In the grammar of a cryptic clue, the fodder — the letters that need to be anagrammed, or the words among which an answer is hidden, or the strings from which we want the first letter or last letter or heart — is actually a noun. In the surface reading those words might be verbs or adjectives or a whole independent clause, but as fodder, it's a noun.
The cryptic reading of the clue has to make grammatical sense just like the surface reading, and so when the cryptic reading has this noun it wants you to modify, there are a couple of clear paths: verbs and adjectives. But you may also come across nouns, and a noun by itself is almost never an indicator. It's easy to see in an acrostic-type clue, such as "Cease showing this obscenity," petitioned leaders (4) for STOP. No doubt that Cease is the definition and that S, T, O and P are the leaders of the subsequent words … but under almost no conception of the English language is showing leader (or, nota bene, leader showing) a way to say S. Showing's leader, sure. Leader of showing, absolutely. But in natural language, showing leader to mean S is tin-eared with nickel hearing aids.
Now, like all Tap the Sign guidelines … is it fine once a puzzle? Only to the extent that any other kind of English language abuse is OK. As Steve Mossberg has observed, one thing that distinguishes UK cryptics is that they have a huge solver base as well as a very large mature solver base, and just like you have to master the rules before you can break them, only advanced solvers are going to have the gimlet-eyed judgment to see how your English language abuse is (rare case) actually clever and not just (common case) hand-wavey clue writing.
In the US, however, we don't have a huge solver base. In our view, any puzzle could be someone's first puzzle, which doesn't mean every puzzle has to be for beginners, but cryptic readings that overrely on cryptic crossword conventions — which are de facto hand waving — as opposed to plain*-spoken** English, can make a puzzle impenetrable to someone trying to learn the ropes. (This is why, for instance, we don't use new as a figure of speech to clue the letter N — that's a well-worn cryptic convention, but if a green solver asked us by what logic new was N, we couldn't tell them. Defenders might say "Well, New York and New Mexico …" but if we started defining stimulated as S or breathing as B because of laser and scuba, future Tap the Sign posts would be coming to you via our decapitated heads on pikes outside the city walls.)
When can you use nouns as indicators? As shown above, get a preposition in there before the fodder, or a possessive after the fodder. But if that wasn't your plan from jump street, that change is guaranteed to wrinkle your surface.
The same holds for anagrams. You may think that tagging disaster or accident directly before or after some anagram fodder is a perfectly reasonable thing to do, but the implied "of" and inverted order make those constructions untenable. Things that looks like nouns but are actually adjectives or adverbs, like novel or cast, are of course fine and can cozy right up to the fodder without preposition, possessive, or other intervention. Now, it's not common English to have post-nounal adjectives — for instance, in One of Dr. Wily's robots in Mustang convertible (4,3) for GUTS MAN, we're saying [FODDER] convertible instead of convertible [FODDER] — but there's plenty of English precedent for such constructions: heir apparent, proof positive, attorney general. It's not a Cryptics 101 construction, but it's not hand-wavey.
The following clues use some noun indicators that we found in the wild, deployed in such a way that we were pulled out of the puzzles we were solving. We've used them to write clues that skirt sign-tapping by using other techniques to put the nouns into phrases that work.
Car's explosion in curve (3)
Choice of potion in error (6)
Realigment of spine for bird (5)
are straightforward clues for ARC, OPTION and SNIPE. But if we come across Bird's spine realignment, the fact that it doesn't quiiite work becomes the thing we see most in the clue.
If you clue mountain top for M, our sign-tapping would be nigh inaudible, and indeed, we would do this, have done this, and will do it again. (Mountaintop, too, but that's a different conversation.) Now, for a formulation like that, we will almost always break out a question mark. Classical cryptic rules would say the question mark should be as close as possible to the dodgy thing, but here in the third decade of the 21st century, we're happy to let an end-of-clue question mark signal some kind of chicanery anywhere in a clue. And mountain top works for two reasons: (a) It's a figure of speech that everyone knows that (b) literally refers to the top of mountain in English as she is spoke, and so that surface meaning carries over into the cryptic meaning. A satisfying cryptic clue has a cryptic parse that slices and dices the surface meaning, where the pleasure of the clue comes from skewering as many of your assumptions about the surface as possible … and this may be the only exception. If we were to clue big top for B, our question mark would be doing a lot of work, because while (a) is still true, that phrase does not literally refer to "the top of big" in E.a.s.i.s. Likewise, we wouldn't do lamp top for L because, while (b) is true in that lamps literally have tops, no one calls it that, and so (a) doesn't apply. (Thanks to Mossberg, George Ho, ryanf and kormad for an edifying conversation about this a few months ago.) Pop quiz to discuss in the comments: How about top dog for D?
Here's how those same (a) and (b) principles might apply in an anagram clue:
Milkshake held up by E.R. actor that played Morrison and Holliday (6)
for KILMER: (MILK)* + E.R. Not a sparkling clue by any stretch, but an example of the rare case of abutting noun to noun and having one reasonably serve as an anagram indicator.
Our next Tap the Sign will conclude our series on anagrams, talking about what makes a good indicator.
* We know.
** We know.