modes.md

Three ways to answer, and the graph as RDF

Modes

curl -sX POST localhost:5100/jobs -H 'content-type: application/json' \
     -d '{"query":"...","mode":"settle"}'

mode	what it does
reason	one run — steps, a graph, an answer
explore	the question is split into questions, each answered by a run of its own, then assembled
settle	explored, explored again, and finished only when two independent runs agree

The same picker sits on the web page.

settle exists so that no single call decides. When the two runs disagree, three checks vote through different lenses — same value, same conclusion, same action — and the tally is shown rather than reduced to a verdict, because 2-1 and 3-0 are different things.

A vote is still opinion, so both answers are then probed. The same sum with its numbers moved is graded against the exact evaluator and settles the matter outright; asking what a knower would also know is suggestive only, and labelled as such, because a model can be confidently and consistently wrong.

explore splits with tapering breadth — 5 questions at the top, then 4, 3, 2 — so a deep exploration is large but still ends. Two embedding guards keep it honest, and neither is the model's to apply: anything that has drifted from the parent question is dropped (the model cannot tell — each step looks reasonable from the one before), and anything already asked is dropped (it cannot tell — it does not hold the earlier questions).

Headless jobs

ID=$(curl -sX POST localhost:5100/jobs -H 'content-type: application/json' \
       -d '{"query":"How many seconds are there in 23 weeks?"}' | jq -r .id)

curl -s localhost:5100/jobs/$ID              # {"state":"running","steps":3,...}
curl -s localhost:5100/jobs/$ID/stream       # N-Triples, live, one triple per line
curl -s localhost:5100/jobs/$ID/rdf          # Turtle, once it has finished
curl -sX DELETE localhost:5100/jobs/$ID      # stop it

The RDF

Two formats for two purposes. N-Triples streams: each line is a complete document, so a consumer can parse what has arrived without waiting for the end. Turtle is prefixed and readable, and needs the whole document, so it is what a finished run serialises to.

The lkg:basis predicate separates kinds of claim, so a consumer can take only the part it can rely on:

basis	meaning
lkg:Embedding	measured similarity — an association, no truth value
lkg:Exact	settled by the exact evaluator — reproducible
lkg:Opinion	a vote — three model readings, counted

<run/a1b2/link/Step1-Step2>  lkg:similarity "0.8371"^^xsd:decimal ;
                             lkg:basis      lkg:Embedding .

<run/a1b2/conversion/1>      lkg:statement  "23 week = 13910400 second" ;
                             lkg:from       <unit/week> ;
                             lkg:to         <unit/second> ;
                             lkg:basis      lkg:Exact .

An exploration also carries its structure: lkg:Finding (each sub-question and what it came back with), lkg:Vote (the tally and what the dispute was about), and lkg:Agreement (how the run finished — by wording or by vote, and in which round). A plain reason run gains none of these, and both formats always carry exactly the same triples.