Phase transitions in GPS C/A signal

I'm reading "Fundamentals of Global Positioning System Receivers" section 7.3 "Maximum data length for acquisition" and it's talking about phase transition:

Theoretically, if there is a navigation data transition, the transition will spread the spectrum and the output will no longer be a cw signal. The spectrum spread will degrade the acquisition result. Since navigation data is 20 ms or 20 C/ A code long, the maximum data record that can be used is 10 ms. The reasoning is as follows. In 20 ms of data at most there can be only one data transition. If one takes the first 10 ms of data and there is a data transition, the next 10 ms will not have one.

In actual acquisition, even if there is a phase transition caused by a navigation data in the input data, the spectrum spreading is not very wide. For example, if 10 ms of data are used for acquisition and there is a phase transition at 5 ms, the width of the peak spectrum is about 400 Hz (2/ (5 × 10−3)).

My understanding was that the C/A signal is a pseudo-random Gold code that is XORed with the data (each data bit being 20*1023 chips long) and the resulting binary signal is BPSK-modulated.

So I would think phase transitions in the C/A signal (caused by the BPSK modulation) would occur much more often, not just between data bits but between chips, every few chips.

What is wrong in my understanding of the modulation of the text?

That text is confusing on its own as the spectrum spreading as transmitted is NOT affected as described. This is suggesting that if you acquire on any arbitrary 20 ms sequence of received data, your acquisition results will vary since a data transition can occur equally likely anywhere in that 20 ms given you select a random starting point for your acquisition. You avoid this whole issue by effectively performing acquisition using a 20 ms sliding correlation on 40 ms of received signal (assuming you want to maximize processing gain over the full 20 ms available).

The BPSK modulation includes the C/A Code specifically making the carrier change back and forth 180°. The C/A code is 1023 chips long; when a C/A code chip is = 1, the carrier prior to data modulation (such as L1 at 1575.42 MHz) is transmitted with 0° phase reference. When a C/A code chip = 0, the carrier is transmitted with a 180° phase reference. The C/A code is at a 1.023 MSps rate, so there are 1023 phase transitions in each 1ms C/A code sequence.

Then there are 20 C/A code sequences in one data bit: Each data bit additionally toggles the carrier between 0 and 180°: So the C/A code is sent 20 times to send a "1" data bit, and the negative of the C/A code (0110... instead of 1001) is sent 20 times to send a "0" data bit. Thus the data rate is 50 bps.

Below is an example of a correlation to SV24 for a single 1023 length sequence associated with that satellite from actual data captured from a GPS antenna with a digital sampling scope. The correlation was done by sliding the sample-matched 1023 length sequence through the received signal that in every other way appeared as white noise. With frequency offsets removed, the result appears as in the plot below showing the correlation peaks every 1 ms at every point in time that the local C/A code aligned with the transmitted C/A code. What we see in the plot is three data symbols (either 1 1 0 or 0 0 1 since phase ambiguity hasn't been removed).  This is the same correlation result plotted on a complex plane (the above plot is the real part of this versus time). This shows how the carrier offset was removed and suggests carrier tracking approaches. 