phronesis · laserbrain
Logic Engine
Classification against rules you declare. It does not reason — it decides which of your categories a text falls into, and then accounts for every decision, including the ones it did not make.
That is the part worth having. “Why did you flag this” is a question anything can answer plausibly, including a model that is guessing. “Why did you not flag this” is what a review actually asks, and a model cannot answer it at all — there is no trace, and asking again may give a different answer. Click any rule below.
your rules · any / all / none, with weights
the text to classify
verdict
margin is a gap to the runner-up, not a probability. This engine has no basis for a confidence and will not invent one.
every rule considered — click one to ask why not
This is a lookup against a trace, not a generated explanation. The same rules and the same text give the same answer every time, and the reason is the actual cause rather than a plausible one.
reference case
No ground set. Everything above is a reading of this text alone — it cannot tell you whether this case has moved, because it has nothing to compare against.
Freeze this as ground and every later reading is measured from it. It is never revised on its own: an instrument that quietly updates its reference follows the thing it is measuring and reports that nothing moved. Stored in this browser only — it is never sent anywhere.
hosted ruleset
Everything above runs in this page and always will. Publishing puts this ruleset somewhere a team shares one version of it, records each decision with the version that made it, and lets a past decision be re-run and checked.
Decisions store a hash of each input, never the input. A classifier pointed at contracts and complaints should not become a copy of them.
What it is not: it has no semantics. It cannot tell “I love this” from “I do not love this” unless you write a none term for the negation. It is a better-behaved regex with an audit trail — which is a defensible thing to hand a regulated buyer, and a dishonest thing to call reasoning.