How Keplar works

The pipeline behind every answer: how a question is understood, which kinds of questions bring in more models, how routing, agreement, disagreement, verification and synthesis work, and where the limits are.

The pipeline, end to endThe eight steps between your question and one answer: understand, select, compare, disagree, verify, synthesize, and answer, with where each can stop early.
How Keplar understands a questionThe classifier that decides task type, complexity and needed capabilities, what signals it uses, and why it is a transparent rule set rather than another model.
Which questions use more modelsA practical guide to what makes Keplar consult one model, three, or five: question type, length, stakes, forecasts, attachments, Deep Research and thoroughness, with examples.
Routing and panelsHow the router picks models: quality, reliability, speed and cost, family diversity, an open-weight seat, vision requirements and your plan.
Model families and diversityWhy a panel of models from different families is more useful than the same model asked three times, and what an open-weight seat is for.
Agreement and the consensus levelHow Keplar groups responses into positions, weights them by support, and turns that into a consensus level, and what the number does and does not mean.
Disagreement detectionHow Keplar finds where models split, how the Disagreements section is built, why it appears only when the split is real, and how to use it.
The verification reviewWhat the verifier or judge actually does: it reads the draft against each model's response and flags reasoning gaps, contradictions and unsupported claims. It does not browse or open sources.
How the final answer is writtenThe draft, review and final synthesis steps: weighing evidence over headcount, keeping minority views honest, and what happens if synthesis fails.
Missing models, timeouts and stand-insWhat happens when a model is slow, refuses or fails: why it is never counted as disagreement, how stand-ins work on Free, and how to retry with all models.
Exact checks for counting and arithmeticWhy counting letters and doing sums are risky for language models, and how Keplar computes them with code and passes the result to the models as a verified fact.
Sources, citations and web lookupsWhat the Sources section really contains, why Keplar does not open or check the links, the status of web lookups, and a checklist for verifying what matters.
Small talk and simple questionsWhy "hi" and "thanks" cost nothing, how Keplar detects casual chat, and what happens to short factual questions.
Why consensus can be wrongThe limits of agreement between models: shared training data, shared blind spots, stale knowledge and leading questions, and what to do about each.
A worked example, step by stepAn illustrative walk through one decision question to show what each stage contributes. The numbers are examples, not a recorded run.