No that’s not how a buck driver works, it does not “shed off” the higher voltage. It is a DC-DC converter that converts the cell voltage to the lower one required by the LED, typically with good efficiency, 90% or more is possible. The output to the LED is smooth DC, with a little ripple. They control the LED brightness by measuring the current through it, usually through a low-value sense resistor, and regulating it to the desired level by modulating the operation of the conversion electronics. Basically they are a mini switch-mode power supply with variable current regulated output.
Buck converters work the same way, but boost the voltage. Necessary to e.g. run a higher voltage LED from a single cell, or a chain of LEDs in series.
Boost-buck converters can operate either way, useful for e.g. a torch designed to accept a variety of cells, such as a single 1.2V NiMH, 1.5V alkaline, 1.7V primary lithium, 3-3.2V primary lithium, 4.2V secondary LiIon, etc. in one cell, two cell or greater multiples using extension tubes.
Being mini SMPS they are complex, require bulky specialised magnetic components, and require considerable skill, and good expensive test equipment to design.
BLF derived drivers such as this one are far simpler things, that just do one job, with one cell type.
7135 constant current drivers are used at the lower levels. Each supplies 350mA. These are PWMed at high (invisible) frequency to reduce the output for lower levels.
There are limitations on how fast they can be PWMed, at very short pulse widths the rise and fall times of the output become significant. Some makes, even batches, are much better than others.
This design uses two banks of 7135s. A single one for the lowest levels, down to moonlight, and a set of 14 for the higher ones (it does have 18 LEDs).
For continuous use, without over heating, that is sufficient.
For the highest levels a FET switch is used to directly connect the cell to the LED, PWMed or DC. There is no current control, the LED has to burn off any excess power as heat, within its junction. Commonly the LED is than operating beyond it’s peak efficiency, or manufacturer’s current and power ratings. PWMing it doesn’t alter this. Still they work. FET lifetime may be reduced, but that’s not usually significant for most. And those that care about fading will probably already have swapped the LEDs for newer ones before then.
When the FET is in use the brightness is un-regulated, and tracks the cell voltage as it discharges, sometimes becoming lower than what is achievable by the 7135s, well before the cell is substantially depleted.
E.g. on this, the LEDs are driven at 290mA each by 7135s. Under FET operation, that will be far higher. Cell internal resistance and peak current capability now become important factors.
Hence our keen interest in the LED transfer characteristics and evaluations posted here by some experts. Cell, springs, wires, FET choice, improving current path, thermal path from the LED, DTP copper MCPCBs, tail current measurement etc.
Boost/Buck drivers can overcome some of these concerns by compensating for many of these losses, controlling LED current through the cell discharge, and, done well, being more efficient (better run time).
These are selling points for commercial torches, where tables of Lumens vs. run-time are studied by consumers.
The MF01 for example has a sophisticated boost driver, though the new version is to have a BLF architecture, which will doubtless reduce manufacturing costs, which should be good for the consumer.
This is a tried and tested architecture that works well, is friendly for firmware developers, uses the minimum of inexpensive readily available, components, understandable, assembly is practical for DIY builders, modders and repairers even with the minimum of tools, and can be easily copy-pasted by any aspiring PCB layout engineer or manufacturer to develop their own variants, using a basic 2-layer PCB.