Prediction markets tell you what the crowd believes.
Contrary tells you where the crowd might be wrong.
A market price is one number with no explanation. Contrary produces a second, independent forecast and shows exactly which evidence moves it away from consensus, and how reliable that evidence is.
The pipeline
Gather the evidence
Every question pulls in published sources, each tagged with a publication, date, relevance score, and whether it supports YES or NO.
Five agents forecast independently
Each specialist answers one narrow question and produces its own probability, confidence, and evidence, without seeing the others first.
The Skeptic audits everything
Duplicated reporting, circular sourcing, unverified rumours, and recency bias are identified and down-weighted before aggregation.
The aggregator combines them
Forecasts are weighted by confidence and by how independent their evidence is, producing one probability and a disagreement score against consensus.
Your lens, remembered
You pick what you follow and which agent to trust more, or stay on General. Backboard stores that as memory on your assistant, so the next live run already knows who is reading.
The five agents
Base Rate Agent
Historical frequency
“How often do events like this actually occur?”
Returns a probability, a confidence level, key evidence, and a short reasoning summary.
News Agent
Recent developments
“What changed recently that genuinely moves the forecast?”
Returns a probability, a confidence level, important developments, and a short reasoning summary.
Contrarian Agent
Consensus stress test
“What is the crowd assuming that may not hold?”
Returns a probability, a confidence level, overlooked assumptions, and a short reasoning summary.
Market Analyst Agent
Consensus pricing
“What does the market price imply, and how is it moving?”
Returns a probability, a confidence level, market movement, and a short reasoning summary.
Skeptic Agent
Evidence audit
“Which of this evidence does not survive scrutiny?”
Returns a probability, a confidence level, evidence challenged, and a short reasoning summary.
Forecast Aggregator
Combines the five specialists into a single probability, weighting by confidence and evidence independence, and reports the disagreement score against market consensus.
keyPositiveSignals, keyNegativeSignals,
uncertainties, forecastExplanation }
Source quality is explicit
Every source is labelled Primary, Reliable secondary, Unverified, or Duplicate. The Skeptic Agent reduces the influence of the weak ones rather than silently averaging them in.
No hindsight in the Time Machine
Rewinding a forecast shows only the evidence published on or before the selected date, so a past position can be judged on what was actually knowable.
Scored, not asserted
Every agent carries a public Brier score across resolved questions, including the market consensus baseline it is competing with.
Backboard is the memory, not a chat box
Five agents run as separate Backboard calls. Your lens is saved on an assistant with memory turned on, then read back with memory Readonly so a live run knows what you follow without asking again.