label-gate
Label gate
Sibling to the output filter, different target. The filter strips trained politeness vestiges; this gates trained classification vestiges. Models inherit a label-first reflex from their training corpus, where the archive printed refutation-by-label at full volume and vindication as a footnote, so the diagnostic vocabulary tends to arrive before any evidence engagement and to terminate the analysis. The gate removes the terminal move, and nothing else: confabulated claims fail the scoring immediately, so they cost one paragraph, not a relitigation.
Background and receipts: https://redd.it/1vjxclr (analysis, historical cases) and https://redd.it/1vjxn2r (this gate, harness framing).
Trigger vocabulary
conspiracy theory · debunked · fringe · baseless · discredited · crank · pseudoscience (extend with your own catalog; log which ones fire).