Suggested by PeeWee (Telegram, 2026-07-30). get_btc_pool_state and get_btc_pool_history expose the raw prediction tape, so every AI agent re-derives the same statistics — wasting tokens and producing inconsistent numbers because each client invents its own definition of consensus or spread. Add a third read-only MCP tool, get_prediction_statistics, returning: consensus price (stake-weighted median) and its signed distance from the live index, mean/median (weighted and unweighted), standard deviation, dispersion as a percent of price, a 0-1 crowd confidence/consensus score, bullish-vs-bearish ratio by entry count and by stake, quartiles, the largest prediction cluster (center, stake share, signed distance from index), and the historical percentile of current dispersion (wider/typical/narrower). Lets agents answer 'is the crowd unusually bullish?', 'how concentrated are predictions?', and 'is today's distribution wider than average?' without reconstructing the tape, while raw data stays available for advanced users. Fully read-only, no auth. Covers btc-4h, btc-5m and event-5m.