re: recover the .slb leading segment — mono, and scoped by measurement

to_xma_riffs now emits the leading headerless segment when it sits at a whole
number of XMA1 packets and carries a non-zero byte. VOICE_D_453 goes from a
0.14 s trailing fragment to a 45116-byte leading sub-wave that dominates it.

I withdrew this exact change earlier for two reasons. Both are now answered
rather than argued away:

* "It recovers no audio" -- it used the STEREO format. At two channels every
  bank yields exactly 1792 bytes, one frame, whatever its size. Mono yields up
  to 113x more.
* "It matches 1524 of 8021 RIFF-bearing entries" -- the byte-level reach is
  still 1524, but the audible reach is not. Across the 84 movie-bound banks
  the segment adds >1 s to exactly 7, the hokyu_*_H tankers on D_453/D_454 --
  precisely the broken ones -- and <=0.25 s to 66 of the rest. The largest
  non-resupply addition is S04A at +0.66 s on a 256 s movie.

The safety oracle is recorded with its limits: 8 of the 84 banks ALREADY
exceed their movie's duration before the change, by hundredths of a second,
so it cannot resolve differences at that scale. It establishes scoping, not
correctness. Callers clamp to the movie length regardless.

VOICE_D_451's all-zero leading region is skipped by the non-zero guard, so
the rule cannot prepend silence to a bank that does not need it. Pinned, as
is the packet arithmetic (n = 8, 1, 7, 22, 29) which has no tunable.

slb_disc, movie_subtitle_disc and movie_manifest_disc all still pass.

NOT verified by ear -- that needs a human.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PMRJjbxLqZtsb5Vb7KunPE

crates/sylpheed-formats/src/slb.rs

@@ -133,7 +133,8 @@ fn parse_voice_clip(name: &str) -> VoiceClip {
 /// tracing confirmed the movie→voice binding; this fixes the *decode* of `RT*`.)
 pub fn to_xma_riffs(slb: &[u8]) -> Vec<Vec<u8>> {
     let mut out = Vec::new();
-    if find(slb, b"RIFF", 0).is_none() {
+    let first_riff = find(slb, b"RIFF", 0);
+    if first_riff.is_none() {
         // Headerless single-stream bank.
         if let Some(data) = slb.get(HEADERLESS_DATA_OFFSET..) {
             if !data.is_empty() {
@@ -142,22 +143,33 @@ pub fn to_xma_riffs(slb: &[u8]) -> Vec<Vec<u8>> {
         }
         return out;
     }
-    // ❌ A "leading headerless stream" rule was tried here and WITHDRAWN.
+    // HYBRID banks: a headerless packet stream followed by RIFF sub-waves, two
+    // SEQUENTIAL SEGMENTS of one clip. The branch above only fires when there is
+    // no `RIFF` at all, so the leading segment used to be dropped — which is why
+    // `VOICE_D_453` decoded to 0.14 s: its line is in that segment and only the
+    // trailing fragment survived.
     //
-    // The structure is real: in all five resupply banks the first `RIFF` sits at
+    // The boundary is arithmetic, not a magic: XMA1 packets are 2048 bytes, so a
-    // exactly `HEADERLESS_DATA_OFFSET + n*XMA1_PACKET` (n = 8, 1, 7, 22, 29), and
+    // leading stream occupies exactly `HEADERLESS_DATA_OFFSET + n*XMA1_PACKET`.
-    // 87 % of `VOICE_D_453` lies in front of it. Emitting that region as a
+    // It decodes as **mono** — at two channels every bank yields exactly 1792
-    // sub-wave raised byte coverage from 5.4 % to 89.9 %.
+    // bytes, one frame, whatever its size.
     //
-    // But byte coverage was the wrong success metric. The emitted streams decode
+    // An earlier version of this was withdrawn for two good reasons, both now
-    // to **1792 PCM bytes** — silence — through the same FFmpeg path that decodes
+    // answered: it recovered no audio (it used the stereo format), and it
-    // the RIFF sub-waves fine, so the region is not XMA1 under the synthesised
+    // matched 1524 of the 8021 RIFF-bearing entries. The byte-level reach is
-    // format. And the rule is not narrow: it matches **1524 of the 8021**
+    // still 1524, but the *audible* reach is not: across the 84 movie-bound
-    // RIFF-bearing entries in `sound.pak`, including `RT*` banks that decode
+    // banks the segment adds >1 s to exactly **7** — the `hokyu_*_H` tankers
-    // correctly today. Landing it would have risked a large regression to fix
+    // bound to `VOICE_D_453`/`454`, i.e. precisely the broken ones — and
-    // five banks it does not actually fix.
+    // ≤0.25 s to 66 of the rest. Callers clamp to the movie length anyway.
-    //
+    if let Some(ri) = first_riff {
-    // See docs/re/voice-bank-leading-region.md.
+        if ri > HEADERLESS_DATA_OFFSET && (ri - HEADERLESS_DATA_OFFSET) % XMA1_PACKET == 0 {
+            if let Some(data) = slb.get(HEADERLESS_DATA_OFFSET..ri) {
+                if data.iter().any(|b| *b != 0) {
+                    out.push(build_riff(&synth_xma1_fmt(1, 0, 48000), data));
+                }
+            }
+        }
+    }
     let mut pos = 0usize;
     while let Some(ri) = find(slb, b"RIFF", pos) {
         // Parse this sub-wave's fmt + data (declared size is honest per sub-wave).

crates/sylpheed-formats/tests/slb_leading_segment_disc.rs

@@ -0,0 +1,86 @@
+//! The `.slb` leading segment — recovered, and scoped.
+//!
+//! `VOICE_D_453` used to decode to 0.14 s because its line lives in a headerless
+//! packet stream *before* the first `RIFF`, and the decoder started at the
+//! `RIFF`. The banks that looked fine were the ones whose leading segment is
+//! silence. One rule, two outcomes.
+
+use std::path::{Path, PathBuf};
+
+use sylpheed_formats::{slb, PakArchive};
+
+fn disc_root() -> Option<PathBuf> {
+    if let Ok(p) = std::env::var("SYLPHEED_DISC") {
+        let p = PathBuf::from(p);
+        if p.join("dat").is_dir() {
+            return Some(p);
+        }
+    }
+    let default = Path::new(
+        "/home/fabi/RE - Project Sylpheed/Project Sylpheed - Arc of Deception (USA, Europe) (En,Ja)",
+    );
+    default.join("dat").is_dir().then(|| default.to_path_buf())
+}
+
+macro_rules! skip_without_disc {
+    ($root:ident) => {
+        let Some($root) = disc_root() else {
+            eprintln!("SKIP: set SYLPHEED_DISC");
+            return;
+        };
+    };
+}
+
+fn bank(root: &Path, n: u32) -> Vec<u8> {
+    let snd = PakArchive::open(root.join("dat/sound.pak")).expect("sound.pak");
+    let path = format!("eng\\etc\\VOICE_D_{n}.slb");
+    let entry = snd.find_by_name(&path).expect("bank present");
+    snd.read(entry).expect("read")
+}
+
+/// The boundary is arithmetic and has no tunable: the first `RIFF` sits at
+/// exactly `HEADERLESS_DATA_OFFSET + n*XMA1_PACKET` in every resupply bank.
+#[test]
+fn leading_segment_is_a_whole_number_of_packets() {
+    skip_without_disc!(root);
+    for (n, packets) in [(450u32, 8usize), (451, 1), (452, 7), (453, 22), (454, 29)] {
+        let b = bank(&root, n);
+        let ri = b.windows(4).position(|w| w == b"RIFF").expect("has a RIFF");
+        assert!(ri > slb::HEADERLESS_DATA_OFFSET, "VOICE_D_{n}");
+        let lead = ri - slb::HEADERLESS_DATA_OFFSET;
+        assert_eq!(lead % slb::XMA1_PACKET, 0, "VOICE_D_{n} not a whole packet count");
+        assert_eq!(lead / slb::XMA1_PACKET, packets, "VOICE_D_{n} packet count");
+    }
+}
+
+/// The two banks whose line lives in the leading segment now yield it.
+#[test]
+fn broken_banks_recover_their_line() {
+    skip_without_disc!(root);
+    // (bank, sub-waves expected, payload of the leading one)
+    for (n, waves, lead_len) in [(453u32, 2usize, 45116usize), (454, 2, 59452)] {
+        let riffs = slb::to_xma_riffs(&bank(&root, n));
+        assert_eq!(riffs.len(), waves, "VOICE_D_{n} sub-wave count");
+        assert_eq!(riffs[0].len(), lead_len, "VOICE_D_{n} leading segment");
+        // It must be the LARGER part: that is the whole point.
+        assert!(
+            riffs[0].len() > riffs[1].len() * 5,
+            "VOICE_D_{n}: leading segment should dominate"
+        );
+    }
+}
+
+/// `VOICE_D_451`'s leading region is all zeros — the guard must skip it, so the
+/// rule cannot prepend silence to a bank that does not need it.
+#[test]
+fn all_zero_leading_region_is_skipped() {
+    skip_without_disc!(root);
+    let b = bank(&root, 451);
+    let ri = b.windows(4).position(|w| w == b"RIFF").unwrap();
+    assert!(
+        b[slb::HEADERLESS_DATA_OFFSET..ri].iter().all(|x| *x == 0),
+        "expected an all-zero leading region"
+    );
+    // Two sub-waves, both from the RIFF section — no synthesised third.
+    assert_eq!(slb::to_xma_riffs(&b).len(), 2);
+}

docs/re/BACKLOG.md

@@ -1057,9 +1057,18 @@ premise was wrong.**
   first `RIFF`. For the first three that discards only silence, so they looked
   fine (2.8 / 1.6 / 2.2 s); for the last two it discards **the line itself**,
   leaving 0.14 s and 0.43 s. One rule, two outcomes, depending on which segment
-  holds the speech. ▶️ The fix is now well-posed: emit the leading region **only
+  holds the speech. ✅ **(same day) LANDED.** `to_xma_riffs` emits the leading segment
-  when it carries signal**, which also avoids the 1524-bank blast radius that
+  wrapped **mono** when it is a whole number of packets and carries a non-zero
-  sank the earlier attempt.
+  byte. Both reasons the first attempt was withdrawn are answered: it used the
+  stereo format (mono yields up to 113× more), and while the byte-level reach is
+  still 1524 entries the **audible** reach is not — across the 84 movie-bound
+  banks it adds >1 s to exactly **7**, the `hokyu_*_H` tankers on
+  `VOICE_D_453`/`454`, and ≤0.25 s to 66 of the rest. ⚠️ The safety oracle is
+  weak and says so: **8 of the 84 already exceed their movie duration before the
+  change**, by hundredths of a second, so it establishes scoping rather than
+  correctness. Pinned by `tests/slb_leading_segment_disc.rs`, including that the
+  all-zero `VOICE_D_451` region stays skipped. ❔ Not verified by ear — that
+  needs a human.
 * ❌ **(2026-08-25) The `.slb` "multi-subwave" guess is REFUTED, and the voice
   decoder is discarding up to 87 % of a bank.** The record table gives a
   **direct** binding `hokyu_DS_s13A -> VOICE_D_452` where the corpus records the

docs/re/voice-bank-leading-region.md

@@ -1,7 +1,8 @@
 # The resupply voice banks — the decoder discards up to 87 % of them
 
-Status: ❌ the recorded "multi-subwave / not cleanly sliced" explanation is
+Status: ✅ **FIXED** — `to_xma_riffs` now emits the leading segment as mono, and
-**REFUTED**. ✅ each shared bank is **one generic line**, which explains the
+`VOICE_D_453`/`454` recover their line (`tests/slb_leading_segment_disc.rs`).
+❌ the recorded "multi-subwave / not cleanly sliced" explanation is **REFUTED**. ✅ each shared bank is **one generic line**, which explains the
 in-game verdict that rejected the `hokyu_DS_s13A` binding — the line really is
 generic. ❌ **my own "audio is missing" conclusion is RETRACTED**: a subtitle cue
 is a START time, not a point inside the clip, and under the correct reading every
@@ -178,6 +179,30 @@ Three of the five are decisive; the other two have their only cue at 0.0 s, whic
 tells us nothing in either direction. So **something is genuinely missing from
 these banks** — independent of anything above, and now measured rather than felt.
 
+## ✅ Landed — and why it is safe this time
+
+`to_xma_riffs` now emits the leading segment, wrapped **mono**, when it sits at a
+whole number of packets and carries a non-zero byte. The first attempt at this
+was withdrawn for two good reasons, and both are answered:
+
+* *"It recovers no audio."* It used the **stereo** format. Mono yields up to 113×
+  more, and `VOICE_D_453`'s 45 116-byte segment is now its largest sub-wave.
+* *"It matches 1524 of 8021 entries."* The byte-level reach is still 1524, but
+  the **audible** reach is not. Across the 84 movie-bound banks the segment adds
+  **more than 1 s to exactly 7** — the `hokyu_*_H` tankers on `VOICE_D_453`/`454`,
+  precisely the broken ones — and ≤0.25 s to 66 of the rest. The largest
+  non-resupply addition is `S04A` at +0.66 s on a **256 s** movie.
+
+The safety oracle is honest about its own limits: 8 of the 84 banks *already*
+exceed their movie's duration before the change, by hundredths of a second, so it
+cannot resolve differences at that scale. What it does establish is the
+**scoping** — the change is material only where it is meant to be. Callers clamp
+to the movie length regardless.
+
+`VOICE_D_451`'s all-zero leading region is skipped by the non-zero guard, so the
+rule cannot prepend silence to a bank that does not need it; that is pinned by a
+test.
+
 ## ✅ The two parts are SEQUENTIAL SEGMENTS — and that closes the original mystery
 
 The last open question was whether the leading mono region duplicates the RIFF