disc Association / ATSC A/52 convention. Because every consumer
+/// reads or writes this single byte, the family is unprefixed — the scope is
+/// "any elementary stream freemkv parses or muxes".
+///
+/// Each constant is `u8`: the spec defines an 8-bit field and the code compares
+/// it directly against a byte read from the buffer, so no casts are needed.
+pub mod coding_type {
+    /// MPEG-2 video (ISO/IEC 13818-1 Table 2-34).
+    pub const MPEG2_VIDEO: u8 = 0x02;
+    /// H.264 / AVC video (ISO/IEC 13818-1 Table 2-34).
+    pub const H264: u8 = 0x1B;
+    /// HEVC / H.265 video (ISO/IEC 13818-1 Table 2-34, 2015 amendment).
+    pub const HEVC: u8 = 0x24;
+    /// SMPTE VC-1 video (BD-ROM convention, ISO user-private range).
+    pub const VC1: u8 = 0xEA;
+
+    /// LPCM audio (BD-ROM convention).
+    pub const LPCM: u8 = 0x80;
+    /// Dolby Digital (AC-3) audio (BD-ROM / ATSC A/52 convention).
+    pub const AC3: u8 = 0x81;
+    /// DTS audio (BD-ROM convention).
+    pub const DTS: u8 = 0x82;
+    /// Dolby TrueHD audio (BD-ROM convention).
+    pub const TRUEHD: u8 = 0x83;
+    /// Dolby Digital Plus (E-AC-3 / AC-3+) audio (BD-ROM convention).
+    pub const AC3_PLUS: u8 = 0x84;
+    /// DTS-HD High Resolution audio (BD-ROM Part 3-1).
+    pub const DTS_HD_HR: u8 = 0x85;
+    /// DTS-HD Master Audio (lossless) (BD-ROM Part 3-1).
+    pub const DTS_HD_MA: u8 = 0x86;
+
+    /// Presentation Graphics — PG subtitle stream (BD-ROM HDMV).
+    pub const PG: u8 = 0x90;
+    /// Interactive Graphics — IG / BD-J menu overlay, NOT a subtitle (BD-ROM HDMV).
+    pub const IG: u8 = 0x91;
+    /// Text subtitle stream (BD-ROM HDMV).
+    pub const TEXT_SUBTITLE: u8 = 0x92;
+
+    /// Secondary Dolby Digital Plus audio (BD-ROM convention).
+    pub const AC3_PLUS_SECONDARY: u8 = 0xA1;
+    /// Secondary DTS-HD audio (lossless MA, not lossy HR) (BD-ROM convention).
+    pub const DTS_HD_SECONDARY: u8 = 0xA2;
+}
+
+/// MPEG PES `stream_id` codes — the byte after the `00 00 01` start-code prefix
+/// that identifies an elementary stream's role in a PES packet (ISO/IEC
+/// 13818-1 Table 2-22). Shared by the program-stream demuxer and the TS/M2TS
+/// muxers, so defined here once. Each is `u8` (matches the byte on the wire).
+pub mod pes_stream_id {
+    /// Video stream (`110x xxxx`; freemkv emits the base id `0xE0`).
+    pub const VIDEO: u8 = 0xE0;
+    /// private_stream_1 — AC-3 / DTS / LPCM / PGS subtitle payloads.
+    pub const PRIVATE_STREAM_1: u8 = 0xBD;
+    /// padding_stream — stuffing bytes only, no payload to demux.
+    pub const PADDING_STREAM: u8 = 0xBE;
+    /// private_stream_2 — DVD navigation (PCI/DSI); carries no muxable ES.
+    pub const PRIVATE_STREAM_2: u8 = 0xBF;
+
+    /// Highest video stream_id — the `110x xxxx` video range tops out at 0xEF.
+    pub const VIDEO_MAX: u8 = 0xEF;
+
+    /// Inclusive range of every PES `stream_id` that carries demuxable payload:
+    /// [`PRIVATE_STREAM_1`] (0xBD) through [`VIDEO_MAX`] (0xEF) — i.e. private
+    /// stream 1/2, padding, MPEG audio (0xC0-0xDF) and video (0xE0-0xEF). The
+    /// pack (0xBA), system-header (0xBB) and program-end (0xB9) codes sit below
+    /// this range and are deliberately excluded: they're structural, not ES.
+    pub const PAYLOAD_RANGE: core::ops::RangeInclusive<u8> = PRIVATE_STREAM_1..=VIDEO_MAX;
+}

src/css/stevenson.rs

@@ -402,7 +402,7 @@ mod tests {
 
     #[test]
     fn crack_unscrambled_returns_none() {
-        let sector = vec![0u8; 2048];
+        let sector = vec![0u8; SECTOR_BYTES];
         assert!(crack_title_key(&sector).is_none());
     }
 
@@ -414,7 +414,7 @@ mod tests {
 
     #[test]
     fn recover_needs_min_plain() {
-        let sector = vec![0u8; 2048];
+        let sector = vec![0u8; SECTOR_BYTES];
         let short_plain = [0u8; 4];
         assert!(recover_title_key(&sector, &short_plain).is_none());
     }

src/decrypt.rs

@@ -373,9 +373,16 @@ pub fn decrypt_sectors(
                     continue;
                 }
                 let crib = css::stevenson::attack_crib(chunk);
-                let original: Option<Vec<u8>> = crib.as_ref().map(|_| chunk.to_vec());
+                // Snapshot the ciphertext into a stack buffer (chunk is exactly
+                // 2048 here — guaranteed by the `< 2048` continue above) only
+                // when there's a crib to validate against, so the common
+                // cache-hit path costs no per-sector heap allocation.
+                let mut original = [0u8; 2048];
+                if crib.is_some() {
+                    original.copy_from_slice(chunk);
+                }
                 css::lfsr::descramble_sector(title_key, chunk);
-                if let (Some(crib), Some(original)) = (crib, original) {
+                if let Some(crib) = crib {
                     if chunk[0x80..0x80 + 10] != crib[..] {
                         // Cached key is stale for this region — restore the
                         // ciphertext and crack this sector's own key.

src/disc/dvd.rs

@@ -106,7 +106,7 @@ impl Disc {
                 })
                 .collect();
 
-            for dvd_title in &ts.titles {
+            for (vts_title_idx, dvd_title) in ts.titles.iter().enumerate() {
                 title_number += 1;
 
                 // Diagnostic dump (--log-level 3): per-cell category table +
@@ -114,13 +114,32 @@ impl Disc {
                 // per-cell IFO detail. No-op unless freemkv::diag is enabled.
                 crate::diag::dump_dvd_cells(ts.vts_number, title_number, dvd_title);
 
-                // Bug-4 leading-cell filter: drop any leading scene-index /
-                // interleaved-angle sub-block cells so the feature starts at the
-                // movie. Conservative — `feature_start_cell` only ever skips a
-                // prefix of secondary-block cells and never truncates a normal
-                // feature (category 0x00 on cell 0 → no-op). See
-                // `ifo::DvdTitle::feature_start_cell`.
-                let feature_start = dvd_title.feature_start_cell();
+                // Feature start cell. Prefer the DVD nav-VM resolver, which
+                // PARKED (#40, menu-at-start playback). The "menu at the start"
+                // symptom (e.g. SOTL) was a sector-mapping fault — the absolute
+                // VOB rebase in `ifo::parse_vts` (`vob_start_sector =
+                // file_start_lba + vtstt_vobs`) — NOT a navigation problem, so
+                // feature-start resolution is unnecessary for correct rips. The
+                // nav resolver + verified VM decoder (`dvdnav`) are kept compiled
+                // but deliberately bypassed; flip `USE_NAV_RESOLVER` to re-enable
+                // once the nav executor is finished. The fallback is the
+                // structural leading-cell filter (`feature_start_cell`), which
+                // drops leading scene-index / interleaved-angle sub-block cells
+                // and is a no-op for a normal feature (category 0x00 on cell 0).
+                // See `dvdnav::resolve_feature_start`.
+                const USE_NAV_RESOLVER: bool = false;
+                let feature_start = if USE_NAV_RESOLVER {
+                    crate::dvdnav::resolve_feature_start(
+                        reader,
+                        udf_fs,
+                        ts.vts_number as u16,
+                        (vts_title_idx + 1) as u16,
+                    )
+                    .unwrap_or_else(|| dvd_title.feature_start_cell())
+                } else {
+                    dvd_title.feature_start_cell()
+                }
+                .min(dvd_title.cells.len());
                 let dropped_secs: f64 = dvd_title.cells[..feature_start]
                     .iter()
                     .map(|c| c.duration_secs)
@@ -136,9 +155,9 @@ impl Disc {
                     );
                 }
 
-                // Build extents from cell sector ranges (absolute = vob_start + cell offset)
-                let extents: Vec<Extent> = dvd_title
-                    .feature_cells()
+                // Build extents from cell sector ranges (absolute = vob_start + cell offset),
+                // starting at the resolved feature-start cell.
+                let extents: Vec<Extent> = dvd_title.cells[feature_start..]
                     .iter()
                     .map(|cell| {
                         let start = ts.vob_start_sector.saturating_add(cell.first_sector);
@@ -553,8 +572,10 @@ mod tests {
         assert_eq!(titles.len(), 1);
         let t = &titles[0];
         assert_eq!(t.extents.len(), 1);
-        // absolute start = vob_start(1000) + first_sector(10) = 1010.
-        assert_eq!(t.extents[0].start_lba, 1010);
+        // absolute start = ifo_lba + vtstt_vobs(1000) + first_sector(10).
+        // The IFO file sits at PART_START(3000) + data_lba(6000) = 9000, so
+        // 9000 + 1000 + 10 = 10010.
+        assert_eq!(t.extents[0].start_lba, 10010);
         // inclusive: 109 - 10 + 1 = 100 sectors.
         assert_eq!(t.extents[0].sector_count, 100);
         // DVD sector = 2048 bytes.
@@ -607,12 +628,85 @@ mod tests {
         assert_eq!(titles.len(), 1);
         let t = &titles[0];
         assert_eq!(t.extents.len(), 1);
-        // Title VOBS (3640) + cell first_sector (0) = 3640 — NOT the menu 44.
+        // ifo_lba(9000) + vtstt_vobs(3640) + first_sector(0) = 12640 — built
+        // from the Title VOBS (0xC4), NOT the menu VOBS (0xC0). The IFO file is
+        // at PART_START(3000) + data_lba(6000) = 9000.
         assert_eq!(
+            t.extents[0].start_lba, 12640,
+            "extent must start at ifo_lba + vtstt_vobs (0xC4), not vtsm_vobs (0xC0)"
+        );
+        // Must not resolve from the menu VOBS (would be 9000 + 44 = 9044), nor
+        // use the raw IFO-relative vtstt_vobs (3640) without the absolute base.
+        assert_ne!(t.extents[0].start_lba, 9044, "must not use the menu VOBS");
+        assert_ne!(
             t.extents[0].start_lba, 3640,
-            "extent must start at vtstt_vobs (0xC4), not vtsm_vobs (0xC0)"
+            "must add the IFO's absolute disc LBA, not use the raw relative value"
+        );
+    }
+
+    /// ABSOLUTE-REBASE regression (THESILENCEOFTHELAMBS / Greenland fix):
+    /// `ifo::parse_vts` now sets `vob_start_sector = file_start_lba(IFO) +
+    /// vtstt_vobs`, so an extent's `start_lba` must equal the sum of THREE
+    /// independent terms — the IFO file's absolute on-disc LBA, the
+    /// IFO-relative `vtstt_vobs` (0xC4), and the cell's `first_sector` — none of
+    /// which may be dropped. The earlier code used the bare relative
+    /// `vtstt_vobs`, placing every extent `ifo_lba` sectors too early (the rip
+    /// opened in the VMGI/menu region before drifting into the movie). The
+    /// other tests fold two of the three terms together (zero cell offset, or
+    /// a single combined expectation); this one keeps all three distinct and
+    /// non-overlapping so a regression to ANY two-term combination is caught.
+    #[test]
+    fn scan_dvd_titles_extent_is_absolute_three_term_sum() {
+        let mut disc = MemDisc::new();
+        let vmg = build_vmg(&[(1, 1, 1)]);
+        // vtstt_vobs (Title VOBS, 0xC4) = 700; one cell first_sector = 33.
+        let vts = build_vts(700, 0x00, &[], &[], &[(33, 132)], false);
+        // IFO data at data_lba 6000 → absolute ifo_lba = PART_START(3000) + 6000.
+        let ifo_lba = PART_START + 6000; // 9000
+        let vtstt_vobs = 700u32;
+        let first_sector = 33u32;
+        let udf = build_video_ts_fs(
+            &mut disc,
+            &[
+                FileSpec {
+                    name: "VIDEO_TS.IFO".into(),
+                    icb_lba: 60,
+                    data_lba: 5000,
+                    contents: vmg,
+                },
+                FileSpec {
+                    name: "VTS_01_0.IFO".into(),
+                    icb_lba: 62,
+                    data_lba: 6000,
+                    contents: vts,
+                },
+            ],
+        );
+        let t = &Disc::scan_dvd_titles(&mut disc, &udf)[0];
+        assert_eq!(t.extents.len(), 1);
+        let got = t.extents[0].start_lba;
+        // The one correct answer: all three terms summed (9000 + 700 + 33).
+        assert_eq!(
+            got,
+            ifo_lba + vtstt_vobs + first_sector,
+            "extent start must be file_start_lba(IFO) + vtstt_vobs + cell.first_sector"
+        );
+        // Each wrong two-term combination must be rejected:
+        assert_ne!(
+            got,
+            vtstt_vobs + first_sector,
+            "must not use the bare relative vtstt_vobs (missing the IFO's absolute LBA)"
+        );
+        assert_ne!(
+            got,
+            ifo_lba + first_sector,
+            "must not drop vtstt_vobs (the Title VOBS pointer)"
+        );
+        assert_ne!(
+            got,
+            ifo_lba + vtstt_vobs,
+            "must not drop the cell's first_sector offset"
         );
-        assert_ne!(t.extents[0].start_lba, 44, "must not use the menu VOBS");
     }
 
     /// Multi-cell title: extents preserve cell order and each maps to its
@@ -648,9 +742,9 @@ mod tests {
         );
         let t = &Disc::scan_dvd_titles(&mut disc, &udf)[0];
         assert_eq!(t.extents.len(), 2);
-        assert_eq!(t.extents[0].start_lba, 500); // 500 + 0
+        assert_eq!(t.extents[0].start_lba, 9500); // ifo_lba(9000) + 500 + 0
         assert_eq!(t.extents[0].sector_count, 100);
-        assert_eq!(t.extents[1].start_lba, 700); // 500 + 200
+        assert_eq!(t.extents[1].start_lba, 9700); // ifo_lba(9000) + 500 + 200
         assert_eq!(t.extents[1].sector_count, 100);
         assert_eq!(t.size_bytes, 200 * 2048);
     }
@@ -1051,8 +1145,9 @@ mod tests {
         assert_eq!(titles[0].playlist, "VTS_01_1.VOB");
         assert_eq!(titles[1].playlist, "VTS_02_2.VOB");
         // Distinct vob_start → distinct extents.
-        assert_eq!(titles[0].extents[0].start_lba, 100);
-        assert_eq!(titles[1].extents[0].start_lba, 200);
+        // VTS_01 IFO @ PART_START(3000)+6000=9000; VTS_02 IFO @ 3000+7000=10000.
+        assert_eq!(titles[0].extents[0].start_lba, 9100); // 9000 + 100
+        assert_eq!(titles[1].extents[0].start_lba, 10200); // 10000 + 200
     }
 
     /// chapter_times from the IFO become Chapter entries with ordinal
@@ -1168,8 +1263,8 @@ mod tests {
         // The leading 0x90 cell is dropped: 2 feature extents, not 3.
         assert_eq!(t.extents.len(), 2, "leading angle sub-block cell dropped");
         // First extent starts at the feature cell (vob 1000 + 100), not at 1000+0.
-        assert_eq!(t.extents[0].start_lba, 1000 + 100);
-        assert_eq!(t.extents[1].start_lba, 1000 + 300);
+        assert_eq!(t.extents[0].start_lba, 9000 + 1000 + 100); // ifo_lba + vtstt + first
+        assert_eq!(t.extents[1].start_lba, 9000 + 1000 + 300);
         // Chapter times shift earlier by the dropped 5s. Program 0 was at the
         // dropped head (clamped to 0); program 1 was at cell 3 =
         // dur(cell0)+dur(cell1) = 5 + 59 = 64s, now 59s after the 5s shift.
@@ -1219,8 +1314,8 @@ mod tests {
         let t = &Disc::scan_dvd_titles(&mut disc, &udf)[0];
         // Nothing dropped: both cells become extents, starting at the very head.
         assert_eq!(t.extents.len(), 2);
-        assert_eq!(t.extents[0].start_lba, 1000); // 1000 + 0, head intact
-        assert_eq!(t.extents[1].start_lba, 1200);
+        assert_eq!(t.extents[0].start_lba, 9000 + 1000); // ifo_lba + vtstt + 0, head intact
+        assert_eq!(t.extents[1].start_lba, 9000 + 1200);
         // Chapter 0 stays at 0.0 (no shift).
         assert!((t.chapters[0].time_secs - 0.0).abs() < 0.01);
     }

src/disc/extract.rs

@@ -24,7 +24,7 @@ use std::io::Write;
 use std::path::{Path, PathBuf};
 use std::sync::atomic::Ordering;
 
-use crate::consts::SECTOR_BYTES;
+use crate::consts::{SECTOR_BYTES, SECTOR_BYTES_U64};
 /// AACS aligned unit = 3 sectors / 6144 bytes. Content reads are issued in
 /// multiples of this so the decrypt step always sees whole units.
 const AACS_UNIT_SECTORS: u32 = 3;
@@ -234,9 +234,12 @@ impl Disc {
 
             // Acquire (rather than Relaxed) on these per-file delta loads:
             // `extract_tree` drives `dec` single-threaded so there is no race
-            // today, but Acquire costs nothing on x86 and gives a happens-
-            // before edge if file extraction is ever parallelised, so the
-            // delta can never read a torn/stale counter across iterations.
+            // today, and Acquire costs nothing on x86. Note this is only half
+            // the synchronisation: the paired counter store
+            // (sector/decrypting.rs `fetch_add`) is Relaxed, so an Acquire
+            // load alone does NOT yet establish a happens-before edge. Before
+            // file extraction is parallelised, upgrade that store to Release
+            // (or stronger) so the delta cannot read a stale counter.
             let before_loss = decrypt_loss.load(Ordering::Acquire);
             let (mut fr, halted) =
                 extract_one_file(&mut dec, dest, pf, total_bytes, &mut done_bytes, opts)?;
@@ -282,7 +285,7 @@ impl Disc {
                 for &(abs_lba, byte_len) in &pf.extents {
                     extents.push(crate::disc::Extent {
                         start_lba: abs_lba,
-                        sector_count: (byte_len as u64).div_ceil(SECTOR_BYTES as u64) as u32,
+                        sector_count: (byte_len as u64).div_ceil(SECTOR_BYTES_U64) as u32,
                     });
                 }
             }
@@ -467,7 +470,7 @@ fn extract_one_file<S: SectorSource>(
         // the per-extent re-anchoring in the mux read paths
         // (`mux/disc.rs`, `sector/prefetched.rs`). No-op for CSS / None.
         dec.set_unit_base(abs_lba);
-        let sectors = (byte_len as u64).div_ceil(SECTOR_BYTES as u64) as u32;
+        let sectors = (byte_len as u64).div_ceil(SECTOR_BYTES_U64) as u32;
         let mut sector_off: u32 = 0;
         while sector_off < sectors {
             let mut batch = (sectors - sector_off).min(READ_BATCH_SECTORS);

src/disc/mod.rs

@@ -627,28 +627,29 @@ impl Codec {
     ];
 
     pub(crate) fn from_coding_type(ct: u8) -> Self {
+        use crate::consts::coding_type as c;
         match ct {
-            0x24 => Codec::Hevc,
-            0x1B => Codec::H264,
-            0xEA => Codec::Vc1,
-            0x02 => Codec::Mpeg2,
-            0x83 => Codec::TrueHd,
-            0x86 => Codec::DtsHdMa,
-            0x85 => Codec::DtsHdHr,
-            0x82 => Codec::Dts,
-            0x81 => Codec::Ac3,
-            0x84 | 0xA1 => Codec::Ac3Plus,
-            0x80 => Codec::Lpcm,
-            // 0x86 (primary) / 0xA2 (secondary) are the DTS-HD MA
-            // lossless pair, parallel to 0x81/0xA1 for AC-3. 0xA2 is
-            // lossless MA, not lossy HR.
-            0xA2 => Codec::DtsHdMa,
-            // 0x90 = Presentation Graphics (PG / subtitles). 0x91 = Interactive
-            // Graphics (IG / menus) and 0x92 = Text subtitles are distinct HDMV
-            // coding types and are NOT PG subtitle streams; only 0x90 maps to
-            // Pgs. IG (0x91) falls through to Unknown so the PMT/STN walker drops
-            // it rather than surfacing a bogus PGS subtitle track for a menu ES.
-            0x90 => Codec::Pgs,
+            c::HEVC => Codec::Hevc,
+            c::H264 => Codec::H264,
+            c::VC1 => Codec::Vc1,
+            c::MPEG2_VIDEO => Codec::Mpeg2,
+            c::TRUEHD => Codec::TrueHd,
+            c::DTS_HD_MA => Codec::DtsHdMa,
+            c::DTS_HD_HR => Codec::DtsHdHr,
+            c::DTS => Codec::Dts,
+            c::AC3 => Codec::Ac3,
+            c::AC3_PLUS | c::AC3_PLUS_SECONDARY => Codec::Ac3Plus,
+            c::LPCM => Codec::Lpcm,
+            // DTS_HD_MA (primary 0x86) / DTS_HD_SECONDARY (0xA2) are the
+            // DTS-HD MA lossless pair, parallel to AC3/AC3_PLUS_SECONDARY for
+            // AC-3. The secondary code is lossless MA, not lossy HR.
+            c::DTS_HD_SECONDARY => Codec::DtsHdMa,
+            // PG (0x90) = Presentation Graphics (subtitles). IG (0x91, menus)
+            // and TEXT_SUBTITLE (0x92) are distinct HDMV coding types and are
+            // NOT PG subtitle streams; only PG maps to Pgs. IG falls through to
+            // Unknown so the PMT/STN walker drops it rather than surfacing a
+            // bogus PGS subtitle track for a menu ES.
+            c::PG => Codec::Pgs,
             ct => Codec::Unknown(ct),
         }
     }
@@ -5068,8 +5069,8 @@ mod tests {
         let mf = Mapfile::load(&disc.mapfile_for(&iso_path)).expect("load mapfile");
         let good = mf.ranges_with(&[SectorStatus::Finished]);
         let bad_ranges = mf.ranges_with(&[SectorStatus::NonTrimmed]);
-        let disc_bytes = sectors as u64 * 2048;
-        const SEC: u64 = crate::consts::SECTOR_BYTES as u64;
+        const SEC: u64 = crate::consts::SECTOR_BYTES_U64;
+        let disc_bytes = sectors as u64 * SEC;
 
         // The first failing batch starts at LBA 320; everything before it read
         // cleanly and must be Finished.

src/disc/patch.rs

@@ -825,12 +825,12 @@ pub(super) fn handle_read_success<R: SectorSource + ?Sized>(
         state.stall_start = (state.now)();
         state.bytes_good_last = bytes_good_now;
     }
-    if (state.now)().duration_since(state.stall_start) > std::time::Duration::from_secs(STALL_SECS)
-    {
+    let stall_elapsed = (state.now)().duration_since(state.stall_start);
+    if stall_elapsed > std::time::Duration::from_secs(STALL_SECS) {
         tracing::warn!(
             target: "freemkv::disc",
             phase = "patch_stall",
-            elapsed_secs = (state.now)().duration_since(state.stall_start).as_secs(),
+            elapsed_secs = stall_elapsed.as_secs(),
             bytes_good = bytes_good_now,
             bytes_good_start = state.bytes_good_start,
             "Patch stalled - no recovery for {}s, exiting pass",
@@ -1107,13 +1107,12 @@ pub(super) fn handle_read_failure<R: SectorSource + ?Sized>(
                 state.stall_start = (state.now)();
                 state.bytes_good_last = bytes_good_now;
             }
-            if (state.now)().duration_since(state.stall_start)
-                > std::time::Duration::from_secs(STALL_SECS)
-            {
+            let stall_elapsed = (state.now)().duration_since(state.stall_start);
+            if stall_elapsed > std::time::Duration::from_secs(STALL_SECS) {
                 tracing::warn!(
                     target: "freemkv::disc",
                     phase = "patch_stall",
-                    elapsed_secs = (state.now)().duration_since(state.stall_start).as_secs(),
+                    elapsed_secs = stall_elapsed.as_secs(),
                     bytes_good = bytes_good_now,
                     bytes_good_start = state.bytes_good_start,
                     "Patch stalled (NOT_READY path) - no recovery for {}s, exiting pass",
@@ -1193,12 +1192,12 @@ pub(super) fn handle_read_failure<R: SectorSource + ?Sized>(
         state.stall_start = (state.now)();
         state.bytes_good_last = bytes_good_now;
     }
-    if (state.now)().duration_since(state.stall_start) > std::time::Duration::from_secs(STALL_SECS)
-    {
+    let stall_elapsed = (state.now)().duration_since(state.stall_start);
+    if stall_elapsed > std::time::Duration::from_secs(STALL_SECS) {
         tracing::warn!(
             target: "freemkv::disc",
             phase = "patch_stall",
-            elapsed_secs = (state.now)().duration_since(state.stall_start).as_secs(),
+            elapsed_secs = stall_elapsed.as_secs(),
             consecutive_failures = state.consecutive_failures,
             bytes_good = bytes_good_now,
             bytes_good_start = state.bytes_good_start,
@@ -1472,13 +1471,14 @@ pub(super) fn check_range_watchdog(
         state.range_bytes_good = bytes_good_now;
         state.range_start = (state.now)();
     }
-    if (state.now)().duration_since(state.range_start).as_secs() >= frame.range_budget_secs {
+    let range_elapsed = (state.now)().duration_since(state.range_start);
+    if range_elapsed.as_secs() >= frame.range_budget_secs {
         tracing::warn!(
             target: "freemkv::disc",
             phase = "patch_range_stall",
             range_lba = frame.range_pos / 2048,
             range_sectors = frame.range_sectors,
-            elapsed_secs = (state.now)().duration_since(state.range_start).as_secs(),
+            elapsed_secs = range_elapsed.as_secs(),
             budget_secs = frame.range_budget_secs,
             bytes_recovered = state.range_bytes_good.saturating_sub(state.bytes_good_before),
             "Range stalled - moving to next range"

src/drive/mod.rs

@@ -433,8 +433,12 @@ impl Drive {
                 }
                 Ok(())
             }
-            // No unlocker matched: not an error — fall through to OEM route.
+            // No unlocker matched, or one matched but only hit a capability
+            // failure (not firmware-unlockable / no OEM VID): not an error —
+            // fall through to the OEM host-cert route.
             Ok(None) => Ok(()),
+            // A genuine transport fault during unlock (UnlockError::Scsi)
+            // propagates here and aborts init — the bus is dead.
             Err(e) => Err(e),
         };
         tracing::info!(

src/dvdnav/mod.rs

@@ -0,0 +1,49 @@
+//! DVD-Video navigation — read-only resolver for the **main-feature start
+//! point** (issue #40). Mirrors what a DVD player's nav VM resolves: First-Play
+//! → menu "Play" → title dispatch → the first cell of the feature, so the rip
+//! starts at the movie rather than at raw cell 0 (e.g. skipping a leading
+//! logo/warning segment when the disc's own navigation does).
+//!
+//! Byte layout follows the DVD-Video specification (VMGI/VTSI headers,
+//! PGC/cell tables, PCI/HLI button packets); the VM command decoder is
+//! verified against libdvdnav's decoder.
+//!
+//! Current contents: [`vmcmd`] — the VM command decoder (proven against the
+//! SOTL/Greenland test discs). The IFO/PCI parsing and the navigation executor
+//! that resolves the start cell build on top of this.
+
+pub mod vmcmd;
+
+use crate::sector::SectorSource;
+
+/// Resolve the feature title's **true start cell** (0-based index into the
+/// title PGC's cell list) by following the disc's own navigation — First-Play →
+/// menu "Play" → title dispatch — the way a player reaches the movie. This is
+/// what lets the rip begin at the feature instead of at raw cell 0 when the
+/// disc's nav enters the title past a leading logo/warning segment (e.g. a
+/// disc whose "Play" resolves to a later cell than cell 0).
+///
+/// Returns `None` when navigation cannot be resolved, so the caller falls back
+/// to the structural leading-cell filter (today's behaviour, ≈ cell 0 / 0:00).
+///
+/// TODO(#40): the IFO/PCI parsing + nav executor (built on [`vmcmd`]) land
+/// incrementally. Until the executor is complete this returns `None`, so wiring
+/// it in is behaviour-neutral; improvements to the resolver take effect here
+/// without touching the call site.
+pub fn resolve_feature_start(
+    reader: &mut dyn SectorSource,
+    udf: &crate::udf::UdfFs,
+    vtsn: u16,
+    vts_ttn: u16,
+) -> Option<usize> {
+    // `reader`/`udf` are the seam inputs the nav executor will consume to read
+    // VIDEO_TS.IFO + the VTS IFOs/menu VOBs. Reserved until that lands.
+    let _ = (reader, udf);
+    tracing::trace!(
+        target: "freemkv::dvdnav",
+        vtsn,
+        vts_ttn,
+        "nav start-cell resolver: unresolved — caller falls back to leading-cell filter"
+    );
+    None
+}

src/dvdnav/vmcmd.rs

@@ -0,0 +1,408 @@
+//! DVD-Video VM command decoder.
+//!
+//! An 8-byte navigation command as found in PGC command tables (pre/post/cell)
+//! and PCI button info. Decoded per the DVD-Video VM instruction set and
+//! verified against libdvdnav's command decoder.
+//!
+//! Bit model: the 8 bytes are a big-endian 64-bit word. `byte0` bits 7-5 are the
+//! command **type**; for type 1, `byte0` bit 4 selects Link (0) vs Jump (1), and
+//! `byte1` bits 3-0 are the sub-command. Compare predicates live in `byte1`
+//! bits 6-4 with the operands in bytes 2-5.
+//!
+//! This module is pure decode + a register model — no I/O, no English (numeric
+//! semantics only), matching libfreemkv conventions. The navigation *executor*
+//! and IFO/PCI parsing build on top of this.
+
+/// A decoded navigation instruction. Only the variants freemkv's start-point
+/// resolver needs are modelled explicitly; everything else is [`Instr::Other`].
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub enum Instr {
+    Nop,
+    /// Stop executing the current command list (resume cell playback).
+    Break,
+    /// Goto command line within the same list (1-based).
+    Goto {
+        line: u8,
+    },
+    /// Leave the current domain.
+    Exit,
+    /// Jump to a VMG title (1-based TT_SRPT index).
+    JumpTt {
+        ttn: u8,
+    },
+    /// Jump to a title within the current VTS (1-based VTS title index).
+    JumpVtsTt {
+        ttn: u8,
+    },
+    /// Jump to a part-of-title (chapter) within a VTS title.
+    JumpVtsPtt {
+        ttn: u8,
+        pttn: u16,
+    },
+    /// Jump to the First-Play PGC.
+    JumpSsFp,
+    /// Jump to a Video-Manager menu (`menu` = menu id).
+    JumpSsVmgm {
+        menu: u8,
+    },
+    /// Jump to a Video-Title-Set menu.
+    JumpSsVtsm {
+        vts: u8,
+        ttn: u8,
+        menu: u8,
+    },
+    /// Jump to a specific VMGM menu PGC.
+    JumpSsVmgmPgc {
+        pgcn: u16,
+    },
+    /// Call a sub-domain (raw retained; resume handled by the executor).
+    CallSs {
+        sub: u8,
+    },
+    /// Link to a PGC number within the current domain.
+    LinkPgcn {
+        pgcn: u16,
+    },
+    /// Link to a part-of-title within the current PGC's title.
+    LinkPttn {
+        pttn: u16,
+    },
+    /// Link to a program number within the current PGC (1-based).
+    LinkPgn {
+        pgn: u8,
+    },
+    /// Link to a cell number within the current PGC (1-based).
+    LinkCn {
+        cn: u8,
+    },
+    /// A link "subset" op (LinkTopCell/NextPG/RSM/…); `sub` is the raw code.
+    LinkSub {
+        sub: u8,
+    },
+    /// Set a GPRM. `op` is the set-op code (1=mov, 3=add, …); value is immediate
+    /// (`imm`) when `immediate`, else the contents of register `src`.
+    SetGprm {
+        reg: u8,
+        op: u8,
+        immediate: bool,
+        imm: u16,
+        src: u8,
+    },
+    /// Set a system parameter / unmodelled set — executor may ignore.
+    SetSystem,
+    /// Anything not individually modelled (kept as raw bytes).
+    Other([u8; 8]),
+}
+
+/// A compare predicate carried by a command (`byte1` bits 6-4). `None` = always.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub struct Compare {
+    /// Compare op: 1=&,2===,3=!=,4=>=,5=>,6=<=,7=<.
+    pub op: u8,
+    /// Left register index (GPRM 0-15, SPRM 128+).
+    pub lhs_reg: u8,
+    /// Right side: immediate when `immediate`, else register `rhs_reg`.
+    pub immediate: bool,
+    pub imm: u16,
+    pub rhs_reg: u8,
+}
+
+/// A fully decoded command: its predicate (if any) and the instruction.
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub struct Command {
+    pub compare: Option<Compare>,
+    pub instr: Instr,
+}
+
+// Command types — `byte0` bits 7-5.
+const TYPE_SPECIAL: u8 = 0;
+const TYPE_LINK_JUMP: u8 = 1;
+const TYPE_SET_SYSTEM: u8 = 2;
+const TYPE_SET_GPRM: u8 = 3;
+
+// Special (type 0) sub-commands — `byte1` bits 3-0.
+const SP_GOTO: u8 = 1;
+const SP_BREAK: u8 = 2;
+
+// Jump/Call (type 1, direct=1) sub-commands.
+const JP_EXIT: u8 = 1;
+const JP_JUMP_TT: u8 = 2;
+const JP_JUMP_VTS_TT: u8 = 3;
+const JP_JUMP_VTS_PTT: u8 = 5;
+const JP_JUMP_SS: u8 = 6;
+const JP_CALL_SS: u8 = 8;
+
+// Link (type 1, direct=0) sub-commands. NOTE: sub-op 0 is NOP/no-link and 1 is
+// the LinkSub form (libdvdnav `decoder.c` `eval_link_instruction`).
+const LK_SUB: u8 = 1;
+const LK_PGCN: u8 = 4;
+const LK_PTTN: u8 = 5;
+const LK_PGN: u8 = 6;
+const LK_CN: u8 = 7;
+
+// JumpSS sub-domain selector — `byte5` bits 7-6.
+const SS_FP: u8 = 0;
+const SS_VMGM_MENU: u8 = 1;
+const SS_VTSM: u8 = 2;
+
+// Operand field widths (spec-defined bit counts).
+const MASK_TTN: u8 = 0x7F; // 7-bit title number
+const MASK_PGN: u8 = 0x7F; // 7-bit program number
+const MASK_LINKOP: u8 = 0x1F; // 5-bit link sub-op
+const MASK_REG: u8 = 0x0F; // 4-bit GPRM index
+const MASK_MENU: u8 = 0x0F; // 4-bit menu id
+const MASK_PTTN: u16 = 0x03FF; // 10-bit part-of-title
+const MASK_PGCN: u16 = 0x7FFF; // 15-bit PGC number
+
+#[inline]
+fn be16(b: &[u8; 8], o: usize) -> u16 {
+    ((b[o] as u16) << 8) | b[o + 1] as u16
+}
+
+// Compare-operand layouts ("if_version"s) per libdvdnav `decoder.c`. The op
+// nibble is always `byte1` bits 6-4; the immediate flag is `byte1` bit 7. The
+// operand *offsets* differ by command family.
+//
+// v1 (special + link): lhs reg = b[3]; rhs imm = bytes4-5 / rhs reg = b[4].
+// v2 (jump + system-set): lhs reg = b[6]; rhs reg = b[7] (registers only).
+// v3 (set-GPRM): lhs reg = b[2]; rhs imm = bytes6-7 / rhs reg = b[6].
+fn if_v1(b: &[u8; 8]) -> Option<Compare> {
+    let op = (b[1] >> 4) & 7;
+    (op != 0).then(|| Compare {
+        op,
+        lhs_reg: b[3],
+        immediate: b[1] >> 7 != 0,
+        imm: be16(b, 4),
+        rhs_reg: b[4],
+    })
+}
+fn if_v2(b: &[u8; 8]) -> Option<Compare> {
+    let op = (b[1] >> 4) & 7;
+    (op != 0).then(|| Compare {
+        op,
+        lhs_reg: b[6],
+        immediate: false,
+        imm: 0,
+        rhs_reg: b[7],
+    })
+}
+fn if_v3(b: &[u8; 8]) -> Option<Compare> {
+    let op = (b[1] >> 4) & 7;
+    (op != 0).then(|| Compare {
+        op,
+        lhs_reg: b[2],
+        immediate: b[1] >> 7 != 0,
+        imm: be16(b, 6),
+        rhs_reg: b[6],
+    })
+}
+
+/// Decode an 8-byte VM command.
+pub fn decode(b: &[u8; 8]) -> Command {
+    let typ = b[0] >> 5;
+    let direct = (b[0] >> 4) & 1;
+    let setop = b[0] & 0x0F;
+    let cmd = b[1] & 0x0F;
+
+    // Compare predicate, with the operand layout for this command family
+    // (libdvdnav `decoder.c` `vm_eval_command` type dispatch).
+    let compare = match (typ, direct) {
+        (TYPE_SPECIAL, _) => if_v1(b),
+        (TYPE_LINK_JUMP, 1) => if_v2(b), // jump
+        (TYPE_LINK_JUMP, 0) => if_v1(b), // link
+        (TYPE_SET_SYSTEM, _) => if_v2(b),
+        (TYPE_SET_GPRM, _) => if_v3(b),
+        _ => None, // 4/5/6 compound — not needed by the resolver
+    };
+
+    // JumpSS sub-domain selector lives in byte5 bits 7-6.
+    let ss_sel = b[5] >> 6;
+
+    let instr = match typ {
+        TYPE_LINK_JUMP if direct == 1 => match cmd {
+            JP_EXIT => Instr::Exit,
+            JP_JUMP_TT => Instr::JumpTt {
+                ttn: b[5] & MASK_TTN,
+            },
+            JP_JUMP_VTS_TT => Instr::JumpVtsTt {
+                ttn: b[5] & MASK_TTN,
+            },
+            JP_JUMP_VTS_PTT => Instr::JumpVtsPtt {
+                ttn: b[5] & MASK_TTN,
+                pttn: be16(b, 2) & MASK_PTTN,
+            },
+            JP_JUMP_SS => match ss_sel {
+                SS_FP => Instr::JumpSsFp,
+                SS_VMGM_MENU => Instr::JumpSsVmgm {
+                    menu: b[5] & MASK_MENU,
+                },
+                SS_VTSM => Instr::JumpSsVtsm {
+                    vts: b[4],
+                    ttn: b[3],
+                    menu: b[5] & MASK_MENU,
+                },
+                _ => Instr::JumpSsVmgmPgc {
+                    pgcn: be16(b, 2) & MASK_PGCN,
+                },
+            },
+            JP_CALL_SS => Instr::CallSs { sub: ss_sel },
+            _ => Instr::Nop,
+        },
+        TYPE_LINK_JUMP => match cmd {
+            // direct == 0 (link). sub-op 0 = NOP/no-link.
+            LK_SUB => Instr::LinkSub {
+                sub: b[7] & MASK_LINKOP,
+            },
+            LK_PGCN => Instr::LinkPgcn {
+                pgcn: be16(b, 6) & MASK_PGCN,
+            },
+            LK_PTTN => Instr::LinkPttn {
+                pttn: be16(b, 6) & MASK_PTTN,
+            },
+            LK_PGN => Instr::LinkPgn {
+                pgn: b[7] & MASK_PGN,
+            },
+            LK_CN => Instr::LinkCn { cn: b[7] },
+            _ => Instr::Nop,
+        },
+        TYPE_SPECIAL => match cmd {
+            SP_GOTO => Instr::Goto { line: b[7] },
+            SP_BREAK => Instr::Break,
+            _ => Instr::Nop,
+        },
+        TYPE_SET_GPRM => Instr::SetGprm {
+            reg: b[3] & MASK_REG,
+            op: setop,
+            immediate: direct != 0,
+            imm: be16(b, 4),
+            src: b[5],
+        },
+        TYPE_SET_SYSTEM => Instr::SetSystem,
+        _ => Instr::Other(*b),
+    };
+
+    Command { compare, instr }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn h(s: &str) -> [u8; 8] {
+        let v: Vec<u8> = (0..8)
+            .map(|i| u8::from_str_radix(&s[i * 2..i * 2 + 2], 16).unwrap())
+            .collect();
+        v.try_into().unwrap()
+    }
+
+    // KATs taken from the real SOTL / Greenland discs (decoded in the PoC).
+    #[test]
+    fn greenland_first_play_is_jumptt_1() {
+        let c = decode(&h("3002000000010000"));
+        assert_eq!(c.instr, Instr::JumpTt { ttn: 1 });
+        assert!(c.compare.is_none());
+    }
+
+    #[test]
+    fn sotl_first_play_is_jumpss_vtsm_root() {
+        // 30 06 ... byte5=0x83 -> sub 2 (VTSM), vts=byte4=1, menu=byte5&0xF=3 (root)
+        let c = decode(&h("3006000101830000"));
+        assert_eq!(
+            c.instr,
+            Instr::JumpSsVtsm {
+                vts: 1,
+                ttn: 1,
+                menu: 3
+            }
+        );
+    }
+
+    #[test]
+    fn sotl_title_dispatch_is_conditional_linkpgn_2() {
+        // 20 a6 ... CmpLink: if GPRM0 == 2 -> LinkPGN 2  (cell 2 = the 5:02 start)
+        let c = decode(&h("20a6000000020002"));
+        assert_eq!(c.instr, Instr::LinkPgn { pgn: 2 });
+        let cmp = c.compare.expect("conditional");
+        assert_eq!(cmp.op, 2); // ==
+        assert_eq!(cmp.lhs_reg, 0); // GPRM0
+        assert!(cmp.immediate);
+        assert_eq!(cmp.imm, 2);
+    }
+
+    #[test]
+    fn sotl_root_button_is_linkpgcn_37() {
+        assert_eq!(
+            decode(&h("2004000000000025")).instr,
+            Instr::LinkPgcn { pgcn: 37 }
+        );
+    }
+
+    #[test]
+    fn greenland_scene_button_is_linkpgn() {
+        assert_eq!(
+            decode(&h("2006000000001401")).instr,
+            Instr::LinkPgn { pgn: 1 }
+        );
+    }
+
+    #[test]
+    fn jumpvts_ptt_decodes_ttn_and_pttn() {
+        // synthetic: 30 05 | ptt(bytes2-3)=0x0002 | ttn(byte5)=1
+        let c = decode(&h("3005000200010000"));
+        assert_eq!(c.instr, Instr::JumpVtsPtt { ttn: 1, pttn: 2 });
+    }
+
+    #[test]
+    fn setgprm_immediate_mov() {
+        // SOTL First-Play pre[0]: 71 00 | reg=byte3=6 | imm(bytes4-5)=0x03e8 -> g6 = 1000
+        match decode(&h("7100000603e80000")).instr {
+            Instr::SetGprm {
+                reg,
+                op,
+                immediate,
+                imm,
+                ..
+            } => {
+                assert_eq!(reg, 6);
+                assert_eq!(op, 1); // mov
+                assert!(immediate);
+                assert_eq!(imm, 1000);
+            }
+            other => panic!("expected SetGprm, got {other:?}"),
+        }
+    }
+
+    // Regression for the libdvdnav cross-check: link sub-op 0 = NOP, 1 = LinkSub.
+    #[test]
+    fn link_subop_zero_is_nop_one_is_linksub() {
+        assert_eq!(decode(&h("2000000000000000")).instr, Instr::Nop);
+        assert_eq!(
+            decode(&h("2001000000000010")).instr,
+            Instr::LinkSub { sub: 0x10 }
+        );
+    }
+
+    // if_version_1 register compare: rhs register is byte4 (not byte5).
+    #[test]
+    fn link_register_compare_rhs_is_byte4() {
+        // 20 26: link, cmp=EQ(2), dircmp=0(register) ; cmd=6 LinkPGN
+        let c = decode(&h("2026000304000002"));
+        assert_eq!(c.instr, Instr::LinkPgn { pgn: 2 });
+        let cmp = c.compare.expect("conditional");
+        assert!(!cmp.immediate);
+        assert_eq!(cmp.lhs_reg, 3);
+        assert_eq!(cmp.rhs_reg, 4);
+    }
+
+    // if_version_2 jump compare: both operands are registers in byte6 / byte7.
+    #[test]
+    fn jump_compare_uses_bytes6_and_7() {
+        // 30 22: jump, cmp=EQ(2) ; cmd=2 JumpTT ttn=byte5=5
+        let c = decode(&h("3022000000050607"));
+        assert_eq!(c.instr, Instr::JumpTt { ttn: 5 });
+        let cmp = c.compare.expect("conditional");
+        assert!(!cmp.immediate);
+        assert_eq!(cmp.lhs_reg, 6);
+        assert_eq!(cmp.rhs_reg, 7);
+    }
+}

src/error.rs

@@ -1199,6 +1199,7 @@ mod tests {
             E_NO_DISC_KEY,
             E_CSS_KEY_MISSING,
             E_AACS_NO_HOST_CERT,
+            E_AACS_BUS_KEY_UNAVAILABLE,
             E_KEYDB_CONNECT,
             E_KEYDB_HTTP,
             E_KEYDB_INVALID,

src/ifo.rs

@@ -440,7 +440,19 @@ fn parse_vts(
     // to the feature and shifted every cell extent back by
     // `vtstt_vobs - vtsm_vobs` sectors, so the rip opened on the parental
     // prompt instead of the movie. The title content lives at `vtstt_vobs`.
-    let vob_start_sector = be_u32(&vts_data, VTSTT_VOBS_OFFSET)?;
+    //
+    // `vtstt_vobs` is a sector address **relative to the start of this VTS_xx_0.IFO
+    // file**, not an absolute disc LBA. The cell `first_sector`/`last_sector`
+    // values are in turn relative to `vtstt_vobs`. To turn them into the absolute
+    // disc LBAs the reader needs, add the IFO file's own on-disc location (from
+    // the UDF FS). Without this rebase every extent started `ifo_lba` sectors too
+    // early — for THESILENCEOFTHELAMBS the feature began at LBA 126 (the VMGI /
+    // VIDEO_TS.VOB main-menu region) instead of 132886 (VTS_03_1.VOB), so the
+    // first ~4.5 min of muxed video was the disc's main menu before the stream
+    // drifted into the movie.
+    let vtstt_vobs = be_u32(&vts_data, VTSTT_VOBS_OFFSET)?;
+    let ifo_lba = udf.file_start_lba(reader, &path)?;
+    let vob_start_sector = ifo_lba.saturating_add(vtstt_vobs);
 
     // Video attributes at offset 0x200 (2 bytes)
     let video = parse_video_attr(&vts_data)?;

src/io/file_sector_source/mod.rs

@@ -72,7 +72,7 @@ use std::path::Path;
 use crate::error::{Error, Result};
 use crate::sector::SectorSource;
 
-use crate::consts::SECTOR_BYTES;
+use crate::consts::{SECTOR_BYTES, SECTOR_BYTES_U64};
 
 /// Bytes-read threshold per `posix_fadvise(DONTNEED)` drop on the
 /// read side. Mirrors `WRITEBACK_CHUNK_BYTES` so the read-side page
@@ -134,7 +134,7 @@ impl FileSectorSource {
             .metadata()
             .map_err(|e| Error::IoError { source: e })?
             .len();
-        let sectors = len / SECTOR_BYTES as u64;
+        let sectors = len / SECTOR_BYTES_U64;
         if sectors > u32::MAX as u64 {
             return Err(Error::IsoTooLarge {
                 path: path.to_string_lossy().into_owned(),
@@ -180,7 +180,7 @@ impl SectorSource for FileSectorSource {
         if count == 0 {
             return Ok(0);
         }
-        let offset = lba as u64 * SECTOR_BYTES as u64;
+        let offset = lba as u64 * SECTOR_BYTES_U64;
         self.file
             .seek(SeekFrom::Start(offset))
             .map_err(|e| Error::IoError { source: e })?;
@@ -494,7 +494,7 @@ mod tests {
     #[test]
     fn dontneed_eviction_does_not_affect_data() {
         // 32 MiB default chunk = 16384 sectors; read a bit past it.
-        let total = (READ_DROP_CHUNK_BYTES_DEFAULT / SECTOR_BYTES as u64) as u32 + 64;
+        let total = (READ_DROP_CHUNK_BYTES_DEFAULT / SECTOR_BYTES_U64) as u32 + 64;
         let dir = tempdir().unwrap();
         let path = dir.path().join("drop.iso");
         make_iso(&path, total);

src/keydb.rs

@@ -1,869 +0,0 @@
-//! KEYDB.cfg updater — HTTP GET, unzip, verify, save.
-//!
-//! Zero external HTTP dependencies. Raw TCP for HTTP GET.
-//! Uses `zip` and `flate2` (already in deps) for extraction.
-
-use crate::error::{Error, Result};
-use std::io::{Read, Write};
-use std::net::{TcpStream, ToSocketAddrs};
-use std::path::PathBuf;
-use std::time::Duration;
-
-/// Network operation timeout (connect / read / write). Keeps the daily
-/// refresh thread from blocking indefinitely on an unresponsive mirror.
-const NET_TIMEOUT: Duration = Duration::from_secs(10);
-
-/// Read timeout — longer than connect/write since the keydb body can be
-/// several MiB over a slow link.
-const READ_TIMEOUT: Duration = Duration::from_secs(30);
-
-/// Maximum redirects to follow before giving up.
-const MAX_REDIRECTS: usize = 5;
-
-/// Upper bound on decompressed keydb size. The published keydb is a few
-/// MiB; 64 MiB is a generous ceiling that still caps a decompression
-/// bomb (a tiny zip/gz can otherwise inflate to GiB and OOM the daily
-/// refresh thread).
-const MAX_KEYDB_BYTES: u64 = 64 * 1024 * 1024;
-
-/// Read a decompressed stream into a String with a hard size ceiling.
-/// Returns `Error::KeydbInvalid` if the input exceeds the cap, or
-/// `Error::KeydbParse` if the bytes are not valid UTF-8.
-fn read_capped_to_string<R: Read>(reader: R) -> Result<String> {
-    let mut buf = Vec::new();
-    // Read one byte past the cap so an exactly-at-cap stream is accepted
-    // but anything larger is rejected.
-    reader
-        .take(MAX_KEYDB_BYTES + 1)
-        .read_to_end(&mut buf)
-        .map_err(|_| Error::KeydbParse)?;
-    if buf.len() as u64 > MAX_KEYDB_BYTES {
-        return Err(Error::KeydbInvalid);
-    }
-    String::from_utf8(buf).map_err(|_| Error::KeydbParse)
-}
-
-/// Build the error returned when the executable's own directory can't be
-/// determined (`std::env::current_exe()` fails or has no parent). This is an
-/// *environment* failure — the process can't locate itself, which typically
-/// signals a stripped container or CI configuration — not a corrupt or
-/// unparseable keydb file. Map it to a `NotFound` I/O error so
-/// display/remediation paths never claim a keydb parse failure for a file
-/// that was never consulted.
-fn no_home_dir() -> Error {
-    Error::IoError {
-        source: std::io::Error::from(std::io::ErrorKind::NotFound),
-    }
-}
-
-/// Standard keydb storage path — the canonical location to write the keydb to.
-///
-/// The keydb lives *next to the executable*: `<dir of current exe>/keydb.cfg`,
-/// where the directory is `std::env::current_exe()`'s parent. This makes
-/// freemkv a portable, standalone binary — drop the exe and its `keydb.cfg`
-/// in the same folder and it works, with no OS-specific config dir
-/// (`%APPDATA%`, `~/.config`, XDG) involved at all.
-///
-/// The CLI's read-side search lives in
-/// `freemkv-keysources::keydb_search_paths`; it resolves the same exe-local
-/// location, so the *write* default used by `save`/`update` and the read-side
-/// search always agree.
-///
-/// Returns a `NotFound` I/O error (via [`no_home_dir`]) if the executable's
-/// own directory can't be determined.
-pub fn default_path() -> Result<PathBuf> {
-    std::env::current_exe()
-        .ok()
-        .and_then(|exe| exe.parent().map(|dir| dir.join("keydb.cfg")))
-        .ok_or_else(no_home_dir)
-}
-
-/// Download a KEYDB from a URL, verify, save to the standard path.
-pub fn update(url: &str) -> Result<UpdateResult> {
-    let body = http_get(url)?;
-    save(&body)
-}
-
-/// Verify and save raw keydb bytes (plain text, .zip, or .gz).
-pub fn save(data: &[u8]) -> Result<UpdateResult> {
-    let text = if data.starts_with(b"PK\x03\x04") {
-        extract_zip(data)?
-    } else if data.starts_with(&[0x1f, 0x8b]) {
-        read_capped_to_string(flate2::read::GzDecoder::new(data))?
-    } else {
-        // Plain-text body: route through the same capped reader as the gz/zip
-        // branches so an oversized uncompressed upload can't bypass
-        // MAX_KEYDB_BYTES.
-        read_capped_to_string(std::io::Cursor::new(data))?
-    };
-
-    let entries = text
-        .lines()
-        .filter(|l| {
-            let t = l.trim();
-            t.starts_with("0x")
-                || t.starts_with("| DK")
-                || t.starts_with("| PK")
-                || t.starts_with("| HC")
-        })
-        .count();
-
-    if entries == 0 {
-        return Err(Error::KeydbInvalid);
-    }
-
-    let path = default_path()?;
-    write_atomic(&path, &text)?;
-
-    Ok(UpdateResult {
-        path,
-        entries,
-        bytes: text.len(),
-    })
-}
-
-/// Write `text` to `path` crash-safely (create parent dir, write a sibling
-/// temp file, fsync, then atomic rename).
-///
-/// keydb.cfg is the single source of AACS truth, and `save`/`update` run
-/// unattended (first-boot download + daily-refresh thread, with a container
-/// restart on every release). A bare in-place `fs::write` truncates the file
-/// before writing, so a SIGKILL (docker stop's grace window), OOM-kill, power
-/// loss, or ENOSPC mid-write would leave the keydb half-written — the prior
-/// good copy already gone. A truncated keydb doesn't error at write time; it
-/// silently breaks key resolution on every later AACS rip. Writing to a temp
-/// file then renaming (POSIX rename is atomic within a filesystem) means an
-/// interrupted update leaves the previous keydb fully intact.
-///
-/// The fsync MUST succeed before the rename: a `sync_all` failure (ENOSPC,
-/// ESTALE on the bind-mounted volume) means the kernel never guaranteed the
-/// bytes reached stable storage, so publishing them via rename would defeat
-/// crash-safety. The temp name is unique per call (pid + monotonic counter)
-/// so a concurrent update can't share a fixed temp path and rename a mangled
-/// file over the keydb.
-fn write_atomic(path: &std::path::Path, text: &str) -> Result<()> {
-    let werr = || Error::KeydbWrite {
-        path: path.display().to_string(),
-    };
-    if let Some(dir) = path.parent() {
-        std::fs::create_dir_all(dir).map_err(|e| {
-            tracing::warn!(error = %e, path = %path.display(), "keydb dir create failed");
-            werr()
-        })?;
-    }
-    let tmp = {
-        use std::sync::atomic::{AtomicU64, Ordering};
-        static TMP_COUNTER: AtomicU64 = AtomicU64::new(0);
-        path.with_extension(format!(
-            "tmp.{}.{}",
-            std::process::id(),
-            TMP_COUNTER.fetch_add(1, Ordering::Relaxed)
-        ))
-    };
-    let write_result = (|| -> std::io::Result<()> {
-        let mut f = std::fs::File::create(&tmp)?;
-        f.write_all(text.as_bytes())?;
-        f.sync_all()?;
-        Ok(())
-    })();
-    if let Err(e) = write_result {
-        let _ = std::fs::remove_file(&tmp);
-        tracing::warn!(error = %e, path = %path.display(), "keydb write/fsync failed; keydb unchanged");
-        return Err(werr());
-    }
-    if let Err(e) = std::fs::rename(&tmp, path) {
-        let _ = std::fs::remove_file(&tmp);
-        tracing::warn!(error = %e, path = %path.display(), "keydb rename failed; keydb unchanged");
-        return Err(werr());
-    }
-    // Durably commit the new dirent: on POSIX filesystems (ext2, some NFS) a
-    // crash right after the rename can lose the directory entry even though the
-    // rename returned. Best-effort (swallowed on failure); no-op on Windows.
-    if let Some(dir) = path.parent() {
-        crate::io::fsync::dir(dir);
-    }
-    Ok(())
-}
-
-/// Result of a KEYDB update -- path written, entry count, and byte size.
-#[derive(Debug)]
-pub struct UpdateResult {
-    pub path: PathBuf,
-    pub entries: usize,
-    pub bytes: usize,
-}
-
-fn http_get(url: &str) -> Result<Vec<u8>> {
-    let (mut host, mut port, mut path) = parse_url(url)?;
-
-    for _ in 0..MAX_REDIRECTS {
-        // Resolve to a concrete socket address so we can bound the connect
-        // with connect_timeout (plain connect() uses the OS default, which
-        // can be minutes).
-        let addr = (host.as_str(), port)
-            .to_socket_addrs()
-            .ok()
-            .and_then(|mut it| it.next())
-            .ok_or_else(|| Error::KeydbConnect { host: host.clone() })?;
-        let mut stream = TcpStream::connect_timeout(&addr, NET_TIMEOUT).map_err(|e| {
-            tracing::debug!(error = %e, host = %host, "keydb connect failed");
-            Error::KeydbConnect { host: host.clone() }
-        })?;
-        stream
-            .set_read_timeout(Some(READ_TIMEOUT))
-            .map_err(|_| Error::KeydbConnect { host: host.clone() })?;
-        stream
-            .set_write_timeout(Some(NET_TIMEOUT))
-            .map_err(|_| Error::KeydbConnect { host: host.clone() })?;
-
-        // HTTP/1.0 forces close-delimited framing: the server cannot reply
-        // with Transfer-Encoding: chunked, so the raw body is the keydb
-        // bytes with no chunk-size lines to de-frame.
-        let request = format!(
-            "GET {path} HTTP/1.0\r\nHost: {host}\r\nConnection: close\r\nAccept-Encoding: identity\r\n\r\n"
-        );
-        stream
-            .write_all(request.as_bytes())
-            .map_err(|_| Error::KeydbConnect { host: host.clone() })?;
-
-        // Read the header block incrementally up to the \r\n\r\n terminator,
-        // bounded to ~64 KiB, BEFORE pulling any body. This avoids buffering up
-        // to 100 MiB per redirect hop just to inspect the status / Location.
-        const MAX_HEADER_BYTES: usize = 64 * 1024;
-        let mut reader = std::io::BufReader::new(stream);
-        let mut header_buf: Vec<u8> = Vec::with_capacity(1024);
-        let mut byte = [0u8; 1];
-        loop {
-            let n = reader
-                .read(&mut byte)
-                .map_err(|_| Error::KeydbConnect { host: host.clone() })?;
-            if n == 0 {
-                // Server closed the connection before the header block
-                // completed: a connection/protocol-level fault, not malformed
-                // keydb content.
-                return Err(Error::KeydbConnect { host: host.clone() });
-            }
-            header_buf.push(byte[0]);
-            if header_buf.ends_with(b"\r\n\r\n") {
-                break;
-            }
-            if header_buf.len() >= MAX_HEADER_BYTES {
-                // Oversized header block from the server: a protocol-level
-                // fault, not a keydb content parse failure. `>=` caps the
-                // buffer at exactly MAX_HEADER_BYTES (the `>` form admitted one
-                // extra byte before tripping).
-                return Err(Error::KeydbConnect { host: host.clone() });
-            }
-        }
-        // header_buf includes the trailing \r\n\r\n.
-        let header_end = header_buf.len() - 4;
-        // Lossy: a stray non-UTF-8 byte in the header block must not blank
-        // out the whole status line / Location header (which would surface
-        // as an undiagnosable KeydbHttp{status:0}).
-        let headers = String::from_utf8_lossy(&header_buf[..header_end]).into_owned();
-        let headers = headers.as_str();
-
-        let status = parse_status(headers).ok_or(Error::KeydbParse)?;
-
-        // Only treat a Location header as a redirect when the status is
-        // actually 3xx; a 200 carrying a stray Location (some proxies) is
-        // not a redirect, and a 3xx without Location is a malformed redirect.
-        if (300..=399).contains(&status) {
-            let location =
-                extract_header(headers, "Location").ok_or(Error::KeydbHttp { status })?;
-            let (next_host, next_port, next_path) = resolve_redirect(&location, &host, port)?;
-            host = next_host;
-            port = next_port;
-            path = next_path;
-            continue;
-        }
-
-        if status != 200 {
-            return Err(Error::KeydbHttp { status });
-        }
-
-        // Now read the body, still bounded by the existing 100 MiB cap. The
-        // BufReader carries any bytes already buffered past the header.
-        let mut body = Vec::new();
-        reader
-            .take(100 * 1024 * 1024)
-            .read_to_end(&mut body)
-            .map_err(|_| Error::KeydbConnect { host: host.clone() })?;
-        return Ok(body);
-    }
-
-    Err(Error::KeydbTooManyRedirects)
-}
-
-/// Resolve a `Location` value against the current request target.
-/// Handles absolute `http://` URLs, scheme-relative `//host/path`,
-/// absolute paths `/path`, and rejects unsupported schemes (e.g.
-/// `https://`, which this dependency-light client cannot fetch) with a
-/// diagnosable error rather than a generic parse failure.
-fn resolve_redirect(
-    location: &str,
-    cur_host: &str,
-    cur_port: u16,
-) -> Result<(String, u16, String)> {
-    let loc = location.trim();
-
-    if let Some(rest) = loc.strip_prefix("//") {
-        // Scheme-relative: //host[:port]/path — inherit http.
-        return parse_url(&format!("http://{rest}"));
-    }
-    if loc.starts_with('/') {
-        // Absolute path on the same host/port.
-        return Ok((cur_host.to_string(), cur_port, loc.to_string()));
-    }
-    if let Some(scheme) = loc.split("://").next() {
-        if loc.contains("://") && !scheme.eq_ignore_ascii_case("http") {
-            return Err(Error::KeydbUnsupportedScheme {
-                scheme: scheme.to_string(),
-            });
-        }
-    }
-    parse_url(loc)
-}
-
-fn parse_url(url: &str) -> Result<(String, u16, String)> {
-    // Reject non-http(s) up front so the caller gets a scheme diagnostic
-    // rather than an opaque parse error.
-    if let Some(scheme) = url.split("://").next() {
-        if url.contains("://") && !scheme.eq_ignore_ascii_case("http") {
-            return Err(Error::KeydbUnsupportedScheme {
-                scheme: scheme.to_string(),
-            });
-        }
-    }
-    let url = url.strip_prefix("http://").ok_or(Error::KeydbParse)?;
-    let (host_port, path) = match url.find('/') {
-        Some(i) => (&url[..i], &url[i..]),
-        None => (url, "/"),
-    };
-    let (host, port) = match host_port.find(':') {
-        Some(i) => {
-            let port_str = &host_port[i + 1..];
-            // A non-empty-but-unparseable port is a malformed URL; only an
-            // omitted port defaults to 80.
-            let port = if port_str.is_empty() {
-                80
-            } else {
-                port_str.parse().map_err(|_| Error::KeydbParse)?
-            };
-            (&host_port[..i], port)
-        }
-        None => (host_port, 80u16),
-    };
-    Ok((host.to_string(), port, path.to_string()))
-}
-
-fn parse_status(headers: &str) -> Option<u16> {
-    headers
-        .lines()
-        .next()
-        .and_then(|l| l.split_whitespace().nth(1))
-        .and_then(|s| s.parse().ok())
-}
-
-/// Locate the end of the HTTP header block (the index of the `\r\n\r\n`).
-/// Retained for the framing unit tests; the live path now reads headers
-/// incrementally in `http_get` so the whole response is never buffered.
-#[cfg_attr(not(test), allow(dead_code))]
-fn find_header_end(data: &[u8]) -> Option<usize> {
-    data.windows(4).position(|w| w == b"\r\n\r\n")
-}
-
-fn extract_header(headers: &str, name: &str) -> Option<String> {
-    // Split on the first ':' rather than byte-indexing at name.len(),
-    // which would panic on a multibyte UTF-8 codepoint straddling that
-    // offset (headers are decoded from untrusted network bytes). Also
-    // accepts single-character values (e.g. "Location:x").
-    for line in headers.lines() {
-        if let Some((key, value)) = line.split_once(':') {
-            if key.trim().eq_ignore_ascii_case(name) {
-                return Some(value.trim().to_string());
-            }
-        }
-    }
-    None
-}
-
-fn extract_zip(data: &[u8]) -> Result<String> {
-    let cursor = std::io::Cursor::new(data);
-    let mut archive = zip::ZipArchive::new(cursor).map_err(|_| Error::KeydbParse)?;
-
-    for i in 0..archive.len() {
-        let file = archive.by_index(i).map_err(|_| Error::KeydbParse)?;
-        if file.name().ends_with(".cfg") || file.name().ends_with(".CFG") {
-            return read_capped_to_string(file);
-        }
-    }
-
-    Err(Error::KeydbInvalid)
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    // Per project convention, tests never touch /tmp (wiped on reboot).
-    // Anchor scratch under the crate's target/ (gitignored), not /tmp.
-    fn scratch(tag: &str) -> std::path::PathBuf {
-        use std::sync::atomic::{AtomicU64, Ordering};
-        static CTR: AtomicU64 = AtomicU64::new(0);
-        let n = CTR.fetch_add(1, Ordering::Relaxed);
-        let d = std::path::PathBuf::from(env!("CARGO_MANIFEST_DIR"))
-            .join("target/test-scratch")
-            .join(format!("keydb-test-{}-{}-{}", std::process::id(), tag, n));
-        let _ = std::fs::remove_dir_all(&d);
-        std::fs::create_dir_all(&d).unwrap();
-        d
-    }
-
-    // Regression: a missing home directory (HOME/USERPROFILE unset) is an
-    // *environment* failure, not a corrupt keydb. It must NOT surface as
-    // E8004 (KeydbParse → "failed to parse the keydb file"), which would
-    // blame a file that was never consulted. It maps to a NotFound I/O
-    // error in the 5xxx (I/O) category instead.
-    #[test]
-    fn no_home_dir_is_io_not_found_not_keydb_parse() {
-        let e = no_home_dir();
-        match e {
-            Error::IoError { source } => {
-                assert_eq!(source.kind(), std::io::ErrorKind::NotFound);
-            }
-            other => panic!("expected IoError(NotFound), got {other:?}"),
-        }
-        // And explicitly: it is not the keydb-parse code.
-        assert_ne!(no_home_dir().code(), Error::KeydbParse.code());
-    }
-
-    // The write default is local to the executable: `<exe dir>/keydb.cfg`.
-    // Under `cargo test`, `current_exe()` is the test binary in `target/…`;
-    // assert against the same computation, not a hardcoded path.
-    #[test]
-    fn default_path_is_local_to_executable() {
-        let expected = std::env::current_exe()
-            .ok()
-            .and_then(|exe| exe.parent().map(|dir| dir.join("keydb.cfg")));
-        match expected {
-            Some(p) => assert_eq!(default_path().unwrap(), p),
-            None => {
-                // No exe parent available → must surface the env failure.
-                assert!(matches!(default_path(), Err(Error::IoError { .. })));
-            }
-        }
-    }
-
-    #[test]
-    fn write_atomic_replaces_existing_and_leaves_no_temp() {
-        let dir = scratch("atomic");
-        let path = dir.join("freemkv").join("keydb.cfg");
-
-        // First write creates the parent dir + file.
-        write_atomic(&path, "0xAAAA = old\n").unwrap();
-        assert_eq!(std::fs::read_to_string(&path).unwrap(), "0xAAAA = old\n");
-
-        // Second write replaces it in place.
-        write_atomic(&path, "0xBBBB = new\n").unwrap();
-        assert_eq!(std::fs::read_to_string(&path).unwrap(), "0xBBBB = new\n");
-
-        // No leftover *.tmp.* sibling — the temp file was renamed, not orphaned.
-        let leftovers: Vec<_> = std::fs::read_dir(path.parent().unwrap())
-            .unwrap()
-            .filter_map(|e| e.ok())
-            .map(|e| e.file_name().to_string_lossy().into_owned())
-            .filter(|n| n.contains(".tmp."))
-            .collect();
-        assert!(leftovers.is_empty(), "stray temp files: {leftovers:?}");
-    }
-
-    #[test]
-    fn write_atomic_failure_preserves_prior_keydb() {
-        // Simulate the crash window: a good keydb already on disk, then an
-        // update whose write target can't be created (parent path is a file,
-        // so create_dir_all under it fails — i.e. ENOTDIR). The rename never
-        // happens, so the existing keydb must survive untouched.
-        let dir = scratch("preserve");
-        let good = dir.join("keydb.cfg");
-        write_atomic(&good, "0xGOOD = keep\n").unwrap();
-
-        // `good` is a regular file; treating it as a directory parent fails.
-        let doomed = good.join("freemkv").join("keydb.cfg");
-        let err = write_atomic(&doomed, "0xBAD = partial\n");
-        assert!(matches!(err, Err(Error::KeydbWrite { .. })));
-
-        // Prior good copy is intact.
-        assert_eq!(std::fs::read_to_string(&good).unwrap(), "0xGOOD = keep\n");
-    }
-
-    #[test]
-    fn parse_url_defaults_and_paths() {
-        let (h, p, path) = parse_url("http://example.com/keydb.zip").unwrap();
-        assert_eq!(
-            (h.as_str(), p, path.as_str()),
-            ("example.com", 80, "/keydb.zip")
-        );
-
-        let (h, p, path) = parse_url("http://example.com:8080").unwrap();
-        assert_eq!((h.as_str(), p, path.as_str()), ("example.com", 8080, "/"));
-    }
-
-    #[test]
-    fn parse_url_rejects_https_scheme() {
-        // TLS is unsupported by this client; surface a scheme diagnostic
-        // rather than a generic parse error.
-        assert!(matches!(
-            parse_url("https://example.com/k.zip"),
-            Err(Error::KeydbUnsupportedScheme { .. })
-        ));
-    }
-
-    #[test]
-    fn parse_url_rejects_malformed_port() {
-        // Non-empty-but-unparseable port must error, not silently fall to 80.
-        assert!(matches!(
-            parse_url("http://example.com:abc/path"),
-            Err(Error::KeydbParse)
-        ));
-        // An empty port still defaults to 80.
-        let (_, p, _) = parse_url("http://example.com:/path").unwrap();
-        assert_eq!(p, 80);
-    }
-
-    #[test]
-    fn redirect_to_https_is_unsupported_scheme_not_parse_error() {
-        // The bplaced-style mirror enabling TLS on a redirect must produce a
-        // diagnosable scheme error, not KeydbParse.
-        assert!(matches!(
-            resolve_redirect("https://mirror.example/keydb.zip", "old.host", 80),
-            Err(Error::KeydbUnsupportedScheme { .. })
-        ));
-    }
-
-    #[test]
-    fn redirect_scheme_relative_and_absolute_path() {
-        // Scheme-relative //host/path inherits http.
-        let (h, p, path) = resolve_redirect("//mirror.example/a.zip", "old.host", 80).unwrap();
-        assert_eq!(
-            (h.as_str(), p, path.as_str()),
-            ("mirror.example", 80, "/a.zip")
-        );
-
-        // Absolute path stays on the current host/port.
-        let (h, p, path) = resolve_redirect("/new/path.zip", "cur.host", 8080).unwrap();
-        assert_eq!(
-            (h.as_str(), p, path.as_str()),
-            ("cur.host", 8080, "/new/path.zip")
-        );
-
-        // Absolute http URL is followed normally.
-        let (h, _, path) = resolve_redirect("http://other.host/x.zip", "cur.host", 80).unwrap();
-        assert_eq!((h.as_str(), path.as_str()), ("other.host", "/x.zip"));
-    }
-
-    #[test]
-    fn parse_status_extracts_code() {
-        assert_eq!(parse_status("HTTP/1.0 200 OK\r\nFoo: bar"), Some(200));
-        assert_eq!(parse_status("HTTP/1.1 301 Moved Permanently"), Some(301));
-        assert_eq!(parse_status("garbage"), None);
-    }
-
-    // ── New comprehensive tests ────────────────────────────────────────────────
-
-    /// find_header_end detects the \r\n\r\n separator (RFC 7230 §3 — HTTP header
-    /// terminator is CRLF CRLF). Returns the byte position of the first \r.
-    /// Mutation: searching for \n\n instead of \r\n\r\n misses the boundary.
-    #[test]
-    fn find_header_end_locates_crlfcrlf() {
-        let data = b"HTTP/1.0 200 OK\r\nContent-Length: 42\r\n\r\nbody starts here";
-        // The \r\n\r\n starts at byte 37 (after the Content-Length line).
-        let pos = find_header_end(data).expect("must find header end");
-        // body starts at pos + 4 (past the \r\n\r\n).
-        assert_eq!(
-            &data[pos + 4..],
-            b"body starts here",
-            "body must begin immediately after the \\r\\n\\r\\n boundary"
-        );
-    }
-
-    /// find_header_end returns None when there is no \r\n\r\n.
-    /// Mutation: returning Some(0) unconditionally makes this fail.
-    #[test]
-    fn find_header_end_returns_none_when_absent() {
-        let data = b"no separator here at all";
-        assert!(find_header_end(data).is_none());
-    }
-
-    /// extract_header is case-insensitive per RFC 7230 §3.2.
-    /// Mutation: using case-sensitive comparison misses "location" vs "Location".
-    #[test]
-    fn extract_header_case_insensitive() {
-        let headers = "HTTP/1.1 301 Moved\r\nlocation: http://new.host/path\r\n";
-        let val = extract_header(headers, "Location").expect("must find Location");
-        assert_eq!(val, "http://new.host/path");
-    }
-
-    /// extract_header with a missing header returns None.
-    /// Mutation: returning Some("") makes the caller proceed on a missing Location header.
-    #[test]
-    fn extract_header_missing_returns_none() {
-        let headers = "HTTP/1.0 200 OK\r\nContent-Type: text/plain\r\n";
-        assert!(extract_header(headers, "Location").is_none());
-    }
-
-    /// extract_header trims leading/trailing whitespace from the value.
-    /// RFC 7230 §3.2.6: optional whitespace around field value.
-    /// Mutation: not trimming the value keeps leading spaces in the URL.
-    #[test]
-    fn extract_header_trims_value_whitespace() {
-        let headers = "HTTP/1.1 301 Moved\r\nLocation:   /new/path  \r\n";
-        let val = extract_header(headers, "Location").unwrap();
-        assert_eq!(val, "/new/path", "value must be trimmed");
-    }
-
-    /// save() rejects data that is not a valid keydb (no recognisable entries).
-    /// Spec: entries are lines starting with "0x", "| DK", "| PK", or "| HC".
-    /// Mutation: dropping the entries==0 check lets an empty file be saved.
-    #[test]
-    fn save_rejects_empty_text() {
-        // Plain text with no valid keydb entries.
-        let garbage = b"this is not a keydb\njust random text\n";
-        assert!(
-            matches!(save(garbage), Err(Error::KeydbInvalid)),
-            "keydb without valid entries must be rejected"
-        );
-    }
-
-    /// save() accepts plain text with at least one "0x"-prefixed entry line.
-    /// Mutation: counting only "| DK" lines ignores the "0x" entry format.
-    #[test]
-    fn save_accepts_plaintext_with_0x_entries() {
-        // Minimal keydb-style file with a VUK entry (0x-prefixed).
-        let content = b"0xDEADBEEFCAFEBABE0102030405060708090A0B0C0D0E0F\n";
-        // We can't predict the HOME path in test environments without
-        // potentially writing to a real location. So only check that save()
-        // accepts this content as valid (may return KeydbWrite if dir exists
-        // but we lack permission — that still proves it passed the parse check).
-        let result = save(content);
-        // Accept either Ok (wrote successfully) or a write error (env issue),
-        // but NOT KeydbInvalid or KeydbParse.
-        match &result {
-            Ok(_) => {}
-            Err(Error::KeydbWrite { .. }) => {}
-            Err(e) => panic!("unexpected error for valid keydb content: {:?}", e),
-        }
-    }
-
-    /// save() accepts content with "| DK" entries (device-key table format).
-    /// Mutation: only accepting "0x" lines rejects DK-format keydb files.
-    #[test]
-    fn save_accepts_pipe_dk_entry_format() {
-        let content = b"| DK 0102030405060708 | 0102030405060708090a0b0c0d0e0f10 |\n";
-        let result = save(content);
-        match &result {
-            Ok(_) => {}
-            Err(Error::KeydbWrite { .. }) => {}
-            Err(e) => panic!("unexpected error for DK-format entry: {:?}", e),
-        }
-    }
-
-    /// save() accepts content with "| PK" entries (processing-key format).
-    /// Mutation: not including "| PK" in the filter rejects PK-format keydb files.
-    #[test]
-    fn save_accepts_pipe_pk_entry_format() {
-        let content = b"| PK 0102030405060708090a0b0c0d0e0f10 |\n";
-        let result = save(content);
-        match &result {
-            Ok(_) => {}
-            Err(Error::KeydbWrite { .. }) => {}
-            Err(e) => panic!("unexpected error for PK-format entry: {:?}", e),
-        }
-    }
-
-    /// save() accepts content with "| HC" entries (host certificate format).
-    /// Mutation: not including "| HC" in the filter rejects HC-format keydb files.
-    #[test]
-    fn save_accepts_pipe_hc_entry_format() {
-        let content = b"| HC 0102030405060708090a0b0c0d0e0f10 |\n";
-        let result = save(content);
-        match &result {
-            Ok(_) => {}
-            Err(Error::KeydbWrite { .. }) => {}
-            Err(e) => panic!("unexpected error for HC-format entry: {:?}", e),
-        }
-    }
-
-    /// save() recognises gzip-compressed input (magic bytes 0x1f 0x8b).
-    /// Spec: gzip format magic is 0x1F 0x8B (RFC 1952 §2.3.1).
-    /// Mutation: treating gzip magic as plain text fails to decompress.
-    #[test]
-    fn save_recognises_gzip_magic() {
-        // Truncated gzip (header only, no valid body) — must not be treated as
-        // plain text (no KeydbInvalid about entries), but as a parse error.
-        let bad_gz = [0x1f, 0x8b, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03];
-        let result = save(&bad_gz);
-        // A truncated gzip is either KeydbParse (decompression error) or
-        // KeydbInvalid (decompressed to empty). Must not be Ok.
-        assert!(result.is_err(), "truncated gzip must not be accepted");
-        // Crucially: must NOT be a plain-text UTF-8 error — gzip magic is not UTF-8.
-        match result.unwrap_err() {
-            Error::KeydbParse | Error::KeydbInvalid => {}
-            e => panic!("wrong error kind for truncated gzip: {:?}", e),
-        }
-    }
-
-    /// save() recognises ZIP magic bytes PK\x03\x04 and routes to extract_zip.
-    /// Spec: ZIP local file header signature is 0x50 0x4B 0x03 0x04 (PKZIP APPNOTE §4.3.6).
-    /// A truncated ZIP must error, but NOT as plain UTF-8 text.
-    /// Mutation: checking gzip magic before ZIP magic means ZIP files are
-    ///           fed to the gzip decoder and produce the wrong error.
-    #[test]
-    fn save_recognises_zip_magic() {
-        // Valid ZIP magic followed by garbage — must be routed to extract_zip.
-        let bad_zip = b"PK\x03\x04garbage that is not a real zip";
-        let result = save(bad_zip);
-        assert!(result.is_err(), "invalid zip must be rejected");
-        // Must be a parse error, not a UTF-8 error.
-        match result.unwrap_err() {
-            Error::KeydbParse | Error::KeydbInvalid => {}
-            e => panic!("wrong error for bad zip: {:?}", e),
-        }
-    }
-
-    /// read_capped_to_string rejects data exceeding MAX_KEYDB_BYTES.
-    /// Spec: doc says "Returns Error::KeydbInvalid if the input exceeds the cap".
-    /// Mutation: removing the length check accepts decompression bombs.
-    #[test]
-    fn read_capped_to_string_rejects_oversized_input() {
-        // Build a reader that reports it has more data than the cap.
-        // We use a Cursor with MAX_KEYDB_BYTES + 1 bytes of content.
-        let too_big = vec![b'A'; (MAX_KEYDB_BYTES + 1) as usize];
-        let cursor = std::io::Cursor::new(too_big);
-        let result = read_capped_to_string(cursor);
-        assert!(
-            matches!(result, Err(Error::KeydbInvalid)),
-            "oversized input must yield KeydbInvalid, got: {:?}",
-            result
-        );
-    }
-
-    /// read_capped_to_string returns KeydbParse (not KeydbInvalid) for
-    /// non-UTF-8 input. Guards the doc/behavior contract: KeydbInvalid is
-    /// reserved for the size-cap violation, a decode failure is a parse error.
-    #[test]
-    fn read_capped_to_string_non_utf8_yields_parse() {
-        // 0xFF is never a valid UTF-8 byte.
-        let cursor = std::io::Cursor::new(vec![0xFFu8, 0xFE, 0xFD]);
-        let result = read_capped_to_string(cursor);
-        assert!(
-            matches!(result, Err(Error::KeydbParse)),
-            "non-UTF-8 input must yield KeydbParse, got: {:?}",
-            result
-        );
-    }
-
-    /// read_capped_to_string accepts exactly MAX_KEYDB_BYTES (at-cap is allowed).
-    /// Spec: doc says "Read one byte past the cap so an exactly-at-cap stream is accepted."
-    /// Mutation: using `>=` instead of `>` in the length check rejects valid at-cap files.
-    #[test]
-    fn read_capped_to_string_accepts_at_cap_size() {
-        let at_cap = vec![b'A'; MAX_KEYDB_BYTES as usize];
-        let cursor = std::io::Cursor::new(at_cap);
-        let result = read_capped_to_string(cursor);
-        assert!(result.is_ok(), "exactly MAX_KEYDB_BYTES must be accepted");
-    }
-
-    /// parse_status returns None for an empty/malformed status line (not a
-    /// meaningless 0). The call site maps None to Error::KeydbParse.
-    #[test]
-    fn parse_status_empty_input_returns_none() {
-        assert_eq!(parse_status(""), None);
-        assert_eq!(parse_status("\r\n"), None);
-    }
-
-    /// Regression: set_read_timeout / set_write_timeout failures must surface as
-    /// KeydbConnect, not be silently swallowed.
-    ///
-    /// We can't easily synthesise a TcpStream whose set_*timeout syscall fails
-    /// without a platform-specific socket hack, so instead we verify that the
-    /// error-mapping expression itself is correct: if set_read_timeout were to
-    /// fail for a given host, the result must be Err(KeydbConnect { host }).
-    ///
-    /// The test constructs the exact Err value the code would return and asserts
-    /// it is KeydbConnect (not, say, silently Ok or a different variant). This
-    /// pins the variant selection so a future refactor that changes the `.ok()`
-    /// pattern back would need to update this test as well.
-    #[test]
-    fn timeout_set_failure_maps_to_keydb_connect() {
-        // Simulate what the propagated error looks like.
-        let host = "hostile.example.com".to_string();
-        // The io::Error that set_read_timeout would return on failure.
-        let io_err = std::io::Error::from(std::io::ErrorKind::InvalidInput);
-        // Apply the same map_err the production code uses.
-        let result: Result<()> =
-            Err(io_err).map_err(|_| Error::KeydbConnect { host: host.clone() });
-        assert!(
-            matches!(result, Err(Error::KeydbConnect { host: ref h }) if h == "hostile.example.com"),
-            "set_timeout failure must map to KeydbConnect, got: {:?}",
-            result
-        );
-    }
-
-    /// Regression: http_get to an unreachable host returns KeydbConnect, not a hang.
-    /// This exercises the connect_timeout path (and thus confirms the overall
-    /// error-propagation chain is wired); the timeout-set propagation is exercised
-    /// by the unit test above.
-    ///
-    /// Uses port 1 on localhost, which is reserved/unassigned and virtually never
-    /// listening. connect_timeout with NET_TIMEOUT will refuse or time out quickly.
-    /// We only assert the error variant, not the host field, since the OS may
-    /// resolve the address differently.
-    #[test]
-    fn http_get_unreachable_host_returns_keydb_connect() {
-        // Port 1 on loopback — almost always refused immediately.
-        let result = http_get("http://127.0.0.1:1/keydb.zip");
-        // Must be an Err; KeydbConnect is expected for a TCP-level failure.
-        // KeydbParse or KeydbHttp would indicate the wrong error path.
-        assert!(result.is_err(), "unreachable host must fail");
-        match result.unwrap_err() {
-            Error::KeydbConnect { .. } => {}
-            e => panic!("expected KeydbConnect for unreachable host, got: {:?}", e),
-        }
-    }
-
-    /// Regression: when the server accepts the connection but closes it before
-    /// sending complete HTTP headers (the `n == 0` byte-read path), that is a
-    /// connection/protocol-level fault. It must surface as KeydbConnect (E8000),
-    /// NOT KeydbParse (E8004) — the keydb content was never received, let alone
-    /// malformed.
-    #[test]
-    fn http_get_server_drops_before_headers_returns_keydb_connect() {
-        use std::io::Read as _;
-        use std::net::TcpListener;
-
-        let listener = TcpListener::bind("127.0.0.1:0").unwrap();
-        let addr = listener.local_addr().unwrap();
-
-        let server = std::thread::spawn(move || {
-            if let Ok((mut sock, _)) = listener.accept() {
-                // Drain the request so the client's write_all completes, then
-                // drop the socket without writing any response. The client's
-                // header read then returns n == 0.
-                let mut buf = [0u8; 512];
-                let _ = sock.read(&mut buf);
-                drop(sock);
-            }
-        });
-
-        let url = format!("http://127.0.0.1:{}/keydb.zip", addr.port());
-        let result = http_get(&url);
-        server.join().unwrap();
-
-        assert!(result.is_err(), "dropped connection must fail");
-        match result.unwrap_err() {
-            Error::KeydbConnect { .. } => {}
-            e => panic!("expected KeydbConnect for dropped connection, got: {:?}", e),
-        }
-    }
-}

src/keysource.rs

@@ -100,6 +100,12 @@ pub struct DiscInputsCtx<'a> {
 impl<'a> DiscInputsCtx<'a> {
     /// Build a context over `inputs`, parsing the encrypted title keys at the
     /// stride for AACS major `version_u8` (1 = V10, else V20/V21).
+    ///
+    /// A present-but-malformed `unit_key_ro` (truncated / wrong magic / wrong
+    /// stride) parses to an empty key set, so a later [`Self::enc_title_keys`]
+    /// returns `Ok(&[])` indistinguishably from a disc that legitimately has no
+    /// title keys — the parse failure is swallowed here, not surfaced as an
+    /// error.
     pub fn new(inputs: &'a DiscInputs, version_u8: u8) -> Self {
         use crate::aacs::{AacsVersion, parse_unit_key_ro};
         let enc_keys = if inputs.unit_key_ro.is_empty() {

src/labels/mod.rs

@@ -634,6 +634,7 @@ fn append_clpi_orphans(
     reader: &mut dyn SectorSource,
     udf: &UdfFs,
 ) -> usize {
+    use crate::consts::coding_type as c;
     // Index existing labels by PID — but StreamLabel doesn't carry
     // PID. Index by (type, language, codec_hint) tuple instead; this
     // is fuzzier than PID matching but the only signal available
@@ -679,9 +680,11 @@ fn append_clpi_orphans(
             // Interactive Graphics (BD-J menu overlay), NOT a user-facing
             // subtitle — skip it, matching the MPLS path which drops IG.
             let stype = match s.coding_type {
-                0x80..=0x86 | 0xA1 | 0xA2 => StreamLabelType::Audio,
-                0x90 => StreamLabelType::Subtitle,
-                _ => continue, // 0x91 IG / video / unknown — skip
+                c::LPCM..=c::DTS_HD_MA | c::AC3_PLUS_SECONDARY | c::DTS_HD_SECONDARY => {
+                    StreamLabelType::Audio
+                }
+                c::PG => StreamLabelType::Subtitle,
+                _ => continue, // IG / video / unknown — skip
             };
             // Same dedup logic as MPLS: normalize language, build codec
             // hint, check against existing label set.
@@ -1121,9 +1124,10 @@ mod registry_tests {
             // and as a marker for "these parsers exist."
             let _ = (name, detect, parse);
         }
-        // The loop above touches every registry entry; iterating a non-empty
-        // fixed-size array is the assertion (a `.is_empty()` check would be
-        // const-folded). The test fails to compile if the tuple shape changes.
+        // The loop above touches every registry entry. The non-empty
+        // invariant is covered separately by `parsers_registry_order_locked`,
+        // whose assert_eq! on the expected order fails if PARSERS is empty.
+        // This test fails to compile if the tuple shape changes.
     }
 }
 

src/labels/mpls_universal.rs

@@ -227,21 +227,22 @@ pub(crate) fn language_display_name(iso: &str) -> String {
 /// bytes (the table covers everything the spec defines, but unknown
 /// values are still possible on malformed discs).
 pub(crate) fn codec_name(coding_type: u8) -> &'static str {
+    use crate::consts::coding_type as c;
     match coding_type {
-        0x02 => "MPEG-2",
-        0x1B => "H.264",
-        0x24 => "HEVC",
-        0x80 => "LPCM",
-        0x81 => "AC-3",
-        0x82 => "DTS",
-        0x83 => "TrueHD",
-        0x84 => "AC-3+",
-        0x85 => "DTS-HD HR", // BD-ROM Part 3-1: 0x85 = DTS-HD High Resolution
-        0x86 => "DTS-HD MA",
-        0x90 => "PG",
-        0x91 => "IG",
-        0xA1 => "AC-3+ Secondary",
-        0xA2 => "DTS-HD Secondary",
+        c::MPEG2_VIDEO => "MPEG-2",
+        c::H264 => "H.264",
+        c::HEVC => "HEVC",
+        c::LPCM => "LPCM",
+        c::AC3 => "AC-3",
+        c::DTS => "DTS",
+        c::TRUEHD => "TrueHD",
+        c::AC3_PLUS => "AC-3+",
+        c::DTS_HD_HR => "DTS-HD HR", // BD-ROM Part 3-1: 0x85 = DTS-HD High Resolution
+        c::DTS_HD_MA => "DTS-HD MA",
+        c::PG => "PG",
+        c::IG => "IG",
+        c::AC3_PLUS_SECONDARY => "AC-3+ Secondary",
+        c::DTS_HD_SECONDARY => "DTS-HD Secondary",
         _ => "",
     }
 }

src/lib.rs

@@ -85,6 +85,20 @@
 //! | E8xxx | Keydb errors (fetch, parse, load) |
 //! | E9xxx | Stream / mux errors (URL, PES, pipeline) |
 
+/// Single source of truth for every freemkv version surface.
+///
+/// `FREEMKV_VERSION` is the package version, overridable at build time via the
+/// `FREEMKV_BUILD_LABEL` env (see `build.rs`); `GIT_SUFFIX` is the git short
+/// hash. The CLI's `--version`, the MKV muxing/writing-application field, and
+/// the FVI generator tag all derive from these two consts, so a binary reports
+/// the exact same label it stamps into the files it produces — no split-brain
+/// where an MKV claims one version and the binary another.
+pub const VERSION_LABEL: &str = concat!(env!("FREEMKV_VERSION"), env!("GIT_SUFFIX"));
+
+/// The muxing/writing-application string written into MKV output
+/// (`"freemkv <version> (g<hash>)"`).
+pub const MUX_APP: &str = concat!("freemkv ", env!("FREEMKV_VERSION"), env!("GIT_SUFFIX"));
+
 pub mod aacs;
 pub(crate) mod clpi;
 pub mod consts;
@@ -93,13 +107,13 @@ pub mod decrypt;
 pub mod diag;
 pub mod disc;
 pub mod drive;
+pub mod dvdnav;
 pub mod error;
 pub mod event;
 pub mod halt;
 pub(crate) mod identity;
 pub(crate) mod ifo;
 pub mod io;
-pub mod keydb;
 pub mod keysource;
 pub mod labels;
 pub(crate) mod mpls;

src/mpls.rs

@@ -179,7 +179,9 @@ pub fn parse(data: &[u8]) -> Result<Playlist> {
             }
             // PG subtitles
             for _ in 0..n_pg {
-                if let Some((entry, next)) = parse_stream_entry(item, spos, STREAM_CATEGORY_PG_SUBTITLE) {
+                if let Some((entry, next)) =
+                    parse_stream_entry(item, spos, STREAM_CATEGORY_PG_SUBTITLE)
+                {
                     streams.push(entry);
                     spos = next;
                 } else {
@@ -196,7 +198,9 @@ pub fn parse(data: &[u8]) -> Result<Playlist> {
             }
             // Secondary audio
             for _ in 0..n_sec_audio {
-                if let Some((mut entry, next)) = parse_stream_entry(item, spos, STREAM_CATEGORY_AUDIO) {
+                if let Some((mut entry, next)) =
+                    parse_stream_entry(item, spos, STREAM_CATEGORY_AUDIO)
+                {
                     entry.stream_type = 5;
                     entry.secondary = true;
                     streams.push(entry);
@@ -213,7 +217,9 @@ pub fn parse(data: &[u8]) -> Result<Playlist> {
             }
             // Secondary video (PiP)
             for _ in 0..n_sec_video {
-                if let Some((mut entry, next)) = parse_stream_entry(item, spos, STREAM_CATEGORY_VIDEO) {
+                if let Some((mut entry, next)) =
+                    parse_stream_entry(item, spos, STREAM_CATEGORY_VIDEO)
+                {
                     entry.stream_type = 6;
                     entry.secondary = true;
                     streams.push(entry);
@@ -240,7 +246,9 @@ pub fn parse(data: &[u8]) -> Result<Playlist> {
             }
             // Secondary PG (PiP subtitles) — must consume to keep spos aligned
             for _ in 0..n_pip_pg {
-                if let Some((mut entry, next)) = parse_stream_entry(item, spos, STREAM_CATEGORY_PG_SUBTITLE) {
+                if let Some((mut entry, next)) =
+                    parse_stream_entry(item, spos, STREAM_CATEGORY_PG_SUBTITLE)
+                {
                     entry.secondary = true;
                     streams.push(entry);
                     // Skip reference data: num_refs(1) + reserved(1) + refs + padding
@@ -256,7 +264,9 @@ pub fn parse(data: &[u8]) -> Result<Playlist> {
             }
             // Dolby Vision enhancement layer
             for _ in 0..n_dv {
-                if let Some((mut entry, next)) = parse_stream_entry(item, spos, STREAM_CATEGORY_VIDEO) {
+                if let Some((mut entry, next)) =
+                    parse_stream_entry(item, spos, STREAM_CATEGORY_VIDEO)
+                {
                     entry.stream_type = 7;
                     entry.secondary = true;
                     streams.push(entry);
@@ -333,6 +343,7 @@ const STREAM_CATEGORY_PG_SUBTITLE: u8 = 3;
 const STREAM_CATEGORY_IG: u8 = 4;
 
 fn parse_stream_entry(item: &[u8], pos: usize, stream_type: u8) -> Option<(StreamEntry, usize)> {
+    use crate::consts::coding_type as c;
     if pos + 2 > item.len() {
         return None;
     }
@@ -387,22 +398,27 @@ fn parse_stream_entry(item: &[u8], pos: usize, stream_type: u8) -> Option<(Strea
     let mut color_space_val = 0u8;
     let mut language = String::new();
 
+    // `stream_type` here is the STN category passed by the caller, which is
+    // only ever a primary category (VIDEO/AUDIO/PG_SUBTITLE/IG). Secondary
+    // audio/video and the DV enhancement layer are parsed through their
+    // matching primary category (identical attribute layout) and re-tagged by
+    // the caller after this returns, so there are no secondary arms here.
     match stream_type {
-        1 => {
+        STREAM_CATEGORY_VIDEO => {
             // Video: coding_type(1) + format_rate(1) + [hdr_info(1) if HEVC]
             if sa.len() >= 2 {
                 video_format = (sa[1] >> 4) & 0x0F;
                 video_rate = sa[1] & 0x0F;
             }
-            if coding_type == 0x24 && sa.len() > 2 {
+            if coding_type == c::HEVC && sa.len() > 2 {
                 dynamic_range = (sa[2] >> 4) & 0x0F;
                 color_space_val = sa[2] & 0x0F;
             }
         }
-        2 => {
+        STREAM_CATEGORY_AUDIO => {
             // Audio: coding_type(1) + format_rate(1) + language(3)
-            // Exception: PGS (0x90/0x91) in audio slot uses PG layout: coding_type(1) + language(3)
-            if coding_type == 0x90 || coding_type == 0x91 {
+            // Exception: PG/IG in an audio slot uses PG layout: coding_type(1) + language(3)
+            if coding_type == c::PG || coding_type == c::IG {
                 if sa.len() >= 4 {
                     language = String::from_utf8_lossy(&sa[1..4]).to_string();
                 }
@@ -416,35 +432,14 @@ fn parse_stream_entry(item: &[u8], pos: usize, stream_type: u8) -> Option<(Strea
                 }
             }
         }
-        3 => {
+        STREAM_CATEGORY_PG_SUBTITLE => {
             // PG: coding_type(1) + language(3).
-            // IG (type 4) is parsed only to advance spos and is then
-            // discarded by the caller, so it deliberately has no arm here.
+            // IG is parsed only to advance spos and is then discarded by the
+            // caller, so it deliberately has no arm here.
             if sa.len() >= 4 {
                 language = String::from_utf8_lossy(&sa[1..4]).to_string();
             }
         }
-        5 => {
-            // Secondary audio: same as primary audio
-            if sa.len() >= 2 {
-                audio_format = (sa[1] >> 4) & 0x0F;
-                audio_rate = sa[1] & 0x0F;
-            }
-            if sa.len() >= 5 {
-                language = String::from_utf8_lossy(&sa[2..5]).to_string();
-            }
-        }
-        6 | 7 => {
-            // Secondary video: same as primary video
-            if sa.len() >= 2 {
-                video_format = (sa[1] >> 4) & 0x0F;
-                video_rate = sa[1] & 0x0F;
-            }
-            if coding_type == 0x24 && sa.len() > 2 {
-                dynamic_range = (sa[2] >> 4) & 0x0F;
-                color_space_val = sa[2] & 0x0F;
-            }
-        }
         _ => {}
     }
 

src/mux/codec/coding.rs

@@ -181,8 +181,13 @@ impl PictureInfo {
         match self.detail {
             CodingDetail::Mpeg2(m) => {
                 if !m.frame_picture {
-                    // A single field picture is inherently interlaced; the
-                    // top_field_first bit names which field this picture is.
+                    // A single field picture is inherently interlaced. Which
+                    // field it actually codes is given by picture_structure
+                    // (top/bottom), not by top_field_first — §6.3.10 constrains
+                    // top_field_first to 0 for field pictures, so it is not the
+                    // spec source here. picture_structure is not retained on
+                    // this carrier, so top_field_first is used only as the lone
+                    // field hint available (best-effort, not spec-derived).
                     Some(if m.top_field_first {
                         FieldOrder::Tff
                     } else {
@@ -203,8 +208,10 @@ impl PictureInfo {
     /// Number of field-display periods this picture occupies — the basis for
     /// soft-telecine (2:3 pulldown) timing. MPEG-2 (ISO/IEC 13818-2 §6.3.10,
     /// ffmpeg `nb_fields = repeat_pict + 2`): a field picture occupies 1 field,
-    /// a normal frame 2, a `repeat_first_field` frame 3 (or 4/6 in a progressive
-    /// sequence). Codecs without pulldown signalling report the normal 2 fields.
+    /// a normal frame 2, a `repeat_first_field` progressive-frame 3 (or 4/6 in a
+    /// progressive sequence); an rff bit on a non-progressive interlaced frame is
+    /// spec-forbidden (§6.3.10) and is treated as 2. Codecs without pulldown
+    /// signalling report the normal 2 fields.
     pub fn nb_fields(&self) -> u8 {
         match self.detail {
             CodingDetail::Mpeg2(m) => {

src/mux/codec/dts.rs

@@ -5,14 +5,14 @@
 //! Buffers across PES boundaries so frames spanning two PES packets
 //! are emitted complete.
 
+use super::startcode::BitReader;
 use super::{CodecParser, Frame, PesPacket, pts_to_ns};
 
 const DTS_CORE_SYNC: [u8; 4] = [0x7F, 0xFE, 0x80, 0x01];
-/// DTS-HD extension substream syncword. The parser delimits an access unit by
-/// the next CORE sync (so every extension between two cores is captured), and
-/// never needs to locate or size the extension itself — so this is referenced
-/// only by the tests that synthesize extension substreams.
-#[cfg(test)]
+/// DTS-HD extension substream syncword. An access unit is delimited by the next
+/// CORE sync; the parser locates and exactly sizes each extension substream (via
+/// `exss_frame_size`) so a false core sync inside the EXSS payload can't split
+/// the AU and truncate the lossless extension.
 const DTS_HD_EXT_SYNC: [u8; 4] = [0x64, 0x58, 0x20, 0x25];
 
 /// DTS / DTS-HD elementary-stream parser. Buffers DTS across PES boundaries so
@@ -336,11 +336,108 @@ enum NextCore {
 /// match the core syncword, so each candidate is validated by decoding its
 /// core size: a match whose decoded size is implausible (< MIN_CORE_FRAME_BYTES
 /// or > MAX_AU_BYTES) is a false sync and is skipped, continuing the search.
+/// Both DTS syncwords (core `0x7FFE8001`, EXSS `0x64582025`) are 32-bit words.
+const SYNCWORD_BYTES: usize = DTS_CORE_SYNC.len();
+
+/// DTS-HD extension-substream (EXSS) header field bit widths (ETSI TS 102 114,
+/// ExtSS header). `bHeaderSizeType` selects the short form (`nuExtSSHeaderSize`
+/// 8 bits, `nuExtSSFsize` 16 bits) or, for larger substreams, the long form
+/// (12 / 20 bits).
+const EXSS_USER_DEFINED_BITS: u32 = 8;
+const EXSS_INDEX_BITS: u32 = 2;
+const EXSS_HEADER_SIZE_TYPE_BITS: u32 = 1;
+const EXSS_HDRSIZE_BITS_SHORT: u32 = 8;
+const EXSS_FSIZE_BITS_SHORT: u32 = 16;
+const EXSS_HDRSIZE_BITS_LONG: u32 = 12;
+const EXSS_FSIZE_BITS_LONG: u32 = 20;
+/// `bHeaderSizeType == 1` selects the long-form field widths.
+const EXSS_HEADER_SIZE_TYPE_LONG: u32 = 1;
+/// Bytes that must be buffered to read the EXSS size fields in the worst case
+/// (long form): the 4-byte sync plus the bits up through `nuExtSSFsize`.
+const EXSS_HEADER_MIN_BYTES: usize = SYNCWORD_BYTES
+    + (EXSS_USER_DEFINED_BITS
+        + EXSS_INDEX_BITS
+        + EXSS_HEADER_SIZE_TYPE_BITS
+        + EXSS_HDRSIZE_BITS_LONG
+        + EXSS_FSIZE_BITS_LONG)
+        .div_ceil(u8::BITS) as usize;
+
+/// DTS-HD extension substream (EXSS) total byte size — INCLUDING the
+/// `0x64582025` syncword — read precisely from its header. `buf` must begin with
+/// `DTS_HD_EXT_SYNC`. `None` when the size fields aren't fully buffered.
+///
+/// `nuExtSSFsize` is the total frame size in bytes minus one. Parsing it lets the
+/// AU framer skip the extension by its exact length instead of scanning its
+/// (arbitrary) payload for a core sync.
+fn exss_frame_size(buf: &[u8]) -> Option<usize> {
+    if buf.len() < EXSS_HEADER_MIN_BYTES {
+        return None;
+    }
+    let mut r = BitReader::new(&buf[SYNCWORD_BYTES..]);
+    let _user = r.read_bits(EXSS_USER_DEFINED_BITS)?; // nUserDefinedBits
+    let _idx = r.read_bits(EXSS_INDEX_BITS)?; // nExtSSIndex
+    let large = r.read_bits(EXSS_HEADER_SIZE_TYPE_BITS)? == EXSS_HEADER_SIZE_TYPE_LONG;
+    let (hbits, fbits) = if large {
+        (EXSS_HDRSIZE_BITS_LONG, EXSS_FSIZE_BITS_LONG)
+    } else {
+        (EXSS_HDRSIZE_BITS_SHORT, EXSS_FSIZE_BITS_SHORT)
+    };
+    let _hdr = r.read_bits(hbits)?; // nuExtSSHeaderSize (not needed for framing)
+    let fsize_minus_one = r.read_bits(fbits)?; // nuExtSSFsize = total bytes - 1
+    Some(fsize_minus_one as usize + 1)
+}
+
+/// Offset where the current access unit ends (the start of the next core
+/// frame). The AU is the core frame plus its trailing DTS-HD extension
+/// substreams, which are skipped PRECISELY by their declared size — so a chance
+/// core syncword inside the XLL lossless payload can never be mistaken for the
+/// next AU boundary (the bug that truncated the extension and produced the
+/// "Failed to decode block code(s)" class). Falls back to the heuristic core-sync
+/// scan only when an extension can't be sized (malformed / truncated input).
 fn next_core_boundary(buf: &[u8], core_size: usize) -> NextCore {
-    let mut from = core_size;
+    let mut pos = core_size;
+    loop {
+        if buf.len() < pos + SYNCWORD_BYTES {
+            return NextCore::NeedMore; // need a syncword to identify the next chunk
+        }
+        if buf[pos..].starts_with(&DTS_HD_EXT_SYNC) {
+            match exss_frame_size(&buf[pos..]) {
+                Some(sz) if sz >= SYNCWORD_BYTES => {
+                    if buf.len() < pos + sz {
+                        return NextCore::NeedMore; // extension not fully buffered
+                    }
+                    pos += sz; // skip the whole extension substream precisely
+                }
+                // Couldn't size it (truncated/garbage header) — heuristic fallback.
+                _ => return scan_for_next_core(buf, pos),
+            }
+        } else if buf[pos..].starts_with(&DTS_CORE_SYNC) {
+            // The bytes right after the precisely-skipped extensions are the next
+            // core frame — the AU boundary.
+            if buf.len() - pos < CORE_HEADER_MIN_BYTES {
+                return NextCore::NeedMore;
+            }
+            let sz = dts_core_frame_size(&buf[pos..]);
+            if (MIN_CORE_FRAME_BYTES..=MAX_AU_BYTES).contains(&sz) {
+                return NextCore::Found(pos);
+            }
+            return scan_for_next_core(buf, pos); // implausible core here — fall back
+        } else {
+            // Neither a known extension nor a core sync at the precise boundary
+            // (padding / junk) — fall back to the heuristic scan.
+            return scan_for_next_core(buf, pos);
+        }
+    }
+}
+
+/// Heuristic fallback (the pre-fix behaviour): scan forward for the next core
+/// syncword whose decoded size is plausible. Used only when precise extension
+/// skipping can't proceed; a chance core syncword in extension payload usually
+/// decodes to an implausible size and is skipped.
+fn scan_for_next_core(buf: &[u8], from: usize) -> NextCore {
+    let mut from = from;
     while let Some(rel) = find_sync(&buf[from..], &DTS_CORE_SYNC) {
         let pos = from + rel;
-        // Need the candidate's core header to judge it.
         if buf.len() - pos < CORE_HEADER_MIN_BYTES {
             return NextCore::NeedMore;
         }
@@ -348,8 +445,7 @@ fn next_core_boundary(buf: &[u8], core_size: usize) -> NextCore {
         if (MIN_CORE_FRAME_BYTES..=MAX_AU_BYTES).contains(&sz) {
             return NextCore::Found(pos);
         }
-        // False sync inside extension payload — skip it and keep searching.
-        from = pos + 4;
+        from = pos + SYNCWORD_BYTES;
     }
     NextCore::None
 }
@@ -399,6 +495,55 @@ mod tests {
         data
     }
 
+    /// A real DTS-HD EXSS substream of `total` bytes (short header form), with an
+    /// optional false DTS core syncword embedded in its payload (decoding to a
+    /// plausible core size) — to prove precise sizing, not a payload scan, bounds
+    /// the extension.
+    fn make_exss(total: usize, false_core_at: Option<usize>) -> Vec<u8> {
+        let mut d = vec![0u8; total];
+        d[0..4].copy_from_slice(&DTS_HD_EXT_SYNC);
+        // Short form: all header fields 0 except nuExtSSFsize = total - 1, laid
+        // out at bit 19 after the sync (byte 6 low 5 bits, byte 7, byte 8 top 3).
+        let fsize = (total - 1) as u32;
+        d[6] = ((fsize >> 11) & 0x1F) as u8;
+        d[7] = ((fsize >> 3) & 0xFF) as u8;
+        d[8] = ((fsize & 0x07) << 5) as u8;
+        if let Some(at) = false_core_at {
+            d[at..at + 4].copy_from_slice(&DTS_CORE_SYNC);
+            let fcs = 512u32 - 1; // decode to a plausible core size — fools the heuristic
+            d[at + 5] = (d[at + 5] & 0xFC) | ((fcs >> 12) & 0x03) as u8;
+            d[at + 6] = ((fcs >> 4) & 0xFF) as u8;
+            d[at + 7] = (d[at + 7] & 0x0F) | (((fcs & 0x0F) << 4) as u8);
+        }
+        d
+    }
+
+    #[test]
+    fn plausible_false_core_sync_inside_real_exss_does_not_split_au() {
+        // EXSS size parse round-trips.
+        assert_eq!(exss_frame_size(&make_exss(600, None)), Some(600));
+
+        // AU = core(512) + a REAL EXSS substream whose XLL payload embeds a DTS
+        // core syncword decoding to a plausible size (512). The heuristic-only
+        // framer would split here and truncate the lossless extension (the
+        // Dunkirk `dca` "Failed to decode block code(s)" class). Precise EXSS
+        // sizing spans the whole extension to the REAL next core.
+        let core = make_dts_core(512);
+        let exss = make_exss(600, Some(40));
+        let next = make_dts_core(512);
+        let mut buf = core.clone();
+        buf.extend_from_slice(&exss);
+        buf.extend_from_slice(&next);
+
+        assert!(
+            matches!(
+                next_core_boundary(&buf, core.len()),
+                NextCore::Found(end) if end == core.len() + exss.len()
+            ),
+            "AU must end at the REAL next core (after the full EXSS), not the false sync inside it"
+        );
+    }
+
     #[test]
     fn parse_empty_pes() {
         let mut parser = DtsParser::new();

src/mux/codec/h264.rs

@@ -325,7 +325,8 @@ impl CodecParser for H264Parser {
 }
 
 /// Parse `(chroma_format_idc, bit_depth_luma_minus8, bit_depth_chroma_minus8)` from
-/// a High-Profile SPS NAL (profile_idc ∈ {100, 110, 122, 144}).
+/// a High-Profile SPS NAL (profile_idc ∈ `HIGH_PROFILES` — the 14 chroma/bit-depth
+/// extended profiles `codec_private` invokes this for).
 ///
 /// SPS RBSP layout (ITU-T H.264 §7.3.2.1.1) up to the fields we need:
 ///   byte 0        NAL header (already known to be type 7)

src/mux/codec/hevc.rs

@@ -56,10 +56,7 @@ fn hevc_num_extra_slice_header_bits(pps_nal: &[u8]) -> Option<u32> {
     br.read_ue()?; // pps_pic_parameter_set_id
     br.read_ue()?; // pps_seq_parameter_set_id
     br.skip_bits(2)?; // dependent_slice_segments_enabled_flag, output_flag_present_flag
-    let mut n = 0u32;
-    for _ in 0..3 {
-        n = (n << 1) | br.read_bit()?;
-    }
+    let n = br.read_bits(3)?;
     Some(n)
 }
 

src/mux/codec/truehd.rs

@@ -138,7 +138,18 @@ impl CodecParser for TrueHdParser {
         // the next PES legitimately begins a new AU and seeds the base.
         if self.buf.is_empty() {
             if let Some(pts) = pes.pts {
-                self.next_pts_ns = pts_to_ns(pts);
+                // Resync to the authoritative PES PTS, but NEVER snap backward.
+                // TrueHD AUs are a fixed sample count (40 @ 48 kHz), so the
+                // per-AU `+AU_DURATION_NS` cadence is sample-accurate — more so
+                // than the disc's per-PES PTS, which carries the source muxer's
+                // own rounding jitter. When the buffer empties exactly on a PES
+                // boundary and that PES's PTS lands a few ticks *below* the
+                // running cadence, an unconditional reset would set the next
+                // AU's timestamp below the AU just emitted, producing the
+                // non-monotonic block timestamps a muxer rejects. Clamp to the
+                // running position so output stays strictly monotonic; a
+                // genuine forward gap/discontinuity is still adopted.
+                self.next_pts_ns = self.next_pts_ns.max(pts_to_ns(pts));
             }
         }
 
@@ -449,6 +460,36 @@ mod tests {
         assert_eq!(frames[1].pts_ns - frames[0].pts_ns, AU_DURATION_NS);
     }
 
+    #[test]
+    fn pes_pts_lagging_the_au_cadence_never_emits_backward() {
+        // Regression: the per-AU cadence is sample-accurate, but a PES boundary
+        // can carry a PTS that lags it slightly (source-muxer rounding jitter).
+        // When the buffer empties exactly on that boundary, an unconditional
+        // reset to the PES PTS snapped the next AU's timestamp BELOW the AU just
+        // emitted — the non-monotonic block timestamps a muxer rejects (the
+        // Top Gun / Dune: Part Two case). The reset must clamp forward-only.
+        let mut parser = TrueHdParser::new();
+        let au = make_truehd_unit(100);
+        // PES1: three complete AUs at pts 90000 — buffer empties, cadence runs
+        // ahead to 90000_ns + 3*AU_DURATION_NS.
+        let mut d1 = au.clone();
+        d1.extend_from_slice(&au);
+        d1.extend_from_slice(&au);
+        let f1 = parser.parse(&make_pes(d1, Some(90000)));
+        assert_eq!(f1.len(), 3);
+        let last1 = f1.last().unwrap().pts_ns;
+        // PES2's PTS (90001) maps to fewer ns than the running cadence — pre-fix
+        // this snapped backward.
+        let f2 = parser.parse(&make_pes(au.clone(), Some(90001)));
+        assert_eq!(f2.len(), 1);
+        assert!(
+            f2[0].pts_ns >= last1,
+            "AU pts must not go backward when PES PTS lags the cadence: got {} after {}",
+            f2[0].pts_ns,
+            last1
+        );
+    }
+
     #[test]
     fn skip_interleaved_ac3() {
         let mut parser = TrueHdParser::new();

src/mux/demux_sink.rs

@@ -273,8 +273,8 @@ const SUP_HEADER_LEN: usize = SUP_MAGIC.len() + 4 + 4;
 const SEG_PCS: u8 = 0x16;
 /// PGS segment type: END of display set.
 const SEG_END: u8 = 0x80;
-/// PCS `composition_state` value: Epoch Start (a fresh display).
-const PCS_COMPOSITION_STATE_EPOCH_START: u8 = 0x80;
+/// PCS `composition_state` value: Normal (an update to the current epoch).
+const PCS_COMPOSITION_STATE_NORMAL: u8 = 0x00;
 /// PGS segment header on the wire (inside `frame.data`): type(1) + size(2 BE).
 const PGS_SEG_HEADER_LEN: usize = 3;
 /// Byte offset of `width`/`height` within a PCS segment (after type+size).
@@ -346,7 +346,7 @@ impl PgsSupWriter {
             PCS_FRAME_RATE,
             0x00,
             0x00, // composition_number
-            PCS_COMPOSITION_STATE_EPOCH_START,
+            PCS_COMPOSITION_STATE_NORMAL,
             0x00, // palette_update_flag
             0x00, // palette_id
             PCS_NO_OBJECTS,

src/mux/fvi_sink.rs

@@ -82,8 +82,16 @@ fn write_fvi_record(w: &mut dyn Write, r: &PictureRecord) -> io::Result<()> {
     // `src` is REQUIRED by the record schema (Appendix A); when provenance is
     // absent the member is still emitted as null — a reader treats null as
     // "position unknown".
+    //
+    // Per `docs/FVI_FORMAT.md` §9, `src.byte` is the offset of the AU's first
+    // byte WITHIN its `sector` (not the absolute source offset). `SourcePos.byte`
+    // is the absolute offset, so reduce it modulo the sector size; `sector`
+    // already carries the whole-sector count.
     let src = match r.source {
-        Some(s) => serde_json::json!({ "sector": s.sector, "byte": s.byte }),
+        Some(s) => serde_json::json!({
+            "sector": s.sector,
+            "byte": s.byte % u64::from(FVI_SECTOR_SIZE),
+        }),
         None => serde_json::Value::Null,
     };
 
@@ -355,8 +363,9 @@ mod tests {
         let dir = tempdir();
         let path = dir.join("movie.fvi");
         let mut sink = FviSink::create(&path, &mpeg2_title(), "iso://m.iso".into(), 1).unwrap();
-        // Video frame on track 0 → indexed.
-        sink.write(&vframe(0, Some(i_pic()), Some(SourcePos::at_byte(2048))))
+        // Video frame on track 0 → indexed. Offset 2148 = sector 1, byte 100
+        // within that sector (exercises the within-sector `src.byte`, §9).
+        sink.write(&vframe(0, Some(i_pic()), Some(SourcePos::at_byte(2148))))
             .unwrap();
         // Audio frame on a non-video track → ignored.
         sink.write(&vframe(7, None, Some(SourcePos::at_byte(9999))))
@@ -387,6 +396,7 @@ mod tests {
         assert_eq!(rec["nb_fields"], 2);
         assert_eq!(rec["pts"], 0);
         assert_eq!(rec["src"]["sector"], 1);
+        assert_eq!(rec["src"]["byte"], 100); // 2148 % 2048 → within-sector (§9)
         assert!(rec.get("dts").is_none(), "no DTS on a frame → omitted");
         assert!(
             rec.get("gop").is_none(),

src/mux/m2ts_mux/mod.rs

@@ -76,15 +76,18 @@ const PCR_INTERVAL_PACKETS: u64 = 40;
 /// the picture it timestamps. 200 ms in 90 kHz ticks.
 const PCR_LEAD_90KHZ: u64 = 90_000 / 5;
 
+// PMT stream-type codes. These are the same elementary-stream coding-type
+// registry as the parse side; the single source of truth is `consts::coding_type`.
+use crate::consts::coding_type;
 /// HEVC stream-type code, ISO/IEC 13818-1 Table 2-34 (2015 amendment).
-const STREAM_TYPE_HEVC: u8 = 0x24;
+const STREAM_TYPE_HEVC: u8 = coding_type::HEVC;
 /// AC-3 / E-AC-3. Not an ISO assignment — sits in the user-private
 /// 0x80-0xFF range and is the Blu-ray Disc Association / ATSC A/52
 /// convention.
-const STREAM_TYPE_AC3: u8 = 0x81;
+const STREAM_TYPE_AC3: u8 = coding_type::AC3;
 /// Dolby TrueHD. Also a private/BD-conventional value in the
 /// user-private 0x80-0xFF range, not an ISO assignment.
-const STREAM_TYPE_TRUEHD: u8 = 0x83;
+const STREAM_TYPE_TRUEHD: u8 = coding_type::TRUEHD;
 
 /// Audio codec hint for [`M2tsMux::new`] / [`M2tsMux::set_audio`]. The
 /// muxer needs to know the codec to pick the right PMT `stream_type`
@@ -251,7 +254,10 @@ impl<W: Write> M2tsMux<W> {
         // step — e.g. a leading audio frame ahead of the first video keyframe),
         // which still floors to 0 per the documented behavior.
         let delta = raw_90k.wrapping_sub(base) & 0x1_FFFF_FFFF;
-        if delta > (1 << 32) { 0 } else { delta }
+        // Signed 33-bit: the sign bit is bit 32 (value 2^32), so the entire
+        // upper half [2^32, 2^33) is negative (frame before base) and floors
+        // to 0. delta == 2^32 is the most-negative value (-2^32), hence `>=`.
+        if delta >= (1 << 32) { 0 } else { delta }
     }
 
     /// Emit one PES payload as a chain of TS packets on `pid`. If `pcr`
@@ -420,7 +426,12 @@ impl<W: Write> M2tsMux<W> {
 
 /// Build a PES packet for a video access unit.
 fn build_video_pes(pts_90k: u64, es: &[u8]) -> Vec<u8> {
-    build_pes_packet(0xE0, pts_90k, es, /* length_in_header */ false)
+    build_pes_packet(
+        crate::consts::pes_stream_id::VIDEO,
+        pts_90k,
+        es,
+        /* length_in_header */ false,
+    )
 }
 
 /// Build a PES packet for an audio access unit.
@@ -430,7 +441,12 @@ fn build_audio_pes(pts_90k: u64, es: &[u8]) -> Vec<u8> {
     // start code. For an access unit larger than ~64 KiB (rare — e.g. a
     // large TrueHD frame) the length field falls back to the unbounded
     // (0x0000) form, which most demuxers tolerate for private_stream_1.
-    build_pes_packet(0xBD, pts_90k, es, /* length_in_header */ true)
+    build_pes_packet(
+        crate::consts::pes_stream_id::PRIVATE_STREAM_1,
+        pts_90k,
+        es,
+        /* length_in_header */ true,
+    )
 }
 
 fn build_pes_packet(stream_id: u8, pts_90k: u64, es: &[u8], length_in_header: bool) -> Vec<u8> {

src/mux/mkv.rs

@@ -28,6 +28,9 @@ const CICP_PRIMARIES_BT470BG: u8 = 5;
 const CICP_PRIMARIES_BT601_525: u8 = 6;
 /// ColourPrimaries = 9 (BT.2020 / BT.2100) — ITU-T H.273 Table 2.
 const CICP_PRIMARIES_BT2020: u8 = 9;
+/// ColourPrimaries = 2 ("unspecified" — colorimetry unknown) — ITU-T H.273
+/// Table 2.
+const CICP_PRIMARIES_UNSPECIFIED: u8 = 2;
 
 /// TransferCharacteristics = 1 (BT.709) — ITU-T H.273 Table 3.
 const CICP_TRANSFER_BT709: u8 = 1;
@@ -41,6 +44,9 @@ const CICP_TRANSFER_PQ: u8 = 16;
 /// TransferCharacteristics = 18 (ARIB STD-B67 / Hybrid Log-Gamma) — ITU-T H.273
 /// Table 3.
 const CICP_TRANSFER_HLG: u8 = 18;
+/// TransferCharacteristics = 2 ("unspecified" — transfer unknown) — ITU-T H.273
+/// Table 3.
+const CICP_TRANSFER_UNSPECIFIED: u8 = 2;
 
 /// MatrixCoefficients = 1 (BT.709) — ITU-T H.273 Table 4.
 const CICP_MATRIX_BT709: u8 = 1;
@@ -50,6 +56,8 @@ const CICP_MATRIX_BT470BG: u8 = 5;
 const CICP_MATRIX_BT601_525: u8 = 6;
 /// MatrixCoefficients = 9 (BT.2020 non-constant luminance) — ITU-T H.273 Table 4.
 const CICP_MATRIX_BT2020NC: u8 = 9;
+/// MatrixCoefficients = 2 ("unspecified" — matrix unknown) — ITU-T H.273 Table 4.
+const CICP_MATRIX_UNSPECIFIED: u8 = 2;
 
 /// Matroska Colour/Range = 1 (broadcast / studio-swing "limited" range). RFC
 /// 9559 Range element. (0 = unspecified, 2 = full.)
@@ -104,7 +112,16 @@ pub(crate) fn cicp_for_video(v: &VideoStream) -> (u8, u8, u8, u8) {
             CICP_PRIMARIES_BT601_525,
             COLOUR_RANGE_LIMITED,
         ),
-        ColorSpace::Unknown => (0, 0, 0, 0),
+        // Unknown colorimetry → CICP "unspecified" (code point 2) for matrix,
+        // transfer, and primaries, with limited range (the disc norm). Both the
+        // MKV sink and the FVI sidecar emit 2 so the two sinks of one title
+        // agree (matches `Colour::from_color_space`'s Unknown mapping).
+        ColorSpace::Unknown => (
+            CICP_MATRIX_UNSPECIFIED,
+            CICP_TRANSFER_UNSPECIFIED,
+            CICP_PRIMARIES_UNSPECIFIED,
+            COLOUR_RANGE_LIMITED,
+        ),
     };
     // Override the transfer for HDR signalled by the HdrFormat (the coarse enum
     // can't express PQ/HLG). Only applies on the enum fallback; a measured CICP
@@ -756,9 +773,8 @@ impl<W: Write + Seek> MkvMuxer<W> {
         }
         // Stamp the freemkv version so any muxed file is traceable to the build
         // that produced it (MediaInfo "Writing application"/"library").
-        const FREEMKV_MUX_APP: &str = concat!("freemkv ", env!("CARGO_PKG_VERSION"));
-        ebml::write_string(&mut writer, ebml::MUXING_APP, FREEMKV_MUX_APP)?;
-        ebml::write_string(&mut writer, ebml::WRITING_APP, FREEMKV_MUX_APP)?;
+        ebml::write_string(&mut writer, ebml::MUXING_APP, crate::MUX_APP)?;
+        ebml::write_string(&mut writer, ebml::WRITING_APP, crate::MUX_APP)?;
         if let Some(t) = title {
             ebml::write_string(&mut writer, ebml::TITLE, t)?;
         }
@@ -1549,6 +1565,41 @@ mod tests {
         );
     }
 
+    /// Unknown colorimetry with no measured CICP and no HDR must emit CICP
+    /// "unspecified" (code point 2) for matrix/transfer/primaries — never 0 — so
+    /// the MKV sink agrees with the FVI sidecar (`Colour::from_color_space`).
+    #[test]
+    fn unknown_color_space_emits_unspecified_cicp() {
+        let v = VideoStream {
+            pid: 0xE0,
+            codec: Codec::Hevc,
+            resolution: Resolution::R1080p,
+            frame_rate: crate::disc::FrameRate::F24,
+            hdr: HdrFormat::Sdr,
+            color_space: ColorSpace::Unknown,
+            display_aspect: None,
+            secondary: false,
+            label: String::new(),
+            measured_cicp: None,
+        };
+        let t = MkvTrack::video(&v);
+        assert_eq!(
+            (
+                t.colour_matrix,
+                t.colour_transfer,
+                t.colour_primaries,
+                t.colour_range
+            ),
+            (
+                CICP_MATRIX_UNSPECIFIED,
+                CICP_TRANSFER_UNSPECIFIED,
+                CICP_PRIMARIES_UNSPECIFIED,
+                COLOUR_RANGE_LIMITED
+            ),
+            "Unknown colorimetry must emit CICP 'unspecified' (2), not 0"
+        );
+    }
+
     /// Helper: search for a 4-byte big-endian EBML ID in a byte slice.
     fn find_id(data: &[u8], id: u32) -> Option<usize> {
         let bytes = id.to_be_bytes();

src/mux/mkvstream.rs

@@ -101,8 +101,12 @@ enum WriteMode {
     /// Header written; muxing live. Boxed (MkvMuxer is large) to keep the enum
     /// small (clippy::large_enum_variant).
     Active(Box<MkvMuxer<Box<dyn WriteSeek + Send>>>),
-    /// Transient placeholder held only across the Pending → Active swap; never
-    /// observed by `read` / `write` / `finish`.
+    /// Sentinel held in `self.mode` while the muxer is being built (across the
+    /// Pending → Active swap). It is also the terminal state left behind after
+    /// `finish()` swaps the muxer out, and the degraded state left behind if
+    /// `activate()` fails partway (the first error still surfaces via `?`). In
+    /// that terminal state a subsequent `write()` no-ops (`Ok(())`) and `finish()`
+    /// does not re-finalize.
     Building,
 }
 
@@ -188,7 +192,11 @@ impl MkvStream {
     /// available), then write the header and replay buffered frames. A no-op if
     /// not pending. The muxer only ever muxes the track it is given — this routes
     /// the parser's measured value onto that track first.
-    fn activate(&mut self, coding: Option<crate::mux::codec::PictureInfo>) -> io::Result<()> {
+    fn activate(
+        &mut self,
+        coding: Option<crate::mux::codec::PictureInfo>,
+        video_picture_seen: bool,
+    ) -> io::Result<()> {
         let mut pending = match std::mem::replace(&mut self.mode, Mode::Write(WriteMode::Building))
         {
             Mode::Write(WriteMode::Pending(p)) => p,
@@ -199,7 +207,7 @@ impl MkvStream {
             }
         };
         if let Some(vt) = pending.video_track {
-            apply_coding_to_track(&mut pending.tracks[vt], coding);
+            apply_coding_to_track(&mut pending.tracks[vt], coding, video_picture_seen);
         }
         // --log-level 3: dump the FINAL TrackEntry metadata (field order set).
         for (i, track) in pending.tracks.iter().enumerate() {
@@ -240,12 +248,20 @@ impl MkvStream {
 /// Set a video track's `FieldOrder` from the MEASURED coding of the first coded
 /// picture — the parser's value, the first time, never a guess.
 ///
-/// A progressive track has no field order (left UNDETERMINED — expected). An
-/// INTERLACED track that reaches here with no measured field order is a
+/// A progressive track — or a progressive picture on an interlaced-flagged track
+/// — has no field order (left UNDETERMINED — expected). An INTERLACED track that
+/// reaches here WITH a video picture but no measured field order is a
 /// parser/source gap (MPEG-2 carries `top_field_first` on every interlaced
 /// picture, so it should never be missing): LOG it loudly so the source can be
 /// debugged, and leave UNDETERMINED — a muxer never fabricates a source fact.
-fn apply_coding_to_track(track: &mut MkvTrack, coding: Option<crate::mux::codec::PictureInfo>) {
+/// `video_picture_seen == false` (an empty title finalized with no frames, or a
+/// cap-triggered build that never saw the video frame) is NOT a defect — the
+/// missing coding is expected there, so log it quietly.
+fn apply_coding_to_track(
+    track: &mut MkvTrack,
+    coding: Option<crate::mux::codec::PictureInfo>,
+    video_picture_seen: bool,
+) {
     // HDR10 static metadata measured from the bitstream (HEVC SEI). Applied for
     // ANY track type that carries it (independent of interlace): the first coded
     // picture's PictureInfo holds it once both HDR10 SEI messages were seen.
@@ -260,17 +276,32 @@ fn apply_coding_to_track(track: &mut MkvTrack, coding: Option<crate::mux::codec:
     match coding.and_then(|c| c.field_order()) {
         Some(FieldOrder::Tff) => track.field_order = ebml::FIELD_ORDER_TFF,
         Some(FieldOrder::Bff) => track.field_order = ebml::FIELD_ORDER_BFF,
-        other => {
+        // A progressive picture on an interlaced-flagged track carries no field
+        // order. Leave UNDETERMINED (not a guess) — there is no parser gap here.
+        Some(FieldOrder::Progressive) => {
+            track.field_order = ebml::FIELD_ORDER_UNDETERMINED;
+        }
+        None if video_picture_seen => {
             tracing::warn!(
                 target: "mux",
-                "interlaced video track reached the muxer with NO measured field order \
-                 (field_order={:?}, coding_present={}); writing FieldOrder=UNDETERMINED \
-                 — NOT a guess. Debug why the source/parser did not set top_field_first.",
-                other,
+                "interlaced video track had a video picture but NO usable field order \
+                 (coding_present={}); writing FieldOrder=UNDETERMINED — NOT a guess. \
+                 Debug why the source/parser did not set top_field_first.",
                 coding.is_some(),
             );
             track.field_order = ebml::FIELD_ORDER_UNDETERMINED;
         }
+        None => {
+            // No video picture was ever measured (empty title finalized with no
+            // frames, or a cap-triggered build before the first video frame).
+            // Coding is legitimately absent, not a parser defect — log quietly.
+            tracing::debug!(
+                target: "mux",
+                "interlaced video track activated with no video picture \
+                 (empty/buffered-only title); writing FieldOrder=UNDETERMINED.",
+            );
+            track.field_order = ebml::FIELD_ORDER_UNDETERMINED;
+        }
     }
 }
 
@@ -427,7 +458,7 @@ impl crate::pes::Stream for MkvStream {
             // Pass the trigger frame's coding only when it IS the video frame; a
             // cap-triggered build never saw the video frame, so nothing measured
             // is passed (apply_coding_to_track then logs + leaves UNDETERMINED).
-            self.activate(if use_coding { frame.coding } else { None })?;
+            self.activate(if use_coding { frame.coding } else { None }, use_coding)?;
             if let Mode::Write(WriteMode::Active(m)) = &mut self.mode {
                 return m.write_frame(
                     frame.track,
@@ -450,7 +481,10 @@ impl crate::pes::Stream for MkvStream {
         // A title that produced no frames (or only buffered ones) is still
         // finalized into a valid MKV: activate now with no measured coding.
         if matches!(self.mode, Mode::Write(WriteMode::Pending(_))) {
-            self.activate(None)?;
+            // No video picture was ever measured for this title (it produced no
+            // frames, or only buffered non-video ones): coding is legitimately
+            // absent, not a parser defect — `video_picture_seen=false`.
+            self.activate(None, false)?;
         }
         if let Mode::Write(WriteMode::Active(m)) =
             std::mem::replace(&mut self.mode, Mode::Write(WriteMode::Building))
@@ -923,7 +957,7 @@ mod tests {
 
         // MEASURED bottom-field-first → BFF (6). The red-flag fix.
         let mut t = interlaced_track();
-        apply_coding_to_track(&mut t, Some(pic(false, false)));
+        apply_coding_to_track(&mut t, Some(pic(false, false)), true);
         assert_eq!(
             t.field_order,
             ebml::FIELD_ORDER_BFF,
@@ -932,27 +966,37 @@ mod tests {
 
         // MEASURED top-field-first → TFF (1).
         let mut t = interlaced_track();
-        apply_coding_to_track(&mut t, Some(pic(true, false)));
+        apply_coding_to_track(&mut t, Some(pic(true, false)), true);
         assert_eq!(
             t.field_order,
             ebml::FIELD_ORDER_TFF,
             "measured TFF → FieldOrder=1"
         );
 
-        // Interlaced track, NO measured coding → UNDETERMINED (logged loudly,
-        // never faked).
+        // Interlaced track, a video picture but NO usable field order →
+        // UNDETERMINED (logged loudly, never faked).
         let mut t = interlaced_track();
-        apply_coding_to_track(&mut t, None);
+        apply_coding_to_track(&mut t, None, true);
         assert_eq!(
             t.field_order,
             ebml::FIELD_ORDER_UNDETERMINED,
             "no measured value → UNDETERMINED, never a guess"
         );
 
+        // Interlaced track activated with NO video picture (empty/buffered-only
+        // title) → UNDETERMINED, logged quietly (not a parser defect).
+        let mut t = interlaced_track();
+        apply_coding_to_track(&mut t, None, false);
+        assert_eq!(
+            t.field_order,
+            ebml::FIELD_ORDER_UNDETERMINED,
+            "empty title → UNDETERMINED, never a guess"
+        );
+
         // Progressive picture on an interlaced-flagged track → UNDETERMINED (no
         // field order applies; not faked to TFF/BFF).
         let mut t = interlaced_track();
-        apply_coding_to_track(&mut t, Some(pic(true, true)));
+        apply_coding_to_track(&mut t, Some(pic(true, true)), true);
         assert_eq!(t.field_order, ebml::FIELD_ORDER_UNDETERMINED);
 
         // A PROGRESSIVE track is never touched — field order stays UNDETERMINED.
@@ -969,7 +1013,7 @@ mod tests {
             measured_cicp: None,
         });
         assert!(!prog.interlaced);
-        apply_coding_to_track(&mut prog, Some(pic(false, false)));
+        apply_coding_to_track(&mut prog, Some(pic(false, false)), true);
         assert_eq!(prog.field_order, ebml::FIELD_ORDER_UNDETERMINED);
     }
 
@@ -1011,18 +1055,18 @@ mod tests {
         let mut t = make();
         assert!(t.hdr10.is_none(), "fresh track has no HDR10");
         let pic = PictureInfo::coding_type_only(CodingType::I).with_hdr10(Some(h));
-        apply_coding_to_track(&mut t, Some(pic));
+        apply_coding_to_track(&mut t, Some(pic), true);
         assert_eq!(t.hdr10, Some(h), "measured HDR10 must reach the track");
 
         // Picture without HDR10 → track stays None (never fabricated).
         let mut t = make();
         let pic = PictureInfo::coding_type_only(CodingType::I);
-        apply_coding_to_track(&mut t, Some(pic));
+        apply_coding_to_track(&mut t, Some(pic), true);
         assert!(t.hdr10.is_none(), "no measured HDR10 → track stays None");
 
         // No coding at all → None.
         let mut t = make();
-        apply_coding_to_track(&mut t, None);
+        apply_coding_to_track(&mut t, None, true);
         assert!(t.hdr10.is_none());
     }
 

src/mux/ps.rs

@@ -23,10 +23,10 @@ const SYSTEM_HEADER_ID: u8 = 0xBB;
 const PROGRAM_END_ID: u8 = 0xB9;
 
 /// Private stream 1 (AC3, DTS, LPCM, subtitles).
-const PRIVATE_STREAM_1: u8 = 0xBD;
+const PRIVATE_STREAM_1: u8 = crate::consts::pes_stream_id::PRIVATE_STREAM_1;
 /// Private stream 2 (0xBF) — DVD navigation (PCI/DSI). Carries no muxable
 /// elementary stream; expected to be dropped on every disc.
-const PRIVATE_STREAM_2: u8 = 0xBF;
+const PRIVATE_STREAM_2: u8 = crate::consts::pes_stream_id::PRIVATE_STREAM_2;
 
 /// Hard cap on the demuxer's reassembly buffer. A length-0 (unbounded) video
 /// PES is delimited by the next PS-layer boundary; if a corrupt stream declares
@@ -107,8 +107,8 @@ impl PsPacket {
     /// mis-routing the packet.
     pub fn dvd_pid(&self) -> Option<u16> {
         match self.stream_id {
-            0xE0..=0xEF => Some(DVD_VIDEO_PID),
-            0xBD => {
+            crate::consts::pes_stream_id::VIDEO..=0xEF => Some(DVD_VIDEO_PID),
+            PRIVATE_STREAM_1 => {
                 let sub = self.sub_stream_id?;
                 dvd_audio_pid(sub).or_else(|| dvd_subtitle_pid(sub))
             }
@@ -348,8 +348,8 @@ fn find_ps_boundary(data: &[u8], from: usize) -> Option<usize> {
 fn is_pes_stream_id(id: u8) -> bool {
     // Video: 0xE0-0xEF, MPEG audio: 0xC0-0xDF, private stream 1: 0xBD,
     // private stream 2: 0xBF, padding: 0xBE, ECM/EMM etc.
-    // We parse anything in the PES range.
-    matches!(id, 0xBD..=0xEF)
+    // We parse anything in the payload-bearing PES range.
+    crate::consts::pes_stream_id::PAYLOAD_RANGE.contains(&id)
 }
 
 /// Parse a single PES packet from a byte slice that starts at the start code.
@@ -365,12 +365,12 @@ fn parse_pes_packet(data: &[u8]) -> Option<PsPacket> {
     let stream_id = data[3];
 
     // Padding stream — skip entirely.
-    if stream_id == 0xBE {
+    if stream_id == crate::consts::pes_stream_id::PADDING_STREAM {
         return None;
     }
 
     // Streams without standard PES header extension.
-    if stream_id == 0xBF {
+    if stream_id == PRIVATE_STREAM_2 {
         let payload = if data.len() > 6 { &data[6..] } else { &[] };
         return Some(PsPacket {
             stream_id,

src/mux/resolve.rs

@@ -12,6 +12,7 @@
 //! | network:// | Yes (listen) | Yes (connect) | host:port (required) |
 //! | stdio:// | Yes (stdin) | Yes (stdout) | empty |
 //! | null:// | -- | Yes | empty |
+//! | demux:// | -- | Yes | directory path (required) — per-track ES demux |
 //! | fvi://  | -- | Yes | file path (required) — per-picture video index |
 //!
 //! Bare paths without a scheme are rejected.
@@ -920,6 +921,30 @@ mod tests {
             !parse_url("dir://x").is_disc_source(),
             "dir:// is a sink, never a disc source"
         );
+        // fvi:// parses to Fvi with the raw remainder as the path, and is a
+        // sink (never a disc source) — parallel to the demux:// coverage above.
+        match parse_url("fvi://out/movie.fvi") {
+            StreamUrl::Fvi { path } => {
+                assert_eq!(path, PathBuf::from("out/movie.fvi"));
+            }
+            other => panic!("fvi:// must parse to Fvi, got {other:?}"),
+        }
+        assert_eq!(parse_url("fvi://x").scheme(), "fvi");
+        assert_eq!(parse_url("fvi://x/y.fvi").path_str(), "x/y.fvi");
+        assert!(
+            !parse_url("fvi://x").is_disc_source(),
+            "fvi:// is a sink, never a disc source"
+        );
+    }
+
+    /// `fvi://` is output-only: `input()` rejects it with StreamWriteOnly
+    /// (E9001 → Unsupported), mirroring `null://` / `demux://`.
+    #[test]
+    fn input_fvi_url_is_write_only() {
+        assert_eq!(
+            input_err_kind("fvi://out/movie.fvi"),
+            std::io::ErrorKind::Unsupported
+        );
     }
 
     /// `dir://` is output-only: `input()` rejects it (StreamWriteOnly →

src/mux/ts.rs

@@ -229,6 +229,12 @@ impl TsDemuxer {
     /// `data` in place. Zero-copy on the bulk path; one 192-byte copy
     /// on the boundary.
     pub fn feed(&mut self, data: &[u8]) -> Vec<PesPacket> {
+        // A plain feed carries no provenance. Reset any base a prior
+        // `feed_at` left behind so mixing the two entry points is safe:
+        // after this call no `SourcePos` is stamped, and the stale running
+        // base can't leak a wrong offset into the boundary packet.
+        self.feed_base = 0;
+        self.has_base = false;
         self.feed_inner(data)
     }
 
@@ -263,15 +269,20 @@ impl TsDemuxer {
                 self.remainder.extend_from_slice(data);
                 return completed;
             }
+            // Capture the remainder length before clearing — it's how many of
+            // the boundary packet's bytes lived in the PREVIOUS feed buffer,
+            // and `feed_base` currently points at the FIRST byte of THIS buffer.
+            let rem_len = self.remainder.len();
             let mut boundary = [0u8; BD_SOURCE_PACKET_BYTES];
-            boundary[..self.remainder.len()].copy_from_slice(&self.remainder);
-            boundary[self.remainder.len()..].copy_from_slice(&data[..need]);
+            boundary[..rem_len].copy_from_slice(&self.remainder);
+            boundary[rem_len..].copy_from_slice(&data[..need]);
             self.remainder.clear();
-            // The boundary packet began in the PREVIOUS feed buffer; stamp it
-            // with the offset just before this buffer (its first bytes' base).
-            let src = self
-                .has_base
-                .then(|| crate::pes::SourcePos::at_byte(self.feed_base.saturating_sub(1)));
+            // The boundary packet's first byte sat `rem_len` bytes before the
+            // current feed_base (in the previous buffer). Stamp it there — not
+            // at `feed_base - 1`, which would be wrong by `rem_len - 1` bytes.
+            let src = self.has_base.then(|| {
+                crate::pes::SourcePos::at_byte(self.feed_base.saturating_sub(rem_len as u64))
+            });
             self.process_packet(&boundary, src, &mut completed);
             offset = need;
         }
@@ -931,6 +942,76 @@ mod tests {
         assert!(result.is_empty());
     }
 
+    /// A boundary packet (one split across two feeds) must be stamped with the
+    /// source offset of its FIRST byte, which sat `remainder.len()` bytes before
+    /// the current feed's base — not at `feed_base - 1`. We feed two 192-byte
+    /// packets via `feed_at`, splitting mid-second-packet so the second packet
+    /// is reassembled at the boundary, and assert its provenance lands exactly
+    /// on its first byte.
+    #[test]
+    fn boundary_packet_source_is_first_byte_not_base_minus_one() {
+        let pid = 0x1011;
+        let base: u64 = 20480; // sector-aligned (10 × 2048)
+        let mut demux = TsDemuxer::new(&[pid]);
+
+        let pkt0 = ts_payload_packet(pid, true, 0, &pes_start(b"AAAA"));
+        let pkt1 = ts_payload_packet(pid, true, 1, &pes_start(b"BBBB"));
+        let mut full = pkt0;
+        full.extend_from_slice(&pkt1);
+
+        // Split mid-pkt1 → pkt1 is reassembled from a 100-byte remainder + the
+        // next feed's head. pkt1's first byte is at absolute offset base + 192.
+        let split = BD_SOURCE_PACKET_BYTES + 100;
+        let out1 = demux.feed_at(base, &full[..split]);
+        assert!(out1.is_empty(), "pkt0's PES is still open");
+
+        // Second feed carries the rest; data[0] is at absolute base + split.
+        let out2 = demux.feed_at(base + split as u64, &full[split..]);
+        // pkt1 (PUSI) flushes pkt0's "AAAA" PES, stamped at pkt0's first byte.
+        assert_eq!(out2.len(), 1, "pkt0's PES completes when pkt1 starts");
+        assert_eq!(
+            out2[0].source.map(|s| s.byte),
+            Some(base),
+            "AAAA PES provenance is pkt0's first byte"
+        );
+
+        // Flush emits pkt1's "BBBB" PES — its source is the boundary stamp.
+        let out3 = demux.flush();
+        assert_eq!(out3.len(), 1, "pkt1's PES flushes out");
+        assert_eq!(
+            out3[0].source.map(|s| s.byte),
+            Some(base + BD_SOURCE_PACKET_BYTES as u64),
+            "boundary packet provenance must be its first byte (base + 192), \
+             not feed_base - 1"
+        );
+    }
+
+    /// Owner decision #7: a plain `feed()` must reset/ignore any base a prior
+    /// `feed_at()` left behind, so mixing the two is safe. After a `feed_at`
+    /// primes a base, the next plain `feed` must stamp `None` on PES packets it
+    /// begins.
+    #[test]
+    fn plain_feed_resets_prior_feed_at_base() {
+        let pid = 0x1011;
+        let mut demux = TsDemuxer::new(&[pid]);
+
+        // Prime a base via feed_at; pkt0's PES stays open.
+        let out1 = demux.feed_at(20480, &ts_payload_packet(pid, true, 0, &pes_start(b"AAAA")));
+        assert!(out1.is_empty());
+
+        // Plain feed must clear the base. pkt1 (PUSI) flushes "AAAA" (which was
+        // stamped during feed_at) and starts "BBBB" with NO provenance.
+        let out2 = demux.feed(&ts_payload_packet(pid, true, 1, &pes_start(b"BBBB")));
+        assert_eq!(out2.len(), 1, "AAAA completes");
+
+        let out3 = demux.flush();
+        assert_eq!(out3.len(), 1, "BBBB flushes");
+        assert_eq!(
+            out3[0].source, None,
+            "PES begun by a plain feed must carry no source after a prior feed_at"
+        );
+    }
+
     // ── scan_streams PMT parsing ──────────────────────────────────────────
 
     /// Wrap a 188-byte TS packet body in a 192-byte BD-TS packet

src/mux/tsmux.rs

@@ -312,11 +312,12 @@ impl<W: Write> TsMuxer<W> {
 
 /// Build a PES packet header for a BD stream.
 fn build_pes_header(pid: u16, pts_90k: u64, data_len: usize) -> Vec<u8> {
+    use crate::consts::pes_stream_id;
     // Determine stream_id from PID range
     let stream_id: u8 = if is_video_pid(pid) {
-        0xE0 // video
+        pes_stream_id::VIDEO
     } else {
-        0xBD // audio, PGS subtitle, or default (private stream 1)
+        pes_stream_id::PRIVATE_STREAM_1 // audio, PGS subtitle, or default
     };
 
     let pes_data_len = data_len + 8; // 3 header bytes + 5 PTS bytes + data
@@ -332,7 +333,7 @@ fn build_pes_header(pid: u16, pts_90k: u64, data_len: usize) -> Vec<u8> {
     // video; `write_frame` splits oversized 0xBD access units so a private
     // stream always fits a bounded u16 length here. The `> 65535` arm
     // remains a defensive fallback for video only.
-    if stream_id == 0xE0 || pes_data_len > 65535 {
+    if stream_id == pes_stream_id::VIDEO || pes_data_len > u16::MAX as usize {
         header.push(0x00);
         header.push(0x00);
     } else {

src/mux/videomap.rs

@@ -37,7 +37,7 @@ pub const FVI_VERSION: u32 = 1;
 
 /// Producing tool tag for the header `"generator"` member
 /// (`docs/FVI_FORMAT.md` §6).
-pub const FVI_GENERATOR: &str = concat!("freemkv/", env!("CARGO_PKG_VERSION"));
+pub const FVI_GENERATOR: &str = concat!("freemkv/", env!("FREEMKV_VERSION"), env!("GIT_SUFFIX"));
 
 /// Header `"timescale"` for all `pts`/`dts` ticks (`docs/FVI_FORMAT.md` §10).
 /// The highway carries presentation timestamps in nanoseconds, so the timescale
@@ -551,6 +551,21 @@ mod tests {
             },
             "measured CICP must override the coarse color_space enum"
         );
+
+        // Unknown colorimetry, SDR, no measured CICP → all code points map to
+        // "unspecified" (2), matching `from_color_space(Unknown)`. Both sinks of
+        // one title must emit 2, never 0.
+        let c = Colour::from_video(&mk(HdrFormat::Sdr, ColorSpace::Unknown, None));
+        assert_eq!(
+            c,
+            Colour {
+                primaries: 2,
+                transfer: 2,
+                matrix: 2,
+                full_range: false,
+            },
+            "Unknown colorimetry must emit CICP 'unspecified' (2), not 0"
+        );
     }
 
     #[test]

src/pes.rs

@@ -107,20 +107,17 @@ impl PesFrame {
     /// truncated `.pes` data would be accepted as a graceful end.
     pub fn deserialize(r: &mut dyn std::io::Read) -> std::io::Result<Option<Self>> {
         // Probe one byte first to distinguish clean EOF from a truncated
-        // header.
+        // header. Loop on EINTR so back-to-back signals don't fail a
+        // recoverable read — symmetric with read_exact's internal retry
+        // on the rest of the header and the data below.
         let mut first = [0u8; 1];
-        match r.read(&mut first) {
-            Ok(0) => return Ok(None), // clean EOF, no frame started
-            Ok(_) => {}
-            Err(e) if e.kind() == std::io::ErrorKind::Interrupted => {
-                // Retry-once on EINTR before committing to the header read.
-                match r.read(&mut first) {
-                    Ok(0) => return Ok(None),
-                    Ok(_) => {}
-                    Err(e) => return Err(e),
-                }
+        loop {
+            match r.read(&mut first) {
+                Ok(0) => return Ok(None), // clean EOF, no frame started
+                Ok(_) => break,
+                Err(e) if e.kind() == std::io::ErrorKind::Interrupted => continue,
+                Err(e) => return Err(e),
             }
-            Err(e) => return Err(e),
         }
 
         let mut header = [0u8; 22]; // 1 + 8 + 1 + 8 + 4

src/unlock.rs

@@ -104,11 +104,17 @@ pub fn register_unlocker(u: Box<dyn Unlocker>) {
 ///   * `Ok(Some((name, vid)))` — a registered unlocker matched, put the drive
 ///     into extended mode, and returned the OEM Volume ID. The caller stashes
 ///     the VID for the handshake phase and need not run the cert handshake.
-///   * `Ok(None)` — no unlocker matched, OR the matching unlocker failed
-///     ([`UnlockError`], logged). Either way the drive is usable in stock mode
-///     and the caller falls through to the in-tree cert handshake. Folding an
-///     unlock failure into `Ok(None)` keeps drive `init()` infallible — a drive
-///     that simply isn't firmware-unlockable must not fail init.
+///   * `Ok(None)` — no unlocker matched, OR the matching unlocker reported a
+///     *capability* failure ([`UnlockError::FirmwareNotUnlockable`],
+///     [`UnlockError::VidUnavailable`], or a cert-auth outcome — all logged).
+///     Either way the drive is usable in stock mode and the caller falls
+///     through to the in-tree cert handshake. Folding a capability failure into
+///     `Ok(None)` keeps drive `init()` infallible — a drive that simply isn't
+///     firmware-unlockable must not fail init.
+///   * `Err(_)` — the matching unlocker hit a genuine SCSI/transport fault
+///     ([`UnlockError::Scsi`]). The bus is broken, not merely unsupported, so
+///     this propagates and aborts init rather than silently falling through to
+///     a cert handshake that would also fail.
 pub(crate) fn route_unlock(
     scsi: &mut dyn ScsiTransport,
     id: &DriveId,
@@ -124,10 +130,29 @@ pub(crate) fn route_unlock(
             let name = u.name().to_string();
             match u.unlock(scsi, id) {
                 Ok(vid) => return Ok(Some((name, vid))),
+                // A genuine SCSI/transport fault is not "this drive can't be
+                // unlocked" — the bus is broken. Propagate so init() aborts
+                // instead of falling through to a cert handshake that will
+                // also fail on the same dead transport. The numeric code from
+                // the originating error is logged; the returned error is the
+                // canonical transport-error variant.
+                Err(UnlockError::Scsi(code)) => {
+                    tracing::error!(
+                        target: "freemkv::unlock",
+                        unlocker = %name,
+                        code,
+                        "unlocker hit a transport fault during unlock; aborting init"
+                    );
+                    return Err(crate::error::Error::ScsiError {
+                        opcode: 0,
+                        status: 0,
+                        sense: None,
+                    });
+                }
+                // A firmware unlocker that can't unlock / has no OEM VID:
+                // fall through to the cert handshake. Debug-only structured
+                // log (variant identifiers, no English prose).
                 Err(e) => {
-                    // A firmware unlocker that can't unlock / has no OEM VID:
-                    // fall through to the cert handshake. Debug-only structured
-                    // log (variant identifiers, no English prose).
                     tracing::warn!(
                         target: "freemkv::unlock",
                         unlocker = %name,
@@ -228,6 +253,10 @@ mod tests {
         /// `None` → `unlock` yields `Err(UnlockError::VidUnavailable)` so
         /// `route_unlock` falls through to the cert handshake.
         vid: Option<[u8; 16]>,
+        /// When `Some(code)`, `unlock` yields `Err(UnlockError::Scsi(code))`
+        /// (a transport fault) instead of consulting `vid`, so `route_unlock`
+        /// propagates an error and aborts init.
+        scsi_err: Option<u16>,
         /// Records whether set_max_read_speed was invoked.
         speed_ran: Arc<AtomicBool>,
     }
@@ -238,6 +267,7 @@ mod tests {
                 ran,
                 // Default: a successful unlock returning an all-zero VID.
                 vid: Some([0u8; 16]),
+                scsi_err: None,
                 speed_ran: Arc::new(AtomicBool::new(false)),
             }
         }
@@ -245,6 +275,10 @@ mod tests {
             self.vid = vid;
             self
         }
+        fn with_scsi_err(mut self, code: u16) -> Self {
+            self.scsi_err = Some(code);
+            self
+        }
         fn with_speed(mut self, speed_ran: Arc<AtomicBool>) -> Self {
             self.speed_ran = speed_ran;
             self
@@ -263,6 +297,9 @@ mod tests {
             _id: &DriveId,
         ) -> std::result::Result<Vid, UnlockError> {
             self.ran.store(true, Ordering::SeqCst);
+            if let Some(code) = self.scsi_err {
+                return Err(UnlockError::Scsi(code));
+            }
             match self.vid {
                 Some(v) => Ok(Vid(v)),
                 None => Err(UnlockError::VidUnavailable),
@@ -359,6 +396,33 @@ mod tests {
         assert!(got.is_none(), "no match → cert fallback");
     }
 
+    /// A matching unlocker that hits a genuine transport fault
+    /// (`UnlockError::Scsi`) makes `route_unlock` PROPAGATE an `Err` rather
+    /// than fold to `Ok(None)`: a dead bus must abort init, not silently fall
+    /// through to a cert handshake that would also fail. Capability failures
+    /// (`VidUnavailable` etc.) still fold to `Ok(None)` — proven by the sibling
+    /// routing tests; this one pins the transport-fault exception.
+    #[test]
+    fn route_unlock_propagates_scsi_transport_fault() {
+        let mut scsi = NoopTransport;
+
+        register_unlocker(Box::new(
+            FakeUnlocker::new("SCSIVNDR", Arc::new(AtomicBool::new(false)))
+                .with_scsi_err(crate::error::E_SCSI_ERROR),
+        ));
+
+        let got = route_unlock(&mut scsi, &fake_id("SCSIVNDR"));
+        assert!(
+            got.is_err(),
+            "a transport fault during unlock aborts init (propagates Err)"
+        );
+        assert_eq!(
+            got.unwrap_err().code(),
+            crate::error::E_SCSI_ERROR,
+            "propagated error is the canonical transport-error code"
+        );
+    }
+
     /// `unlocker_set_max_read_speed` consults the FIRST matching unlocker's
     /// `set_max_read_speed`. A matching unlocker is invoked; a non-match is a
     /// safe no-op (nothing invoked, `Ok(())`).

tests/fvi_pipeline.rs

@@ -160,7 +160,7 @@ fn two_gop_image() -> Vec<u8> {
     data[..a.len()].copy_from_slice(&a);
 
     let mut b = ps_pack_header();
-    b.extend_from_slice(&video_pes(&gop_es(), 3003)); // ~0.1s later
+    b.extend_from_slice(&video_pes(&gop_es(), 3003)); // one frame (~33 ms) later at 29.97 fps (3003/90000 s)
     let off = 3 * 2048;
     data[off..off + b.len()].copy_from_slice(&b);
 
@@ -276,13 +276,21 @@ fn fvi_sink_indexes_real_mpeg2_pipeline_output() {
         vec![0, 3],
         "stamped src sectors must reach the .fvi in arrival order; got {src_sectors:?}"
     );
-    // For each stamped record, sector == byte / 2048 (SourcePos::at_byte). The
-    // byte offset is exact (here 14 into each region, just past the pack
-    // header), so it is NOT sector-aligned — provenance is byte-exact.
+    // Per FVI_FORMAT.md §9, `src.byte` is the offset of the AU's first byte
+    // WITHIN its 2048-byte `src.sector`, so it is always < 2048. The two
+    // stamped sources sit 14 bytes into their sector (just past the pack
+    // header), so the within-sector byte is exact (14) — provenance is
+    // byte-exact, carried unchanged from demux.
     for r in &records {
-        if let (Some(sector), Some(byte)) = (r["src"]["sector"].as_u64(), r["src"]["byte"].as_u64())
-        {
-            assert_eq!(sector, byte / 2048, "src.sector must equal src.byte / 2048");
+        if let Some(byte) = r["src"]["byte"].as_u64() {
+            assert!(
+                byte < 2048,
+                "src.byte is a within-sector offset (§9), must be < 2048; got {byte}"
+            );
+            assert_eq!(
+                byte, 14,
+                "stamped src.byte is 14 (just past the pack header)"
+            );
         }
     }
 
@@ -338,7 +346,7 @@ fn fvi_sink_indexes_non_mpeg2_frames_codec_agnostically() {
     assert_eq!(recs[0]["type"], "I");
     assert_eq!(recs[0]["pts"], 1234);
     assert_eq!(recs[0]["src"]["sector"], 4); // 8192 / 2048
-    assert_eq!(recs[0]["src"]["byte"], 8192);
+    assert_eq!(recs[0]["src"]["byte"], 0); // 8192 is sector-aligned → within-sector offset 0 (§9)
     assert!(
         recs[0].get("field_order").is_none() && recs[0].get("nb_fields").is_none(),
         "coding-absent record omits field_order/nb_fields"