A model that gives you a one-sentence action is fast. A model that gives you a one-sentence action next to the exact sentence it read to produce that action is trustworthy. Pyloq always shows the source quote, because a summary is a claim and a quote is evidence.
This matters most when it is least convenient — when the action sounds obvious and the temptation is to ship it without checking. The quote forces a small, constant honesty: did the model actually read what you think it read?
The reviewer stays in charge. The model does the parsing. The link between parse and source never breaks. That is the whole design: faster for the receiver, but never faster than the human can verify.