# Wwise v150 routing support Status: Experimental Scope: `@carbonenginejs/runtime-resource` BNK readers consumed by `@carbonenginejs/runtime-audio` Audience: Runtime-resource and runtime-audio maintainers Summary: Defines implemented dry-output routing and the remaining Wwise bus-processing work. ## Current support `runtime-resource` exposes typed complete v150 Audio Bus, Auxiliary Bus, Fx ShareSet, and Fx Custom bodies and retains every music node's qualified NodeBase. `runtime-audio` consumes the routing records only in its optional builder. Portable SFX sounds and music tracks carry their effective output bus, ordered cycle-safe ancestry, summed authored base Bus Volume, summed Make-Up Gain, and the Output Bus Volume authored by the NodeBase that supplies the effective output-bus override. Playback adds live Bus Volume action state, global Voice/Bus Volume RTPC curves, and global Bus Volume State contributions without changing the application SFX or music sliders. Qualified graph routes apply static Bus Volume, global Bus Volume RTPC, and Immediate State gain at one shared post-effect fader per physical Bus. The route-local stage retains Make-Up Gain, effective NodeBase Output Bus Volume, Bus Volume actions, and collapsed-route ducking; SFX Voice Volume remains a separate pre-Bus stage. The exact SFX Aux shape keeps Voice-target ducking before its split and applies Bus-target ducking after the additive State filter on each complete dry/wet leg. It also projects v150 Audio Bus auto-ducking and coordinates actual scheduled SFX and built-in music source activity through one shared clock. Routed static Wwise Parametric EQ slots are projected into one portable catalog and realized on the corresponding collapsed dry routes. Static Wwise Delay is decoded directly from the portable graph and realized only at its shared Bus. The strict shared mixer also decodes source-proven v150 Wwise Meter records and may omit only audio-transparent instances that cannot feed telemetry back into the graph. It may likewise omit a static user-aux send only when the complete return is proven to remain at or below Wwise's `-96 dB` silence threshold, or realize the narrow exact SFX return shape documented below. ## Music NodeBase contract All four v150 music HIRCs inherit output routing from `CAkParameterNodeBase::NodeBaseParams`. The typed records retain the existing exact music-tail validation and expose the retained prefix as `nodeBase` using these ranges: - Segment, Playlist, and Switch: `[1, anchor)`; byte zero is `uFlags`. - Track: `[headEnd, typeAt)` or `[headEnd, simpleAt)`. The audio builder walks `directParentId` from a leaf and selects the first nonzero `overrideBusId`. Only that same NodeBase contributes Output Bus Volume; a value stored on a descendant that does not override the output bus is not an active routing contribution. Track `iLookAheadTime` is read as a signed `s32`. The music event projection also retains a narrow ordered `music.programs` table for v150 Pause and Resume actions. Qualification requires game-object scope, element or all mode, no Bus flag, exceptions, delay, probability, or randomized transition, standard action flags (Pause 7, Resume 6), and a Wwise curve from 0 through 9. Element targets must be retained music nodes; all mode must target zero. Element execution is currently limited to nodes that are also postable music roots; retained descendants stay fail-closed because one browser music instance cannot pause only a nested Wwise subtree. Repeated actions are not deduplicated because their order and count define nested pause depth. The music scheduler executes this subset by freezing at the authored fade boundary, retaining source offsets, the live playlist iterator, pending preparation, and musical envelope progress, then shifting the retained timeline on Resume. All/game-object controls are also retained in the SFX program because Wwise target zero has no music/SFX discriminator. Events that mix a retained music control with music Play, Stop, or setter actions stay fail-closed until one ordered cross-family program can represent their authored order. ## Audio, Auxiliary Bus, and effect contract For bank version 150, HIRC 8 is Audio Bus, HIRC 18 (`0x12`) is Auxiliary Bus, and HIRC 19 (`0x13`) is LFO. The reader accepts only HIRC 8 and 18 as bus records; type 19 remains an LFO. `inspectBNK()` removes the leading object ID, so the current entry payload starts with: ```text +0 u32 overrideBusId // parent bus; zero means root +4 u32 outputDeviceId // root only ... AkPropBundle: u8 propertyCount u8 propertyIds[propertyCount] u32 propertyValues[propertyCount] ``` For a non-root bus the property count is at `+4`; for a root it is at `+8`. The property-value array is parallel to the ID array. In v150, property `0x04` is finite little-endian float32 Bus Volume. Keep an absent value as `null`, distinct from authored zero. `0x0d` Output Bus Volume and `0x05` Make-Up Gain are separate properties. The BNK reader boundary exposes this typed bus catalog: ```js buses: Map ``` The typed catalog is limited to bank version 150 and consumes the complete bus body exactly. It bounds-checks conditional fields, parallel property arrays, positioning and aux routing, duck rules, ordered effect/metadata slots, RTPCs, and state data. Duplicate or invalid properties, non-finite typed values, truncation, and trailing data reject the typed record while preserving raw inspection with a diagnostic. Absent projected gain values stay `null`, distinct from authored zero. Children are derived by reversing parent links rather than serialized into the record. The same boundary exposes v150 HIRC 16 Fx ShareSet and HIRC 17 Fx Custom records. Their generic typed form preserves plug-in identity, the exact opaque parameter block, media index-to-source mappings, RTPC curves, state values, and initial plug-in property values. The resource layer does not interpret opaque plug-in parameters or media as audio. Across EVE build 3444265's 29 audio banks, the exact-end readers decode all 133 serialized bus bodies and all 410 serialized effect bodies with no failures. The later-bank-wins catalogs contain 130 buses and 381 effects. The effect records cover 16 plug-in classes; the one Convolution Reverb ShareSet maps media index zero to source 154360724, whose `PLUG` payload is opaque plug-in media rather than WEM audio. ## Remaining work The dry-output route retains Bus Volume and Make-Up Gain when those properties occur in the selected dry ancestry. A qualified shared graph moves only Bus Volume to exact per-Bus post-effect faders. It separately applies the effective NodeBase Output Bus Volume and evaluates every qualified scaling-2 Voice Volume and Bus Volume RTPC on the route. SFX Voice Volume owns a distinct gain before Bus Volume and effects; built-in music continues to consume Bus Volume only. The builder keeps raw graph values because Wwise interpolates them before converting `-1` to `-96.3 dB` and other values with `20 * log10(value + 1)`. Keeping these contributions distinct preserves their future placement when real bus and effect stages replace the collapsed gain. Static v150 Parametric EQ and Wwise Delay are the implemented DSP adapters. One bounded source-local EVE-v150 Parametric EQ Game Parameter form is also implemented; it does not make general dynamic effect controls admissible. Feedback-free v150 Meter records have a qualified audio-transparent omission contract, but Meter telemetry is not implemented. One qualified SFX-only static user send is implemented when its neutral-filter Auxiliary return rejoins the dry ancestry; all other audible auxiliary sends and complete ordered effect chaining, dynamic effect controls, nonlinear effects, general wet-path duck placement, effect-tail bus activity, Meter telemetry, virtual-voice behavior, and spatial diffraction remain separate runtime slices. Those features need their complete effective send/property projection, qualified plug-in adapters, and signal semantics; the typed catalogs do not imply that playback is implemented. Bus-target Voice Volume actions remain Voice Volume, not Bus Volume. The bounded runtime form admits only an absolute, non-randomized game-object Set whose target is the first/output Bus of every affected Sound and whose route has no NodeBase or Bus Aux sends. One per-voice gain stores the posting emitter's Bus-keyed state and sits before the route's Bus processing, including for voices posted by a later event. Fixed delay and transition timing are retained. Reset, global, relative/randomized, ancestor-Bus, wet, and music forms remain barriers because they need broader target enumeration or stage-aware insertion inside the Bus ancestry. EVE build 3444265 has 104 Parametric EQ definitions. Exactly two static Custom instances occur on realized dry routes: one reaches 3,717 SFX leaves and applies a peaking band at `-13 dB`, `120 Hz`, `Q=5` plus an authored-neutral high shelf; the other reaches 172 SFX leaves and is entirely neutral. Their combined reach is 3,889 SFX leaves and zero music tracks. Neither routed instance has RTPC, State, property-value, or media controls. Eleven EQ definitions outside the qualified Audio Bus slots do have controls and are not silently promoted; their SFX NodeBase effect slots remain unsupported. The later EVE build 3453885 source-local path resolves the first effective NodeBase effect override for every retained Sound. A descendant override replaces its parent list, an explicit empty override clears it, and a root list is effective even when its override bit is clear. Complete admitted Parametric EQ/Wwise Delay/Matrix Reverb/RoomVerb/Meter chains are projected once per voice before Voice LPF/HPF and the emitter/auxiliary split. The exact demo currently retains 457 qualified EQ Sound leaves and 87 Delay leaves. Of the EQ leaves, 361 are acoustically non-neutral; 170 use the exact live `ship_Roll` frequency form, 17 belong to complete admitted Tremolo/EQ chains, and 24 precede RoomVerb. Sound `350811697` adds the static EQ-to-Tremolo skyhook chain; its `processLfe:false` form is admitted for decoded mono/stereo voices and keeps the complete chain dry above two channels. Other EQ leaves with live controls remain dry-playback approximations. Overlapping property-value, unsupported-plug-in, and other independent-LFE barriers are not partially applied. The builder and shared mixer follow pinned wwiser's version-150 56-byte parameter layout, validate exact boolean bytes and ShareSet/Custom slot identity, preserve bus, slot, and band order, and reject routed Bus `processLfe:false` until an independent browser LFE branch exists. A fully qualified graph route now realizes one ordered Web Audio EQ chain per Bus after SFX spatialization or music route envelopes. Blocked and graphless paths retain the distributed source-route fallback. The field decoding and graph placement are exact, but Web Audio biquads are not claimed to be native Wwise DSP. EVE's reachable EQ follows an active Compressor in Wwise slot order, so its chain remains blocked rather than stacking a nonlinear stage per voice. The source-local exception accepts only the empirically evidenced v150 combination `ParamID 2`, control type Game Parameter, exclusive accumulation, and log-frequency scaling. It maps the curve to Band 1 Frequency and preserves object RTPC, global RTPC, then STMG default precedence. All current EVE records use `ship_Roll` over `0..360`, producing approximately `160..3650 Hz`. Scheduling samples the existing Wwise interpolation over known control transition boundaries. The `ParamID 2` mapping is an exact corpus adaptation, not a published universal Wwise plug-in enum. These records also author `processLfe:false`; they are realized only for decoded mono/stereo voices, where no independent LFE channel exists. Multichannel playback keeps the whole chain dry. Parameters 1 and 7 are driven by Wwise Modulator controls in this corpus. The largest exact unresolved family is ShareSet `1738007123`, inherited by 27 Sound leaves across seven medium/XXL engine-on events. Its first curve is the already-admitted `ship_Roll` Band 1 Frequency mapping. Its second is additive `ParamID 1`, scaling 0, from `+0.2` to `-24 dB`; the v150 band field order and corpus value domain pin this narrowly to Band 1 Gain. LFO HIRC `841531431` drives that gain with 82.6-percent depth at a base `0.2 Hz`, while a nested exclusive `ship_Roll` curve drives LFO Frequency toward `100 Hz`. The object has zero attack, default Sine waveform, two-percent smoothing, and 50-percent PWM. Runtime-resource currently identifies that v150 type-19 object but retains its 96-byte payload raw. Runtime-audio does not duplicate the owning BNK decoder, so this otherwise realizable Oscillator-to-biquad-gain adapter remains blocked until a typed LFO record is exposed. Exact scope sharing, phase/lifecycle, smoothing, and pause behavior must then be qualified before the 27 leaves can leave complete-chain dry fallback. Pinned wwiser proves the v150 Wwise Delay's 18-byte layout: float32 Delay Time, Feedback, Wet/Dry Mix, and Output Level followed by one-byte Enable Feedback and Process LFE booleans. The shared adapter validates the ranges documented by the [official Wwise Delay reference][wwise-delay], requires Process LFE until an independent browser LFE branch exists, and rejects every dynamic, media, or malformed record before allocating nodes. It realizes one shared Web Audio dry/wet split, optional feedback loop, and output gain in authored Bus/slot order. This is an explicit browser DSP adaptation, not a native Wwise claim. The same stage may be voice-owned for a qualified source override. Its post-source feedback tail is deliberately not a lifecycle clock: Web Audio does not expose Wwise's plug-in tail completion, so disposal at decoded dry-source completion cuts residual feedback. Pause and seek likewise reuse browser node state rather than claiming native plug-in-state parity. [wwise-delay]: https://www.audiokinetic.com/en/public-library/2025.1.3_9039/?id=wwise_delay_plug_in&source=Help EVE build 3444265 contains seven active, static, control-free Delay Custom instances on two buses and 14 SFX leaves. Every instance authors a one-second, 100-percent-wet delay with `processLfe:true`; six disable feedback while one enables 45-percent feedback. Every affected dry route later crosses the unsupported root Peak Limiter, and ten also cross audible auxiliary routing, Compressor, and Parametric EQ stages. Delay support therefore unlocks no EVE route by itself: the strict mixer retains all 14 barriers until the complete ordered chain is qualified. EVE build 3444265 authors 60 Bus Volume RTPC curves on 56 buses, driven by 18 Game Parameters. Every one of the 16,263 serialized SFX leaves and 2,484 music tracks reaches at least one affected bus through its dry ancestry. All 60 curves use Game Parameter control type 0, additive accumulation 2, and dB scaling 2, so the current per-route scalar is signal-equivalent on the dry path. The catalog is stored once per bus rather than duplicated on every leaf. The same build authors seven property-0 Voice Volume RTPC curves on seven buses, all driven by `advanced_settings_atmosphere` (`3045458040`). They use Game Parameter control type 0, additive accumulation 2, dB scaling 2, and linear interpolation. Their raw reach is 2,797 SFX references and no music references. Exactly 963 SFX references have no auxiliary send or other barrier and cross only feedback-free Meter effects; those routes now realize the control on a separate pre-bus gain. The remaining 1,834 stay blocked by feedback-capable Meter, auxiliary, unsupported-RTPC, or ordered effect barriers. The same build authors 117 additive in-dB Bus Volume State values across 113 bus/State-Group occurrences on 60 buses and 42 distinct State Groups. Of those buses, 49 occur on current dry routes; all 16,263 SFX leaves and all 2,484 music tracks reach at least one affected bus. Matching groups and distinct buses add their decibel contributions, while a missing State case remains neutral. The portable catalog embeds the exact STMG default and directed transition rules so runtime blending is interruptible and independent of the optional SFX graph. That unified version-2 catalog also retains one Bus Pitch value (`-100` cents), 24 LPF values (`-70..100`) on 18 buses, and six HPF values (`20..45`) on four buses. Pitch reaches 12,678 SFX leaves and no music tracks. LPF reaches 12,909 SFX leaves and 2,482 music tracks; routed HPF ancestry reaches 13,707 SFX leaves and 2,482 music tracks. EVE's STMG filter behavior is additive. Runtime sums signed weighted values across groups and buses before the final clamp, so the negative LPF cases are not discarded early. SFX Pitch uses the existing transport integration; built-in music deliberately ignores Audio Bus Pitch as required by Wwise while applying LPF/HPF on its dry routes. The same build authors 14 auto-ducking source buses and 36 links to 20 unique target buses. Nine source buses occur on SFX dry routes, three occur on music routes, and two are currently unreachable; 33 links can therefore activate. All 20 targets occur on a realized SFX or music route. The target-property split is 24 Voice Volume and 12 Bus Volume rules, with curves 1, 2, 4, 6, 7, and 8, Recovery Times of 0, 1000, or 2000 ms, and fade pairs ranging from 100/5000 through 5000/3000 ms. The portable catalog preserves all 36 rules rather than duplicating them onto 16,263 SFX leaves or 2,484 music tracks. Runtime activity begins at the physical source's scheduled `start()` time, not at event post or media acquisition. It is binary per source bus: overlaps hold one duck and do not stack, while different source buses add. The last source end starts Recovery and then the authored Fade In; reactivation during Recovery cancels release, and reactivation during Fade In rebases into Fade Out. Curves are evaluated with the shared Wwise interpolation table in linear gain. The current Audiokinetic definition treats Maximum Ducking Volume as the collective source-bus floor across its listed targets; a v150 differential Authoring fixture remains desirable because older documentation used target-oriented wording. Voice Volume and Bus Volume remain separate typed target properties. Voice Volume applies before the qualified SFX dry/wet split, while Bus Volume targets are evaluated after State filtering over each complete route leg. A duck source that would place part of its collective maximum floor at Voice Volume and part at Bus Volume fails closed, as does a duck source or Voice target on the wet-only ancestry. This narrow placement proof does not claim general wet-path equivalence: source-side effect tails may extend activity and unsupported Aux shapes remain blocked. See Audiokinetic's [Auto-Ducking reference](https://www.audiokinetic.com/en/public-library/2025.1.3_9037/?id=auto_ducking_tab&source=Help) and [voice pipeline](https://www.audiokinetic.com/en/library/edge/?id=understanding_voice_pipeline&source=Help). All but one occurrence author Immediate synchronization. The exception is State Group `video_overlay` (`2603658559`) on Audio Bus `2609808943`: State `on` (`1651971902`) contributes `-96 dB` and authors Next Grid. That bus reaches 12,678 SFX leaves and no music tracks. Wwise documents that a bus containing only Actor-Mixer Hierarchy sound objects ignores the authored music synchronization point and applies the State immediately. The catalog therefore preserves `syncType: 1` while recording `effectiveSyncType: 0`. The builder rejects any future non-immediate bus State on a music route until a qualified music-grid scheduler exists. See Audiokinetic's [State-change synchronization rules](https://www.audiokinetic.com/en/public-library/2024.1.7_8863/?id=defining_points_within_music_objects_for_state_changes&source=Help). The pinned wwiser source remains the binary-format guide for State property, accumulation, and synchronization enums. Its report layer explicitly does not model dynamic State synchronization timing, so the route-dependent effective behavior above follows Audiokinetic's runtime semantics rather than a report inference. EVE build 3444265 contains three sounds with an effective nonzero Output Bus Volume: `1030460440` (`+3 dB`), `315857869` (`+8 dB`), and `577594853` (`+6 dB`). They all route through bus `3429521127`; its dry ancestry contains only a Wwise Meter, which is signal-transparent. The current scalar realization therefore produces the authored amplitude for these sounds while meter behavior itself remains unsupported. In EVE build 3444265, the two buses that author Make-Up Gain are Auxiliary Buses. No SFX leaf or music track reaches either bus through its dry-output ancestry. A structural walk finds 516 candidate SFX leaves, but eight descendants replace the inherited user-aux list; the qualified effective count is 508. Both routes contain effect processing. The Convolution Reverb also consumes an opaque `PLUG` payload rather than WEM audio, and the complete paths include a Wwise Peak Limiter. Consequently, the implemented dry route retains the correct separate Make-Up Gain contract but does not yet make those EVE values audible. The remaining general aux/effect slice must realize the effective auxiliary routing together with its ordered qualified effect chains; routing those sends as dry parallel copies would not be a parity implementation. The portable `busGraph` checkpoint now preserves the complete reachable topology. EVE 3444265 projects 14 authored bus aux links from ten source buses to five targets. A wwiser-correct NodeBase walk finds 3,645 SFX Sounds with effective direct user sends (4,288 send references); root Actor-Mixer `513700882` is the important inheritance edge because its clear override bit is still effective at the root. Eight descendants replace that list, leaving 80 Sounds feeding Auxiliary Bus `2266223546`, whose authored send to `518379211` is the fourteenth bus link. Music authors no direct NodeBase send, though 2,482 tracks traverse bus `3991942870` and therefore its bus-level send to `1475559705`. The five wet targets divide cleanly. `3845478417`, `3243121821`, and `1475559705` return through a `-96 dB` parent and have only bypassed EQ/Meter slots. Their sends are static, use neutral send filters, and their installed RTPC/State upper bounds cannot amplify the return. The mixer therefore omits only those provably silent paths and allocates no wet nodes for them. This adds 3,177 qualified SFX references and 2,347 music references: totals become 7,596 of 16,255 SFX and 2,349 of 2,484 music, across 50 SFX and seven music route records. Absolute and positive-relative Bus Volume Set actions are projected as `busVolumeMayIncrease` barriers so a future authored action cannot invalidate the proof. Every Set or Reset Bus Volume target also carries the separate `busVolumeActionControlled` marker. The strict mixer rejects that marker on an audible effect ancestry even though static/global Bus gain now has exact post-effect placement: Bus Volume actions are scoped to a playing instance or game object and therefore cannot safely drive one physical fader shared by unrelated voices and music. Otherwise an action could attenuate only new Delay input while an existing feedback tail continued at the old gain. Auxiliary Bus `3235074386` supplies the first exact audible return. Eleven SFX references on routes 83 and 199 author one static neutral-filter send at `-14` or `-20 dB`; their dry and wet branches rejoin at Bus `3161330387`. The route entry fans out after spatialization. Each branch adds State LPF/HPF across its complete ancestry before one final clamp, then applies its complete Bus-target duck contribution before entering the physical buses and sharing the common suffix. Voice Volume and Voice-target ducking remain before the split. This raises the qualified SFX total to 7,607. Route 68 remains blocked: only its wet branch crosses Bus `2609808943`, whose active State authors `-100` cents of Pitch, and one source playback rate cannot pitch only the wet copy. It is exactly one SFX reference, leaf `602217068`; supporting it requires synchronized dry and wet source lanes, not merely relaxing the mixer gate. The admitted EVE leaves are `308284283` on route 83 and `32756389`, `179741258`, `180881900`, `311848135`, `552000032`, `746615656`, `859755539`, `865594407`, `1046944186`, and `1049068182` on route 199. They are reached by `triglavian_controlled_gate_inactive_ambience_play` and the four `medium_structure_amarr_play`, `medium_structure_caldari_play`, `medium_structure_gallente_play`, and `medium_structure_minmatar_play` events. `1912148245` cascades to `518379211`; their audible paths contain ordered Compressor, Parametric EQ, Convolution Reverb, and Peak Limiter stages. The Convolution ShareSet `3019852427` references embedded source `154360724`, a 3,179,376-byte `PLUG` payload in `hangar.bnk`, not WEM. The catalog retains that cross-bank media identity and opaque parameter block. Shared DSP placement and a qualified PLUG decoder remain prerequisites for audible parity. Pinned wwiser decodes the v150 Convolution Reverb's 57-byte parameter envelope, but still labels its final float32 and byte unknown and does not decode `PLUG` media. The EVE payload contains `hash`, `junk`, and `data` chunks; the apparent 48 kHz field inside `data` is not sufficient evidence for a PCM layout. Raw effect media already survives library construction, install, tools-core range acquisition, and HTTP delivery as `application/octet-stream`, so transport is not the blocker. Complete built-in support first needs an independently qualified PLUG-to-impulse decoder in runtime-resource, then a cancellation-safe effect-media preparation seam in runtime-audio before the shared Bus graph is realized. Passing the opaque payload to `decodeAudioData` or treating its data chunk as PCM would be guesswork. The ShareSet affects 508 SFX references on routes 53, 54, and 250. A strict qualification simulation gives Convolution support zero immediate unlocks, both alone and together with hypothetical Compressor and Peak Limiter support, because all three routes still require the general audible auxiliary-return topology and its complete wet-side gain/State placement. The shared-runtime seam resolves immutable route handles. One controller belongs to each enabled audio-system generation and is shared by SFX and music; install-time checks require every playable routed Sound/track to agree with its catalog dry projection. Exact leaf and track IDs survive scheduling, while library disposal invalidates the controller before the backend nodes are disconnected. SFX realization now keeps one lazy branch per exact route handle and spatial mode inside each emitter generation. Same-route 3D voices share a synchronized panner, different authored routes use different panners, and 2D signals remain on a distinct flat branch. Branches retain the old generation's placement, scaling, and object-RTPC snapshot across unregister/re-register and disconnect only after that generation drains. A branch enters shared Bus topology only when the strict mixer qualifies its complete path; every other branch retains the legacy SFX destination. Qualified music tracks now attach through separate per-segment transition lanes and per-instance exact-route lanes. Those stages mirror authored crossfades and Play/Stop fades before the shared music category input, while blocked tracks retain the legacy music path. The system owns a fail-closed shared mixer contract. It allocates stable SFX/music category entries and one shared input node per common ancestor only for strict default-channel paths plus the exact SFX Aux shape above. Decoded positioning and HDR override metadata is neutral only when positioning and HDR are both inactive; unknown flags and active behavior remain barriers. Effect records are decoded from the authoritative graph bytes before any node is allocated. A qualified path may contain a complete static Parametric EQ/Delay sequence plus feedback-free Meter omissions. Each physical Bus realizes one fader after its ordered effects and before its parent; that fader owns its static Bus Volume, global Bus Volume RTPC, and Immediate State gain. Qualification requires the route's authored Bus-volume aggregate to exactly equal the sum of the physical Bus values and requires live RTPC/State readers whenever those catalogs are used. Make-Up Gain, effective NodeBase Output Bus Volume, and Bus Volume actions stay on the existing per-voice or per-track stage. Voice Volume and State Pitch stay source-local. Immediate State LPF/HPF retain one additive whole-route filter pair; the exact SFX Aux path evaluates separate whole dry and wet ancestries, followed by their Bus-target duck gains. Incoming duck targets count as route controls even though the portable graph declares each duck on its source Bus. Unsupported RTPC bindings, dynamic effect controls, media, rendered slots, unsupported auxiliary sends, and mixed/unknown effects remain barriers. Voice-/route-local controls also remain barriers when they share an audible effect Bus or occur on its ancestor side. The one bounded exception is a strict-descendant control below an opted-in Peak Limiter, as detailed in the dynamics section. Qualified SFX branches now connect after their flat/spatial route stage to the shared SFX category input. When analyser support exists, each qualified branch uses its own analyser before that input and `GetGameObjLevel()` sums their sample frames with the legacy emitter analyser. Blocked paths still connect to the legacy SFX destination and allocate no partial mixer graph. Qualified music routes likewise retain route identity through segment and instance envelope lanes; their category volume is applied by the shared mixer rather than the legacy music output. A full build-and-realize audit of EVE 3444265 qualifies 7,607 of 16,255 SFX references and 2,349 of 2,484 music references across its 262 dry-route records. Before silent-aux omission, the dry-only population was 4,419 SFX references and two music references. That SFX subtotal comprised 1,050 effect-free RTPC/State references, 976 RTPC/State references with only feedback-free Meter stages, 76 effect-free RTPC/State/ducking references, and 1,354 RTPC/State/ducking references with only feedback-free Meter stages, plus 963 Voice Volume RTPC references whose effects are only feedback-free Meters. The additional 3,177 SFX and 2,347 music references cross only the three proven-silent static aux returns. The final 11 SFX references use the exact audible return through `3235074386`; no other audible send is rendered. Pinned wwiser proves the v150 Wwise Meter's 28-byte layout: five float32 attack, release, minimum, maximum, and hold values; four one-byte infinite-hold, mode, scope, and downstream-volume fields; and one uint32 Game Parameter ID. The shared mixer accepts exact boolean and Peak/RMS plus Global/GameObject enum values. [Audiokinetic documents](https://www.audiokinetic.com/library/2024.1.0_8669/?id=wwise_meter_plug_in_effect&source=Help) that Meter measures without modifying the signal and that `Apply Downstream Volume` changes which inherited gains the meter displays. By default the mixer therefore omits the telemetry stage when the Game Parameter ID is zero and the effect has no other controls or media, regardless of that measurement flag. EVE's main feedback-free Meter `651869473` reaches 12,678 SFX dry paths, but only the subset with fully projected Voice/Bus Volume RTPC/State controls and no other barrier qualifies. Meter `902247780` similarly qualifies only the subset whose RTPC/State/ducking catalogs are complete. Other reachable EVE Meters write nonzero Game Parameters and remain barriers under the default `wwiseMeterFeedback: "strict"` policy. All active reachable nonzero Meter targets feed audio-observable RTPCs: 34 unique dry-route records and 1,243 SFX references cross those feedback paths. The only globally unconsumed nonzero target belongs to an unreachable bypassed slot and unlocks no route. Meter telemetry itself is not implemented. The 1,243 feedback-Meter references are an overlapping upper ceiling, not an immediate-unlock count. All reachable EVE Meters use Global scope, but both Peak and RMS modes occur and one active record includes downstream volume in its measurement. An exact adapter therefore needs render-thread metering, an audio-rate Meter-to-Game-Parameter feedback path, and downstream-aware gain measurement; main-thread `AnalyserNode` polling cannot preserve its timing. Pinned wwiser proves the record layout, not the detector window or envelope law, so Wwise golden vectors remain required before these stages can leave fail-closed qualification. `wwiseMeterFeedback: "omit-telemetry"` is the explicit audible approximation for the signal-transparent subset: it accepts a static Meter with a nonzero Game Parameter target, passes the audio through the slot, and does not produce the target value. The downstream-volume flag changes the omitted value, not the audible route. Every other qualification gate still applies. Pinned wwiser's v150 property table identifies decimal Audio Bus RTPC parameter `53` (`0x35`) as `MaxNumInstances`. EVE build 3444265 contains one such curve, on bus `3168327127`, mapping control `3108463768` from `0 -> 1` and `100 -> 280`. The bus reaches 328 SFX references and also has a supported Bus Volume RTPC. The builder therefore emits both `"rtpc"` and `"voice-limits"` reasons. `wwiseVoiceLimits: "ignore"` omits the dynamic voice-count/eviction policy; it adds no routes by itself because all 328 paths also cross a feedback Meter, but adds the full 328 when combined with Meter telemetry omission. Together the two explicit policies add 1,186 SFX references in the audited qualification simulation. These are route-admission gains, not newly audible media voices. EVE's reachable ordered graph contains five active 22-byte Wwise Compressors and one active 22-byte Wwise Peak Limiter. All are static, channel-linked, and configured to process LFE. The root limiter is ShareSet `3134687450` on bus `4085017428`: threshold `-1 dB`, ratio `10`, lookahead `0.01 s`, release `0.1 s`, and `0 dB` output. Pinned wwiser proves the limiter layout, and an internal runtime decoder validates that static 22-byte record. Pinned wwiser has no Compressor parameter parser; the coherent Compressor field order therefore remains an empirical v150 corpus interpretation rather than a source-proven adapter contract. The default `wwiseDynamics: "strict"` policy keeps all six stages outside shared routing. The affected media still uses the legacy audible route with authored dynamics omitted. The explicit `"approximate-web-audio"` policy admits the static, linked, Process-LFE subset. Web Audio's `DynamicsCompressorNode` has a fixed 6 ms lookahead, automatic makeup and different detector/envelope behavior, and a maximum ratio of 20 while one EVE Compressor authors 20.1. The adapter uses a hard knee, compensates the mandatory makeup before authored output gain, and pads limiter output latency above 6 ms where possible; it cannot extend the native detector window. This is adapted route qualification, not exact Wwise DSP. The backend's independent safety compressor is not an authored Wwise limiter and is never reused as one. The same runtime policy admits complete source-local Compressor and Peak Limiter overrides. EVE build 3453885 contains 2,114 qualified Compressor Sound leaves across nine complete chain signatures, reachable from 539 retained events. Eight of those leaves place one already-qualified Parametric EQ after the Compressor. Another 73 Sound leaves inherit the single source-proven Custom Peak Limiter `754157063` under `refinery_l_play`. Strict mode omits each complete source chain rather than applying only a supported sibling. The portable library retains the authored base dynamics record, while each opted-in physical voice owns its browser dynamics and limiter lookahead stages. Missing browser primitives keep the voice audible through the dry fallback. Pinned wwiser still does not decode Compressor parameters, so that larger source population uses the same explicitly empirical v150 field order and makes no golden-vector DSP equivalence claim; the Peak Limiter layout is source-proven. Pinned wwiser also proves Wwise Flanger `0x007D0003` as an exact 59-byte v150 record. The builder admits a source-local override only when the whole active slot list is otherwise supported, the Flanger has no media/State controls, and it uses sine modulation with zero phase offset/spread and the Left-Right phase mode. `wwiseModulation: "strict"` leaves that complete chain dry. Explicit `"approximate-web-audio"` realizes the unified-comb topology with Gain and Delay nodes and, when LFO is enabled with nonzero depth, one voice-owned sine Oscillator. Its browser delay is `D * (1 + (depth/100) * sin(2*pi*f*t))`; Blend, feedforward, feedback, Wet/Dry, and output gain retain their authored placement. The oscillator starts once at the first physical source start and remains phase-running through pause/source recreation until final voice disposal. This adaptation is deliberately not Wwise-equivalent. It processes every decoded channel although both qualifying EVE definitions author Center and LFE bypass, clamps feedback to `[-0.999, 0.999]`, does not reproduce Wwise's sample interpolation or channel phase law, and cuts the feedback/delay tail at decoded dry-source completion. Missing required Web Audio primitives omits the entire source chain and preserves audible dry playback. Shared-Bus Flanger is still a route barrier because the shared mixer does not own scheduled-source lifecycle. The exact EVE dynamic exception is not a Wwise Modulator. Four v150 Flanger ShareSets use ordinary Game Parameter control `3712448215`, named `ship_Distance`, while STMG binds that parameter to built-in Distance `1`. Their additive scaling-0 `ParamID 1` and matching property ID `1` drive the decoded Wet/Dry Mix base. Runtime-audio admits only that complete shape and evaluates it per voice from emitter/listener distance. Control type `4`, other bindings, ramps, names, targets, accumulations, scalings, and versions remain barriers. EVE build 3453885 projects nine isolated static-Flanger Sound leaves across `dungeon_brothel_atmo_play`, three `ecx_*explosive*` events, and `worldobject_jumpgate_activity_play`. Eight use ShareSet `2906410516`; one uses Custom effect `290827855`. Another twelve booster leaves place a static Flanger in slot 0 before the admitted live `ship_Roll` Parametric EQ in slot 1, raising the static Flanger population to 21 leaves across nine retained events. The built-in-Distance exception adds 28 leaves across 18 more events: ten idle Tremolo chains, sixteen engine-on EQ chains (ten also with live Tremolo), and two XXL warp-blast leaves. Two M-afterburner leaves still fail closed on companion EQ `1738007123`. The total is 49 Flanger leaves across 27 events. Wwise Tremolo `0x00830003` uses a separate evidence boundary. Pinned wwiser identifies the plug-in and shows the corresponding depth/frequency/waveform, smoothing/PWM, and phase-field sequence inside Flanger, but it does not contain a Tremolo parameter decoder. The EVE corpus supports the inference. The implementation therefore treats EVE-v150's exact 38-byte record as empirical rather than source-proven: depth and frequency floats, waveform integer, smoothing/PWM floats, phase offset/mode/ spread, output gain, and Center/LFE flags. Admission is further bounded to a static bank-version-150 Sine record, an unsmoothed 50%-duty Square, or an unsmoothed Triangle record. Square and Triangle normally require zero offset and zero all-channel spread. One exact EVE OSSE Square preset instead carries 9% smoothing, 15% PWM, Circular phase mode, and 180-degree spread; all admitted records require bounded phase fields and both channel flags enabled. The official authoring order and EVE preset corpus together pin numeric waveform IDs 0, 1, and 2 to Sine, Square, and Triangle. Matching records associated with any other bank version or waveform shape are rejected before projection. The `wwiseModulation: "approximate-web-audio"` adapter maps that subset to a voice-owned Gain/Oscillator stage with `gain(t) = 1 - depth/2 + (depth/2) * sin(2*pi*f*t + phase)`, followed by authored output gain. Nonzero Sine global phase uses a custom `PeriodicWave`; qualified Square and Triangle normally use Web Audio's band-limited native oscillators. One exact 15%-duty OSSE Square preset uses a bounded custom Fourier pulse instead. The record retains phase mode and spread, but the browser applies one carrier to all channels and omits Wwise's per-channel Left-Right, Front-Rear, Circular, or Random distribution. It starts and disposes the oscillator with the physical voice. This preserves the unipolar `[1-depth, 1]` range, global phase, and slot order, but does not claim Wwise's native oscillator/channel law or sample behavior. Smoothing and PWM are normally shape-validated but not retained or applied; Square is admitted only when they are exactly zero and 50 percent. The exact OSSE exception retains 9% smoothing and 15% PWM and applies the duty cycle, but its finite Fourier carrier does not reproduce Wwise's exponential smoothing filter. Its two leaves share mono media, so the authored Circular 180-degree channel spread is inert. Triangle is admitted only with zero smoothing; its PWM is intentionally ignored after range validation because Wwise applies PWM only to Square. Strict mode, missing primitives, other dynamic controls, other waveforms, and shared-Bus Tremolo keep the complete chain audible and dry. Two dynamic EVE-v150 Sine forms are admitted separately. The common form has exactly two object Game Parameter curves sharing `booster_intensity`: additive scaling-0 `ParamID 1` drives Depth and exclusive scaling-3 `ParamID 2` drives Frequency. One `pirate_radio_guristas_play` Sound carries only that Depth curve and retains a static `0.24 Hz` frequency. Its complete admission fingerprint is 72.6% base Depth, 35% smoothing, 50% PWM, 53-degree global phase, Random phase mode, and 55-degree spread. The browser preserves the global phase, omits the per-channel Random spread, and does not reproduce the native 35% smoothing. Those numeric targets and the single-curve preset are corpus-proven mappings, not universal Wwise enums. The builder also requires the matching base-Depth property and STMG ramp type 2 with the exact two-second up/down times. Runtime-audio approximates the control filter independently per voice with Wwise's documented 99.5%-at-authored-time exponential law, samples the admitted curves, clamps Depth to `0..100` and Frequency to `0.02 Hz..Nyquist`, and automates the oscillator plus both unipolar-gain terms. A voice created during an existing object's filter ramp begins at the currently readable target rather than reconstructing object-level history. Set Game Parameter transitions are likewise resampled as targets rather than being convolved with this plug-in-local filter. This is still the same native-oscillator approximation; missing metadata/readers/primitives keep the whole chain dry. EVE build 3453885 projects 190 Tremolo stages on 186 Sound leaves. Twenty-one dynamic leaves cover the original medium-engine set, ten newly complete engine-on chains, and the single Depth-controlled Guristas radio loop. Two exact OSSE Square leaves add the Amarr and Minmatar big-screen events. The static population spans 91 retained events: 117 Tremolo-only, 14 Tremolo followed by qualified static EQ, three qualified static EQ followed by Tremolo, eight Tremolo followed by Matrix Reverb, six Delay followed by Tremolo, two Tremolo followed by Delay, one double-Tremolo chain, three Square-carrier leaves under `phased_asteroid_impossible`, and one four-stage chain under `jita_OSSE_bigscreen_1_play`. The latter contains Triangle effects `566093542`, `579691495`, and `672599830` around Sine effect `126257322`; all are static, control-free, and zero-phase. They use 70 effect identities and 24 exact parameter records. Sound `350811697` contributes the static skyhook EQ-to-Tremolo chain. Its `processLfe:false` EQ realizes for decoded mono/stereo audio, where there is no independent LFE channel, while multichannel playback remains wholly dry. Ten additional static Tremolo-to-distance-Flanger leaves cover six idle engine events. Admitting bounded phase fields raises qualified source-effect leaves from 2,666 to 2,740 without changing media or event reachability. Wwise Guitar Distortion `0x007e0003` is source-proven separately. Pinned wwiser decodes six 17-byte EQ records followed by distortion type, Drive, Tone, Rectification, output gain, and Wet/Dry mix. EVE build 3453885 has 69 retained Sound leaves across 23 events using 18 static effect identities and 12 exact raw records (11 audible decoded parameter sets). All are v150, control-free, fully wet, and use Overdrive or Heavy. Another 136 leaves across 35 events use three dynamic effect identities. Their exact common shape is an object Game Parameter, additive accumulation, scaling 0, property and RTPC `ParamID 61`, with the property value equal to base Drive. The controls are `ship_health_hull` (111 leaves), `ship_warp_direction` (one), and `booster_intensity` (24). ParamID 61 is pinned empirically to Drive only for this EVE-v150 corpus; pinned wwiser does not publish a plug-in RTPC enum. `wwiseDistortion: "approximate-web-audio"` maps enabled pre/post bands to authored-order biquads around a 4x-oversampled WaveShaper. Dynamic Drive uses scheduled pre/post Gain nodes around a fixed maximum-Drive curve. For normalized WaveShaper inputs this preserves the existing static approximation family without rebuilding the curve. Its normalized tanh curve and Rectification blend are deliberate CarbonEngineJS approximations; wwiser establishes no native transfer, Drive scaling, Tone law, oversampling, or channel behavior. Tone is retained but currently inert. Strict mode, missing primitives, non-v150/non-fully-wet records, unsupported distortion types, other dynamic shapes, a missing live RTPC reader, and shared-Bus Guitar Distortion keep the complete chain audible and dry. Static admission raised qualified source-effect leaves from 2,506 to 2,575. Live Drive raised the post-Meter aggregate from 3,043 to 3,179 leaves and from 3,201 to 3,337 records without changing media or event reachability; the later modulation and skyhook-EQ slices set the prior totals to 3,196 and 3,358; built-in-Distance Flanger and the single-Depth Tremolo now set the current totals to 3,225 leaves and 3,423 records. Wwise Harmonizer `0x008a0003` remains a measured DSP barrier rather than an unclassified plug-in. Pinned wwiser proves the v150 layout for two pitch voices, voice-local filters and gains, input mode, dry/wet levels, window size, Process LFE, and dry synchronization. EVE build 3453885 has 101 effectively reachable Harmonizer Sound leaves across 12 retained events. Ten static leaves belong to `sun_yellow_play`; another 90 belong to the five `sun_*` and five `trig_sun*` events, and one belongs to `warp_ship_init_play`. The 100 sun leaves enable one zero-cent voice, disable voice two, mix dry and wet at `0 dB`, and synchronize dry latency. Zero pitch does not make this a transparent effect: the processed windowed branch and latency-aligned dry branch still define the sound, while Web Audio exposes no duration-preserving pitch-shift node. The 90 dynamic leaves also use Envelope Modulator `53388567`: object scope, 1.674-second attack, curve 49, 1-second decay, 100-percent sustain, and 0.268-second release. It drives additive dB plug-in `ParamID 3`, but pinned wwiser supplies no Harmonizer RTPC enum and the corpus does not distinguish Voice 1 Gain from another gain-like target. The runtime must not guess that mapping. Effect `100527105` is the narrow exception: its wet level is `-96 dB`, so its `booster_intensity` RTPC cannot audibly alter the dry signal and omission is safe. It reaches only the single warp Sound. Since existing dry fallback already produces that audible result, projecting a one-record no-op would add metadata rather than fidelity. Harmonizer therefore remains fully unprojected until a controlled Wwise parameter pair and custom voice-owned pitch-shift DSP exist; all affected media remains audible through the complete-chain dry fallback. Pinned wwiser proves Matrix Reverb `0x00730003` and its v150 default-delay record: float32 Reverb Time and HF Ratio, uint32 delay count, float32 Dry and Wet levels and Pre-Delay, a Process-LFE byte, and uint32 delay mode. The exact static form is 29 bytes; custom mode appends one float32 per delay and remains unsupported. The builder admits only control/media-free default-mode records, Process LFE, and standard 4/8/12/16 delay counts. `wwiseReverb: "approximate-web-audio"` realizes that source-local record as a bounded four-line cyclic feedback-delay network. It preserves Dry/Wet and Pre-Delay, estimates nominal T60 from Reverb Time, and maps HF Ratio to a fixed logarithmic low-pass curve. The authored delay count stays in metadata because the reduced browser topology does not reproduce Wwise's proprietary matrix, mixing, damping, channel, or LFE laws. Strict mode, missing primitives, dynamic/custom-delay records, and shared-Bus placement retain complete-chain dry playback. Natural source completion disposes the network and cuts its remaining tail. EVE build 3453885 contains and projects all 50 Matrix Reverb Sound leaves across 22 retained events and five effect identities. Forty-two are Matrix-only. The other eight place a Tremolo ShareSet before Matrix Reverb; the portable Tremolo retains its 108-degree global phase, Random mode, and 66-degree spread, while the browser explicitly omits the Random per-channel distribution. Every Matrix record itself is v150, 29-byte, 12-delay, default-mode, Process-LFE, and control/media-free. The initial Matrix slice raised qualified source-effect leaves from 2,575 to 2,617; the later Tremolo phase admission completed the remaining chains without changing media or event reachability. Pinned wwiser proves RoomVerb `0x00760003` and its exact 186-byte v150 record. It contains Decay Time, HF Damping, Diffusion, Stereo Width, three tone-filter records and insert positions, input/output channel levels, Dry/Early/Late levels, the early-reflection pattern and timing, Reverb Pre-Delay, Room Size, Density, Room Shape, quality/reverb-unit count, and the algorithm-tuning tail. The builder admits only static, control/media-free records with the exact audited EVE tuning fingerprint and mono/stereo-compatible channel levels. `wwiseRoomVerb: "approximate-web-audio"` realizes those source-local records with two deterministic cached procedural convolution branches rather than the Matrix feedback network. The early branch approximates the reflection pattern, room-size timing, and stereo width. The late branch preserves Pre-Delay and approximates nominal T60, time-varying HF damping, diffusion, density, room shape, and quality. Authored Dry/Early/Late levels are preserved, and enabled tone filters map to Web Audio biquads at their retained branch positions. This is not Audiokinetic's proprietary reflection table, reverb-unit algorithm, early-reflection front/back timing, surround/LFE/center routing, or channel law. Strict mode, missing primitives, a decoded source above two channels, dynamic records, and shared-Bus RoomVerb keep the complete chain audible and dry. Voice disposal cuts the remaining convolution tail. EVE build 3453885 projects 52 RoomVerb Sound leaves across 34 retained events. They use 11 effect identities representing ten exact parameter records. Every record has an active late tail; 50 also have audible early reflections. Twenty- eight chains contain only RoomVerb and 24 place a qualified static Parametric EQ before it. Decay ranges from 1.2 to 6.7 seconds, Stereo Width is 88 or 180 degrees, and two single-leaf records use active tone shaping. RoomVerb raised the qualified source-effect population from 2,740 to 2,792 Sound leaves and the retained record count from 2,781 to 2,857; the later live-EQ block raises the totals to 2,962 leaves and 3,039 records without changing media or event reachability. The same source-local projection retains pinned wwiser's exact 28-byte v150 Wwise Meter record when it is static and control-free. A Meter without a Game Parameter target is an exact signal-transparent omission. A target-bearing record keeps the whole source chain dry under the default `wwiseMeterFeedback: "strict"`; explicit `"omit-telemetry"` lets audible sibling effects run in authored order while allocating no Meter node and producing no Game Parameter value. This reuses the shared-Bus policy rather than baking host policy into the portable library. EVE build 3453885 has 130 such Sound leaves. The original 49 span 39 retained events; another 81 across 51 events use Meter `277510878`, whose downstream- volume flag changes the omitted measurement, followed by Compressor `554802347`. Retaining that complete chain raises the current aggregate after Tremolo, Matrix, RoomVerb, live EQ, and this correction to 3,043 leaves and 3,201 records; the later live-Drive admission raises those totals to 3,179 and 3,337 without changing media or event reachability. Bounded static Tremolo admission raises those totals to 3,183 and 3,344; the paired dynamic Tremolo slice raises those totals to 3,193 and 3,354; the exact OSSE Square slice raises those totals to 3,195 and 3,356; the static skyhook EQ-to-Tremolo slice set the prior totals to 3,196 and 3,358; built-in-Distance Flanger and the single-Depth Tremolo now set the current totals to 3,225 and 3,423. Playback still requires both explicit policies: Meter telemetry omission and Web Audio dynamics approximation. The omitted `sovhub_upgrades_meter` Game Parameter feeds a cross-bank Voice Volume RTPC on Structures actor-mixer `572768013`, so the structure subtree loses up to about `0.94 dB` of authored ducking. This is an audible opt-in approximation, not an exact omission. A fail-closed qualification simulation that treats only these dynamics stages as supported, while leaving every other route gate intact, bounds their EVE build 3444265 payoff. Compressor support alone adds all 135 remaining music references and no SFX. Peak Limiter support alone adds 36 SFX references. Supporting both adds 144 SFX and 135 music references, moving combined adapted qualification from 9,956/18,739 (53.1%) to 10,235/18,739 (54.6%) and bringing built-in music to 100%. Of the Compressor-reachable SFX, most still cross independent Aux or convolution barriers, so the raw reach of 3,889 SFX is not an attainable dynamics-only gain. EVE build 3453885 also contains four SFX route shapes where all route-local controls belong to bus `1017660616` or one of its strict descendants and the only audible ancestor effect is root Peak Limiter `3134687450` on bus `4085017428`. The source/route filters, Voice Volume, Bus actions, and duck source activity already run before that shared root stage; the physical Bus faders remain on their existing sides of the limiter. None of these routes has an incoming duck target or audible Aux send. The mixer therefore admits this specific order under `wwiseDynamics: "approximate-web-audio"` while continuing to reject same-Bus/ancestor controls, incoming duck targets, and every other effect family. The four routes cover 36 SFX leaves and 24 media across `Ambience_Hangar_Caldari_Play`, `jita_hangar_play`, `Ambience_Hangar_Gallente_Play`, `OSSE_Gallente_steamalter_play`, `chjita_initial_docking_entrance_lights_play`, and `Ambience_Hangar_Minmatar_Play`. Only `OSSE_Gallente_steamalter_play` becomes fully shared-routed; the other five retain additional blocked branches. This is a stage-order qualification gain, not new media/event audibility, and the Peak Limiter remains the documented Web Audio approximation.