AVAudioFile.processingFormat, only Float32 is allowed?

Here is some code I have to create an AVAudioFile instance based on Int16 samples.

let format = AVAudioFormat(commonFormat: .pcmFormatInt16, sampleRate: 44100.0, channels: 2, interleaved: false)!
let audioFile = try AVAudioFile(forWriting: outputURL, settings: format.settings)

When writing to the file I get the following runtime error, presumably from CoreAudio.

CABufferList.h:184 ASSERTION FAILURE [(nBytes <= buf->mDataByteSize) != 0 is false]:

I read this as a size mismatch between what is specified in the format used to create the file and the file's own internal processingFormat property, which is read-only. Here is my debugger console output showing the input format I created, along with the resulting AVAudioFile fileFormat and processingFormat properties.

(lldb) po format
<AVAudioFormat 0x300e553b0:  2 ch,  44100 Hz, Int16, deinterleaved>
(lldb) po format.settings
▿ 7 elements
  ▿ 0 : 2 elements
    - key : "AVNumberOfChannelsKey"
    - value : 2
  ▿ 1 : 2 elements
    - key : "AVLinearPCMBitDepthKey"
    - value : 16
  ▿ 2 : 2 elements
    - key : "AVFormatIDKey"
    - value : 1819304813
  ▿ 3 : 2 elements
    - key : "AVLinearPCMIsNonInterleaved"
    - value : 1
  ▿ 4 : 2 elements
    - key : "AVLinearPCMIsBigEndianKey"
    - value : 0
  ▿ 5 : 2 elements
    - key : "AVLinearPCMIsFloatKey"
    - value : 0
  ▿ 6 : 2 elements
    - key : "AVSampleRateKey"
    - value : 44100
(lldb) po audioFile.fileFormat
<AVAudioFormat 0x300ea5400:  2 ch,  44100 Hz, Int16, interleaved>
(lldb) po audioFile.processingFormat
<AVAudioFormat 0x300ea5450:  2 ch,  44100 Hz, Float32, deinterleaved>

Please note that the input format I'm using does not match either the audio file fileFormat or processingFormat properties. The file format is interleaved even though I specified de-interleaved. This makes sense to me as working with audio files that are growing is much easier and more efficient with interleaved data. The head-scratcher is the processingFormat. I specified Int16 samples and it is expecting Float32? According to the format settings dictionary, we are specifying the correct key/value pairs.

Is this expected behavior? Does Apple always insist on Float32 internally or is this a bug?