15. One capability per question the lexicon is asked

Context

Thirteen catalogued rows declared lexicon.nouns. They wanted five different things: word membership (charade, semordnilap, word_square), an ordered noun list (n_plus_7, s_plus_7), synonyms (kangaroo_word, synonymic_substitution, chimera), antonyms (antonymic_substitution, antonymic_translation) and glosses (the definitional_* family).

A row asking for a thesaurus while declaring it needs a noun list is the same class of error as a catalogue definition promising more than its checker verifies.

Decision

lexicon.words and lexicon.nouns are separate capabilities with a pack behind them. lexicon.synonyms, lexicon.antonyms and lexicon.glosses are named on the rows that need them and have no pack behind them, which is the honest state.

Naming a capability before there is a checker that can honestly use it would repeat the mistake being fixed here, so those three stay unimplemented until a procedure can answer them without guessing.

Consequences

denckring status is unchanged in shape but truer in content: the eight remaining rows now say what they are actually waiting for.

The membership oracle is deliberately broad, and broad in a way callers inherit. It is the union of WordNet's noun lemmas and CMUdict's headwords, so it answers "could this be a word" rather than "is this in a dictionary of standard English" — CMUdict lists tac, which means cat reverses into something this oracle calls a word. semordnilap rests on that and says so.

The noun list is restricted to purely alphabetic lemmas. N+7 walks it, and a displacement has to be a word the tokeniser gives back whole: cat's-paw comes back as three tokens and would break the correspondence between source and result.