From b5e6f638812abd6b3058bac8212071ee3d4010b9 Mon Sep 17 00:00:00 2001
From: Lucas Winther <lucwin@ece.au.dk>
Date: Fri, 30 Oct 2026 03:19:40 +0100
Subject: [PATCH] test: tighten demucs separation assertions (ruling R7)

Bass absolute SDR floor, guitar/piano order pinned by stem energy,
finite + non-silent checks on the mono and 3 s slice paths, and the
test.failing comment cites ruling R7.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 tests/pipeline/separate.test.ts | 35 +++++++++++++++++++++++++++++++--
 1 file changed, 33 insertions(+), 2 deletions(-)

diff --git a/tests/pipeline/separate.test.ts b/tests/pipeline/separate.test.ts
index 051ea09..fed24b8 100644
--- a/tests/pipeline/separate.test.ts
+++ b/tests/pipeline/separate.test.ts
@@ -45,6 +45,23 @@ let out: Record<Stem, Audio>;
 let secsPerMin = 0;
 const sdrs: Record<string, number> = {};
 
+function energy(a: Audio): number {
+  let e = 0;
+  for (const ch of a.channels) for (const v of ch) e += v * v;
+  return e;
+}
+
+/** Every sample finite, and the stems together carry a real share of the input's energy (not silence). */
+function expectFiniteNonSilent(input: Audio, stems: Record<Stem, Audio>): void {
+  let total = 0;
+  for (const s of STEMS) {
+    for (const ch of stems[s].channels) for (const v of ch) if (!Number.isFinite(v)) throw new Error(`${s}: non-finite sample ${v}`);
+    total += energy(stems[s]);
+  }
+  expect(energy(input)).toBeGreaterThan(0);
+  expect(total).toBeGreaterThan(0.1 * energy(input));
+}
+
 function sum(...as: Audio[]): Audio {
   const len = as[0]!.channels[0]!.length;
   return { sampleRate: 44100, channels: [0, 1].map((c) => {
@@ -64,6 +81,9 @@ beforeAll(async () => {
     sdrs[`${stem}_mixBaseline`] = sdr(oracle(role), data.mix);
   }
   sdrs.guitarPlusPiano = sdr(oracle("guitar"), sum(out.guitar, out.piano));
+  sdrs.guitarRefVsPianoStem = sdr(oracle("guitar"), out.piano);
+  sdrs.guitarStemEnergy = energy(out.guitar);
+  sdrs.pianoStemEnergy = energy(out.piano);
   console.log(`separation t1b/1: ${secsPerMin.toFixed(1)} s per minute of audio`, sdrs);
 }, SLOW);
 
@@ -82,12 +102,21 @@ test("separates t1b/1: drums > 3 dB, bass beats the mix, guitar(+piano) > 3 dB a
   expect(sdrs.drums!).toBeGreaterThan(3);
   expect(sdrs.drums!).toBeGreaterThan(sdrs.drums_mixBaseline!);
   expect(sdrs.bass!).toBeGreaterThan(sdrs.bass_mixBaseline!);
+  expect(sdrs.bass!).toBeGreaterThan(0.5); // absolute floor: a silent estimate scores 0 dB
   expect(sdrs.guitarPlusPiano!).toBeGreaterThan(3);
   expect(sdrs.guitarPlusPiano!).toBeGreaterThan(sdrs.guitar_mixBaseline!);
 });
 
-// Brief's literal criterion. Unmet by the model on this song (guitar 1.85 dB, bass 1.67 dB); kept as an
-// expected failure so it flips loudly if separation improves. Needs a controller ruling.
+test("separates t1b/1: guitar and piano stems are not swapped (guitar stem carries more energy)", () => {
+  // Pins the manifest stem order (… guitar, piano). Energy, not SDR: the piano stem's SDR against the oracle
+  // guitar (2.55 dB measured) exceeds the guitar stem's (1.85 dB) because both hold parts of the same guitar.
+  // Measured energies: guitar 19521, piano 12721.
+  expect(sdrs.guitarStemEnergy!).toBeGreaterThan(sdrs.pianoStemEnergy!);
+});
+
+// Brief's literal criterion, relaxed by controller ruling R7: HTDemucs-6s routes part of the SONiVOX GM guitar
+// into its piano stem (synthetic-timbre gap), so guitar is 1.85 dB and bass 1.67 dB on t1b/1. Kept as an expected
+// failure so it flips loudly if separation improves.
 test.failing("brief criterion: guitar and bass stems each > 3 dB and > mix baseline", () => {
   expect(sdrs.guitar!).toBeGreaterThan(3);
   expect(sdrs.guitar!).toBeGreaterThan(sdrs.guitar_mixBaseline!);
@@ -100,6 +129,7 @@ test("a 3-second slice (shorter than one segment) returns 3 s stems", async () =
   const slice: Audio = { sampleRate: 44100, channels: data.mix.channels.map((c) => c.slice(44100, 44100 + n)) };
   const out = await separate(slice);
   for (const s of STEMS) for (const ch of out[s].channels) expect(ch.length).toBe(n);
+  expectFiniteNonSilent(slice, out);
 }, SLOW);
 
 test("mono input is accepted and returns stereo stems of the same length", async () => {
@@ -110,6 +140,7 @@ test("mono input is accepted and returns stereo stems of the same length", async
     expect(out[s].channels.length).toBe(2);
     expect(out[s].channels[0]!.length).toBe(n);
   }
+  expectFiniteNonSilent(mono, out);
 }, SLOW);
 
 test("rejects sample rates other than 44100", async () => {