Five Pro surfaces, one terminal.
Each surface is a transformation. Scattered prices into ranked spreads. Raw confidence into calibrated edge. Quote-side P&L into spread, rebate, inventory, and adverse selection. Real-world exposure into a sized hedge. Individual trades into measured skill.
The trade is the gap net of friction.
Same event, different venue, different price. Polymarket and Kalshi list the same FOMC question, the same election, the same weather contract. Prices diverge for reasons that have little to do with belief: fee schedules, oracle wording, settlement currency, capital cost. The headline spread is not the trade.
nijinn matches contracts by underlying event, not by title. Oracle-gated entity resolution catches the cases where two contracts read the same but resolve differently, which is where naive matchers blow up. The unified price across matched contracts is liquidity-weighted. The actionable gap is the dispersion that survives ε, the friction sum: taker fees on both legs, slippage at your fill size, settlement-currency basis, the cost of being short a binary down to resolution.
Five arb types classified on entry: pure cross-venue, oracle-asymmetry, settlement-basis, resolution-window, liquidity-imbalance. Each type has its own ε profile and its own decay curve. You see the type, the net annualised return, the depth at fill size, and the time-to-resolution before you commit capital.
Log the prior before you look at the price.
The discipline is older than prediction markets. The instrument layer is what makes it stick. Without the log there is no calibration history. Without the calibration history there is no answer to 'am I actually getting better.'
nijinn scores your edge against consensus, sizes the position with Quarter-Kelly, and decomposes the trade into 6 sources of mispricing on entry: stale information, narrative anchor, oracle ambiguity, liquidity-induced drift, recency overreaction, and consensus collapse. Each source has a known half-life. The interface tells you which one you think you are exploiting before the trade goes on.
The post-trade view is the audit. Calibration is bucketed by your stated confidence: 0.55-0.60, 0.60-0.65, and so on up to 0.95+. Drift in any bucket shows up as a number, not a feeling. If you are systematically overconfident at 0.85 and well-calibrated at 0.65, the platform tells you in 30 trades, not 300.
Four fair-value reads. P&L decomposed every fill.
Quoting on a binary needs 4 fair-value reads at minimum: mid, depth-weighted mid, microprice, trade-flow VWAP. The 4 disagree the most exactly when you most need them to, which is when the book is thin or when one side is loading a directional view.
nijinn computes the 4 in real time and surfaces the one currently most informative. P&L is decomposed every fill: how much from spread capture, how much from venue rebate, how much from carry on inventory, how much absorbed as adverse selection. The four lines tell you whether your edge is widening or whether the rebate is the only thing keeping the book green.
Toxicity alerts fire when the flow turns informed: order imbalance shifting in one direction and your inventory accumulating on the wrong side at the same time. The alert lets you widen, pull, or hedge before the next fill becomes the one you regret.
Am I good at this. Can I get out at fair value.
The audit stack equity desks take for granted does not exist for prediction markets. Brier scoring is the table-stakes diagnostic and most practitioners have never plotted their own calibration curve. The calibration curve answers 'am I actually accurate.' MaxDD with Triple-Penance recovery answers 'how long does the next bad streak take to claw back.' Amihud-rolling answers 'can I get out at fair value before the count.'
nijinn ships the full stack. Brier and Brier Skill Score versus a uniform reference. Calibration curves bucketed by stated confidence with 95% confidence intervals. Drawdown decomposition with Triple-Penance recovery applied to the path. VaR at 95 and 99, Sharpe and Sortino over rolling windows, Amihud-rolling exit cost computed from venue depth. The exit-cost number is the one most practitioners do not have and most need.
Two questions answered. Both with numbers.
Right size is the deliverable.
Prediction markets are the only liquid instrument for many event-driven exposures: tariff outcomes, monetary-policy paths the rates curve does not cleanly span, weather events outside the WX-index range, geopolitical resolution dates. The hedge problem is not whether to use the contract. The hedge problem is what notional gets you the actual offset.
nijinn estimates the hedge ratio empirically: rolling correlation between the prediction-contract price path and the underlying P&L driver, log-odds transformed to handle the binary boundary effect, validated across multiple windows before it reinforces an edge in the influence graph. The output is a notional, not a ratio. The notional already accounts for the contract's payoff structure (binary versus continuous-outcome), the resolution window, the ε to enter at size, and the basis risk between settlement currency and your reporting currency.
Wrong-sized hedges are common and cost real money. The platform answers in dollars, not in coefficients.