Is it always possible ... ?
It is worth keeping in mind, documentation is always... a little bit crap.
This is nothing unique or new to software -- it may be most notorious in that domain, right now, but this has always been the case, in software, electronics, or everything else. Documentation is only as good as the time spent making it, effort put into it, feedback received and utilized. Someone has to write the thing, the thing doesn't come out as an automatic one-to-one byproduct of the design process, the design process itself is subject to bugs anyway, it has to be updated from time to time as the design itself changes, and there are very often real business (marketing, IP, etc.) reasons to hide information and oversimplify technical aspects that would otherwise be given in the datasheet.
So, first of all:
Flat out, it is not always possible.
Some datasheets just don't tell you anything at all. There are datasheets that basically acknowledge that a device exists, maybe that it has some pins and package, and that's it. Some may give pin names, maybe voltage and current ratings. Maybe they even give some characteristics and a functional description. The full range of what you will find in datasheets, extends well beyond the relative richness provided by major brands.
Already, you can see that providing a block diagram can almost be... something of a luxury.
To put the finest point in that:
Your presumption that a datasheet contains a block diagram at all, is erroneous, strictly speaking.
Mind, I would ordinarily interpret a statement like this, not as a strict error, but a curable statement. Here, you have the claim: "when X, then Y". But X is not an independent variable, here. X is an optional characteristic of object W. I am inclined to interpret this instead as: "when a W contains X, then Y" -- a contingency rather than a type error.
Which I could refine still further, as "when W contains X with further characteristics Q, R, S.., then...". Namely, that the datasheet W, contains block diagram X, with aspects suggesting current-mode control [Q, R, S..]; and then what we can conclude about it.
But this, too, is an assumption on my part -- and in any case, clearly a miscommunication has already occurred. Maybe that miscommunication is curable, maybe it's reasonable to make an assumption, but that assumption could always be wrong. Feedback is required to ensure reliable communication -- I give exactly this critique to illustrate 1. the exact contours of my denial above; 2. that I do not accept the premise exactly as communicated, and 3. a clearer communication of the claim is desirable.
Getting back to the block diagram, then:
The block diagram itself, is almost certainly simplified, partly for clarity, but also partly because they have IP to hide and they aren't going to reveal exactly how things work internally. A wave-of-the-hands, "it's sort of like this", is very common. It's been a very long time since even equivalent schematics (let alone exact literal ones) were commonly seen -- back in the 60s and 70s when ICs were far simpler than they are today.
To wit: the example you show, is absurd if taken literally. They have the high-side driver tied with a high-side MOSFET and a low-side "current sense", a preposterous (if not actively destructive) arrangement.
What they actually meant, might be something like this:
It is possible to make multi-source MOSFETs; indeed these have been a standard (if niche) component from time to time, but it's particularly easy to create one in an IC. Construction is simple: isolate a corner of the source metallization layer, and bond it to a new pin. When the two source pins are held at the same (or near enough) voltage, their ratio of currents is given by the area ratio*. Another way to express this, is two MOSFETs connected in parallel (D to D, G to G), except for separate S's, and having some ratio of channel width / cross-sectional area between them.
*Power MOSFETs are made with high cell density, evenly distributed across the die; the die area itself can be taken as a uniform continuous area for this purpose. The exact measure will be the cross-section of cells thus connected, but physical die area will be a close analog. This will be more exactly designed (based on process parameters and foundry specs) in an IC.
"Near enough" is relative to ID * RDS(on): the local source region can't pull itself any higher than the drain voltage, so clearly you'll get 100% current error if the source voltage is allowed to rise unconstrained (i.e., not sinking ID Ass / A current from it, where Ass is the second source area and A is total area). Typically, a small shunt resistor (Rsh ? RDS(on) A / Ass) is used, and some 10s to 100s mV is practical in this connection.
Alternately, not sinking current at all, but simply using the second source pin as a sensor for VDS(on), could be used. Note that this does not measure current, per se: RDS(on) has a substantial tempco, and may or may not be usable directly in a control circuit.
In any case, what they might've meant, is something like the above: Q1 actually has a 2nd source terminal, whose voltage is servoed (with error amp EA) equal with the main source, and that current is cascoded down to a ground-referred voltage CS with Q2. (Exactly as shown, this would require a negative supply for EA -- negative relative to SW, that is -- or a depletion PMOS, which, I suppose could be done.)
Other manufacturers show this by simply drawing a circle around the drain-supply wire (as if a clamp-on ammeter), making some kind of hand-wave "it senses current, don't worry about how we managed to do it". Microchip definitely loses points for making such a boneheaded "equivalent" circuit though.
So, we have reason to doubt the literal veracity of the block diagram already.
Is there reason to doubt its structure, too?
I don't think so, not here anyway. The voltage error amp, current sense comparator, PWM latch (with clock SET), level-shift/driver (oh, they don't show level-shifting either, so there you have it again) and output, is a textbook peak current mode buck converter control. They could've glossed over all this and just shown some dumb "control" block, they could've mucked up the error amp and comparator* and not really cared about clarifying any of it, etc. They drew the thing as expected, so they seem to wish to communicate this as fact. Truth is constructed from a collection of related, self supporting facts; they have shown such a pattern here.
*A recent example I saw: JRC's equivalent to MC34063, seemed to show an error amp instead of a hysteretic comparator -- but without compensation network, it likely operates the same way anyway, i.e. voltage-mode hysteretic. I would be inclined to suspect it's actually a lightly-modified clone, not a wholesale improved version (e.g., compare with NCV3063).
Still, though, do we really know?
Of course not. The datasheet is only worth what we paid for it; there's no particular tie of its claims to whatever part we might buy.
Indeed, we might buy parts claiming to be such, on eBay or Amazon or whatever, and find they behave (or even look) nothing like what the datasheet says.
An optical or electron microscope is required to inspect the bare die directly (along with some nasty chemicals and careful handling, to expose the die itself, and probably remove some layers of metal obscuring the fine structure). This is not at all infeasible, but it is more effort than most are going to invest in such an otherwise trivial thing. It's also destructive; you can examine one unit, but who knows what's inside the other thousand on that reel?
Mainline distributors have a vested interest in maintaining brand recognition, product quality, and honoring manufacturers' branding in turn (and the value of their supply contracts). Which does show up in the bottom line (you're paying for legitimate parts, hopefully). Financial incentives are another kind of "proof by work"; somewhere, somehow, work was performed to earn those dollars, and agreeing to spend them in some excess, implies some level of trust in that relationship.
Perhaps that's still not a sufficient degree of knowledge, of trust. How else could we establish a part is what it says it is? Tracing documents, following the parts through the fab, packaging, and various levels of distribution and warehousing, might be considered. This is standard for aerospace grade components, for example. It's (roughly speaking) a reason the military might buy a $200 hammer or etc. All that meta-data tracking who handled it, where and when, isn't cheap -- even with streamlined tools (digital records, RF tracker tags, etc.), it takes effort to do. It also implies a contractual obligation: that a particular customer has such a vested stake in the veracity of these parts, that they are willing to pay for such data, and how to use it -- for example, to be allowed to look up, in distributor's/manufacturer's records, or visit facilities, or etc., under agreeable terms, to confirm that, whatever records they shipped with the part, aren't just randomly generated noise, but are a meaningful and trustable part of the entire manufacturing process.
The better question, then, is: who needs to know? How much does it matter?
I am perfectly fine making a professional decision to use this part (or a similar one) in a commercial application. The level of detail here is sufficient to make a design decision in that context.
Something high cost, high reliability, or high risk (like aerospace as I just mentioned, also various medical devices, and "mission critical" weapons systems), might demand much higher confidence in the part -- you might not actually be paying for anything physical with a $100 regulator IC, it might be a 100% standard commercial part -- but you're not paying for material quality, you're paying for the certainty that, what you bought and installed, is the thing that you specified in the first place.
There could also be legal accountability. Perhaps a sales contract includes terms of, if defective, improper, counterfeit, etc. parts are found, at more than some rate, compensation for the investigation, production down-time, product warranty, etc. will be transferred. Not everything can be signed off in a contract; a company's legal accountability in case of, say, medical device malfunction, can only be deflected so far. Companies generally minimize liability, and it would be foolish for a distributor to accept liability for faulty parts when it's the manufacturer at fault; having a paper trail also supports a legal defense showing where the culpability lies. (Which might simply be at the end of the paper trail, but it could be in the middle as well -- documents can be forged, and forgery is a legally cognizable crime.)
So, putting all that together. A more nuanced question, could be answered affirmatively:
Given a regulator datasheet that provides 1. a description indicating control type, 2. a block diagram supporting claim (1), and 3. measured characteristics consistent with claims (1) and (2), is it always possible to make a design decision, of commercial-grade level of confidence, about the regulator?
To this, I would say yes. But, you will note, this adds whole layers of nuance to the initial prompt.
Pay attention to the details! The details matter!
Cheers, and good luck.
