Add server-initiated-volume-curve scenario (#64)

README.md

@@ -18,6 +18,7 @@ Current scenarios:
 - `server-initiated-external-source` (Client enters and leaves `external_source` mid-stream): start the server first, then the client. The client transitions to `external_source` and back; the harness verifies the server emits the right `group/update` and `stream/end` and that the previous group is restored. Per the audit only SendspinKit exposes a client-side API; no client in the matrix declares the `external-source-client-api` capability yet, so every case fails fast
 - `server-initiated-multi-server-arbitration` (Client arbitrates between two connected servers): two servers connect to one client with different `connection_reason` values; the harness verifies the client sends `client/goodbye` with `'another_server'` on the right socket and ends up speaking to the right server per the spec's decision table. Per the audit only sendspin-cpp and sendspin-cli implement these rules; no client in the matrix declares the `multi-server-arbitration` capability yet, so every case fails fast
 - `server-initiated-static-delay` (Client applies configured `static_delay_ms`): start the server first, then the client. The client is configured with a non-zero `static_delay_ms`; the server pre-compensates and the harness asserts the client's actual emission time matches `T − static_delay_ms` within the spec's drift tolerance. Per the audit, `sendspin-rs` and `SendspinKit` parse the field but never apply it (highest-severity bug); seven SDKs don't persist across restarts. No client in the matrix declares the `static-delay-applied` capability yet, so every case fails fast
+- `server-initiated-volume-curve` (Server calibrates client volume curve): start the server first, then the client. The server sets `volume = 50`, the client plays a known fixture, and the harness measures the peak amplitude and asserts the spec's perceptual curve (`0.5^1.5 ≈ 0.354`, ≈ −9 dB) vs. the non-conformant linear gain (≈ −6 dB). Per the audit, sendspin-go and sendspin-js apply linear `vol/100` gain today. No client in the matrix declares the `volume-perceptual-curve` capability yet, so every case fails fast
 
 ## Current coverage
 

src/conformance/implementations.py

@@ -69,6 +69,7 @@
                 "external-source-client-api",
                 "multi-server-arbitration",
                 "static-delay-applied",
+                "volume-perceptual-curve",
             ),
         ),
     ),

src/conformance/scenarios.py

@@ -196,6 +196,26 @@
 )
 
 
+SERVER_INITIATED_VOLUME_CURVE = ScenarioSpec(
+    id="server-initiated-volume-curve",
+    display_name="Server calibrates client volume curve",
+    description=(
+        "Start the server first, then the client. The server sets `volume = 50`, "
+        "the client plays a known fixture, and the harness measures the peak "
+        "amplitude of the recorded output and asserts it matches the spec's "
+        "perceptual curve (`0.5^1.5 ≈ 0.354` × original, ≈ −9 dB). A "
+        "non-conformant linear-gain client outputs ≈ `0.5` × original (≈ −6 dB). "
+        "Per the audit, sendspin-go and sendspin-js apply linear `vol/100` gain "
+        "today; sendspin-jvm and sendspin-dotnet need follow-up audits."
+    ),
+    initiator_role="server",
+    preferred_codec="pcm",
+    required_role_families=("player",),
+    verification_mode="capability-only",
+    required_capability="volume-perceptual-curve",
+)
+
+
 SERVER_INITIATED_STATIC_DELAY = ScenarioSpec(
     id="server-initiated-static-delay",
     display_name="Client applies configured `static_delay_ms`",
@@ -290,6 +310,7 @@
     SERVER_INITIATED_EXTERNAL_SOURCE,
     SERVER_INITIATED_MULTI_SERVER_ARBITRATION,
     SERVER_INITIATED_STATIC_DELAY,
+    SERVER_INITIATED_VOLUME_CURVE,
 )
 
 SCENARIOS: dict[str, ScenarioSpec] = {scenario.id: scenario for scenario in SCENARIO_LIST}