That specific story might not entirely work as-is, but that was an example. You could address it as a fairy tale or urban legend or as a similar account. Someone else seeing something similar doesn't necessarily have to directly relate. It could be two separate issues with a common solution. The other person could have just been seeing things while you could actually have it be something.
For example, with my story example, the other person could've actually been seeing someone actually there, but who moved out of sight when they turned, perhaps hinting at a secret passageway or such.
There are definitely more ways to do it, I just gave a few suggestions/ideas.
Another option would be to assume that the players are going to fail the 'puzzle' and brute force it. For this, you would want to try to balance the items so that the player would really have to try to fail. Make it so that all the items are viable purchases, not just the chocolate bar.
Also, I managed to hardlock myself in a different way. Wasted all my money on health recovery food and not sanity recovering food. Went into the mirror hall, saw Momoko turned around, saved, and exited. Now I can't get past because all my saves are impossible to beat. That actually would have been an interesting solution if you could escape the room and have the Momokos despawn and stay despawned, but all the Momokos respawn on re-entry.
I also want to mention that sometimes clues come in forms that you might not expect.
Like Demon Souls and Bloodborne, the enemy and boss room designs were also clues on how to beat the enemies and what attacks the enemies would use. If you paid attention, just looking at the boss areas pre-boss could tell you how to fight some bosses.
I mentioned the ghost thing with having the clues being part of gameplay.
It can also be leading the eye. This is done by making it so that your eye naturally rests on a certain point, by using very specific patterns in the texture, etc.
It can also be training. Slowly teaching the player to do a certain set of actions. This does mean avoiding doing the opposite of this training though.
There is also association. For example, you find a bow and arrow and see a target, you automatically think: bow and arrow == shoot target. If you show a moving red line of light that seems to be looking around, then on finding the player, starts to follow them. Many players that have played certain stealth games or movies will assume laser/sniper and look for cover. The important thing to note with this is that it only works if you can reasonably expect the playerbase to all have the associations. If you show a laser line in a game to a group of people who have never heard of a laser or laser sight before, it just becomes a moving red line.
Related to association, even something like a name can be a clue. For example of a vague one, Autumn. You find yourself at a cliff. Now, another way to say Autumn is Fall. A more concrete example would be Ariadne. You see a spiderweb on a door and you have the name Ariadne. Ariadne gave a thread, spiders spin threads. As can be seen, the problem is that name clues are usually weak clues. (That doesn't have to be the case, but it relies heavily on associations). This can be made stronger by attaching those names to something more concrete, such as calling a boss 'Achilles' and giving it a weak point of the heel.
That said, there are games, especially older games, that don't do a good job at providing the clues, but there is really no reason to copy the bad parts of older games. If we wanted to do that, we'd get rid of saves and have the player find codes that they can input to start the game at specific points, use horrible camera angles to make the game harder, ignore accessibility and make the game hard for colorblind and deaf players to play, have horrible controls that people just get used to after several iterations, etc.
