Ritara Whitepaper
An AI Copilot for Prediction Markets
This document describes Ritara's product vision and intended technical direction. It is not a promise of future features, not financial advice, and not an offer to sell securities or provide investment services.
Abstract
Turning market pages into informed decisions
Prediction markets turn beliefs into prices, but participating responsibly still demands too much manual work: reading market rules, understanding settlement conditions, tracking the information environment, and managing risk across fragmented platforms. Ritara is building the intelligence layer that sits directly in the trader's workflow.
Ritara starts as a read-only Chrome extension that acts as an AI copilot inside prediction market pages, summarizing markets, surfacing key constraints, highlighting uncertainty, and helping users reason from evidence. We begin on Base, initially targeting Limitless.Exchange, a Base-based prediction market platform focused on short-term markets. [1] [2]
Ritara's roadmap extends beyond read-only assistance into cross-market coverage across popular prediction markets, and ultimately into personalized trade suggestions powered by an opt-in learning loop based on user behavior and outcomes.
1. Motivation
Why prediction markets need a copilot
Prediction markets are compelling because they compress distributed information into a single number: price. But the practical experience of trading them often breaks down into friction.
- Market comprehension friction: settlement conditions, resolution criteria, time windows, liquidity structure, fees, collateral, and edge cases.
- Research friction: evidence is scattered across sources; the why behind price movement is rarely embedded in the market interface.
- Cross-market fragmentation: the same or similar events can exist on multiple platforms with different rules, liquidity, and pricing.
- Decision friction: even when a user has an opinion, translating it into a risk-managed trade is non-trivial.
Ritara exists to reduce these frictions without requiring users to leave the page they are already on.
2. What Ritara Is Today
A read-only copilot delivered as a Chrome extension
Ritara is a read-only AI copilot delivered as a Chrome extension for prediction markets.
Current characteristics
- Read-only: Ritara does not place trades and does not require custody of user funds.
- In-context: Ritara runs in the browser and operates on the market page a user is viewing.
- Decision support, not certainty: Ritara's job is to clarify, not to claim omniscience.
3. Vision
The intelligence layer for prediction markets
Ritara becomes the intelligence layer for prediction markets - making markets easier to understand, safer to trade, and simpler to compare across platforms.
Workflow-native intelligence
Insight should appear where the decision is made: directly on the market page.
Transparency and uncertainty
Ritara should show why it believes something and how confident it is.
User control first
Ritara begins read-only, then moves toward suggestions, with opt-in personalization and explicit user agency.
Cross-market normalization
A market should be representable in a standard schema regardless of platform, enabling comparison, search, and consistent analysis.
4. Starting Scope: Base + Limitless.Exchange
Anchoring to Base and Limitless.Exchange
Why Base
Base is described by Coinbase as an Ethereum Layer 2 chain built on the OP Stack, aiming for low-cost, developer-friendly on-chain applications. [2]
Why Limitless.Exchange
Limitless.Exchange is described as a prediction market platform built on Base, focused on short-term price markets and fast settlement. [1]
Some ecosystem sources describe Limitless resolution as using Pyth Network oracles and USDC collateral. [3]
Oracle-aware reasoning
Oracles matter because they define how real-world data becomes on-chain resolution. Pyth describes itself as a first-party oracle network that provides real-time price feeds across many chains. [4]
Base documentation includes Pyth as an oracle option and describes Pyth's pull-oracle design and low-latency updates. [5]
Ritara's role is not to be the oracle - it is to help users understand what the oracle implies for settlement, timing, disputes, and risk.
5. Core Product: The Ritara Copilot
Decision support on the market page
What Ritara does
- Market summary: what is being asked, what counts as Yes or No, when it resolves, and what source resolves it.
- Settlement and edge-case checklist: identify ambiguous terms and highlight resolution dependencies (oracle feed, timestamp, market rules).
- Price interpretation: translate prices into implied probabilities, and highlight what price movement would mean.
- Evidence workspace: help the user gather and evaluate relevant evidence with explicit uncertainty.
- Risk framing: non-prescriptive guidance on position sizing concepts, slippage, fee awareness, and scenario thinking.
What Ritara does not do (yet)
- Trade execution
- Custody
- Guaranteed edge
- Hidden personalization or training
This restraint is intentional: trust comes before autonomy.
6. Technical Architecture (Conceptual)
A modular pipeline for scale
Ritara is designed as a modular system so that adding platforms and moving up the autonomy ladder does not require rebuilding everything.
High-level pipeline
Page Understanding (Client)
Detect platform and market page; extract structured info from the DOM (title, outcomes, price or odds, expiry, rules, liquidity signals).
Normalization (Core Layer)
Convert platform-specific data into a common Market Object schema.
Retrieval (Evidence Layer)
Pull relevant sources: platform rules, oracle details, related markets, user-provided links.
Reasoning (Model Layer)
Produce structured outputs: summary, key risks and unknowns, assumptions, and an evidence map.
Presentation (UI Layer)
Render insights inside the extension overlay.
Market Object schema (draft)
This schema is a key moat: once normalized, Ritara can compare across markets, build search, and evaluate models consistently.
7. Trust, Safety, and Privacy
Trust before autonomy
Read-only first
Ritara begins read-only because it materially reduces risk: no funds handling, lower blast radius for bugs, and easier user trust.
Data minimization and consent
- Default: do not collect sensitive content unnecessarily.
- Opt-in: any telemetry used for personalization or model improvement requires clear opt-in.
- User controls: export and delete controls should exist for user-associated data.
Why this matters
Browser extensions can be high-risk collection points if implemented irresponsibly. Reporting has highlighted cases where popular extensions harvested users' AI prompts and responses, underscoring why strict privacy defaults and transparent permissions are essential. [6]
Transparency as a feature
- What sources were used
- What assumptions were made
- What is uncertain
- What data (if any) is being stored or sent
8. Roadmap
A staged path to trust and autonomy
This roadmap is intentionally staged: credibility first, autonomy later.
Read-only Copilot
Chrome extension that sits on market pages and surfaces clarity without taking action.
- Support a small set of market pages.
- Market summaries and rule highlights.
- Evidence assistance and uncertainty-aware reasoning.
Platform Coverage on Base
Broaden coverage across Base and harden normalization.
- Broaden support across Base prediction market surfaces.
- Improve the Market Object normalization layer.
- Add robust diffing for what changed since last visit.
Search, Compare, and Monitor
Normalize across platforms and expose cross-market intelligence.
- Cross-platform search for related markets.
- Comparison of rules, liquidity, and prices.
- Watchlists, alerts, and portfolio visibility (read-only).
Trade Planning (Not Execution)
Introduce trade plans rather than commands.
- Suggested directions with rationales.
- Scenario analysis and change-your-mind prompts.
- Risk framing and sizing concepts with explicit disclaimers.
Preferences and Context
Use consented signals to tailor analysis and alerts.
- Topics of interest.
- Preferred risk posture.
- Explanation style and depth.
- Alerts and summaries.
Ritara-trained Suggestion Model
Train and evaluate models that suggest trades based on signals and outcomes.
- Market states (prices, liquidity, time-to-expiry).
- Evidence signals that are structured and cited.
- User constraints and preferences with explicit consent.
- Outcomes and feedback loops.
9. Model Training and Evaluation (Planned)
Learning with consent and measurement
Data sources (opt-in)
- Explicit feedback: thumbs up or down, this was helpful, this missed a key point.
- Behavioral signals: watchlist additions, alert creation, time spent reading analysis.
- Outcome grounding: resolved market results (when available).
- User-declared profile: risk tolerance, topics, time horizon.
Evaluation standards
- Forecasting metrics: calibration, Brier score, log loss (for probabilistic outputs).
- Decision usefulness: whether suggestions reduce user effort and increase clarity.
- Safety metrics: hallucination rate, citation accuracy, refusal correctness.
- A/B testing: only when privacy and consent standards are met.
10. Risks and Limitations
Clear-eyed limits
AI limitations
AI can hallucinate. Ritara must be engineered to ground outputs in sources and show uncertainty.
Market integrity
Prediction markets can be manipulated and can reflect narrative momentum rather than truth in the short term. Ritara should present alternative hypotheses and warn when information is low quality.
Regulatory and compliance considerations
Trade suggestion features may trigger additional compliance requirements depending on jurisdiction and implementation. Ritara prioritizes decision support and user agency, with conservative rollout of any features that could be interpreted as individualized financial advice.
11. Conclusion
Clarity before conviction
Ritara starts with a simple wedge: a read-only Chrome extension that makes prediction markets easier to understand directly in the user's workflow. From that foundation, Ritara expands into a cross-market intelligence layer and, only after earning trust, moves toward personalized trade suggestions built on opt-in learning and rigorous evaluation.
The goal is not to predict the future. The goal is to make prediction markets usable, interpretable, and safer so users can make better decisions with less friction.
Appendix A: Glossary
Shared terms
Prediction market
A market whose prices represent beliefs about future outcomes.
Market Object
Ritara's internal normalized representation of a market across platforms.
Oracle
Infrastructure that brings external data on-chain for settlement.
CLOB
Central limit order book, an order-driven liquidity model.
Calibration
Whether predicted probabilities match observed frequencies over time.
Sources