I'd guess that the battery only powers the settings SRAM, and not the MCU, because a receiver is pretty useless without a main power source. Some receivers used EDLC supercaps for settings retention, powering a SRAM with 2.5 to 4.5 V.
What could be happening is that the settings SRAM gets scrambled when the power fails, confusing the MCU. The MCU isn't programmed to clear the SRAM on reset because user settings are stored there. So for example, if half of the bits in the SRAM get flipped, it doesn't know that the contents are no longer valid.
After being powered off, an SRAM can power on with a random bitpattern, or be influenced partially by the data stored there before.