The Love Equations Lab

Matching is not proof. Word overlap can find a relevant concept on this site; it cannot establish that either side is true.

Coverage describes your text only. A high matched share means your text talks about things this site covers—not that it represents anyone else.

Unmatched is useful. It may be noise, a source error, a gap on this site, or a lead worth chasing. The Open Questions view keeps those possibilities separate.

Irony is read straight. The stance label reads negation, quotation, attribution, and qualification from the sentence itself. It has no way to detect sarcasm, so a passage that mocks an idea by stating it is recorded as stating it. Where a stance label looks backwards, check the tone first.

Word shapes are approximated too. A short list of technical words stops a term like provider being read in its vendor sense, and that list is compared against the passage by stripping suffixes rather than by knowing English. It reaches too far in places and not far enough in others: paying is treated as payment, while carers is not treated as care. Where a plainly relational sentence is rejected, or a technical one is accepted, the word's shape is the first thing to check.

Clauses are guessed from punctuation. Deciding who is claiming what, and which words a borrowed term is standing next to, both need to know where one clause ends and the next begins. That boundary is approximated from commas, semicolons and dashes rather than parsed, so coordination without a comma, a negator inside a subordinate clause, an appositive, and a chain of attributions can all put a word in the wrong place. Flagged cases route to the feedback pipeline, where they accumulate as the argument for a real parser.