Skip to main content

​
Oracle Overview

A pragmatic, LLM‑assisted oracle that settles prediction markets with verifiable, time‑bounded evidence. It blends multiple AI providers with a strict settlement rubric, and commits outcomes on‑chain when decisive—or safely triggers refunds when ambiguity remains.
  • What it does: Decides YES/NO market outcomes or identifies non‑resolvable/ambiguous cases.
  • How it thinks: Evidence‑first prompts that check listed sources, extract specific numbers, and compute thresholds explicitly—minimizing “vibes” answers.
  • Where it runs: Backs market platforms with an API layer and a Solana settlement module.
  • Providers: Uses Perplexity and Grok 4 for real‑time, wide‑coverage retrieval. Their unbiased, up‑to‑date context makes this oracle perfectly suited for prediction markets.

​
Two‑Step Workflow

​
1) Criteria generation (filtering and clarity)

After market creation, the oracle generates or validates market‑specific settlement criteria before the market can be made tradeable :
  • Resolvability gate: Confirms whether the question is objectively answerable by the deadline.
  • Authoritative sources: Builds a concise list of official or governing‑body sources to check later.
  • Objective rubric: Clarifies exact metrics (scores, counts, prices, timestamps) and how to compute the outcome.
  • Token/coin markets: For coin markets, criteria incorporate token‑specific context and data pathways without exposing internal data methods.
Result: Markets flagged as non‑resolvable are safely filtered (and can be routed to refund paths). Resolvable markets move to step 2.

​
2) Settlement (decision and on‑chain finalization)

At/after market end:
  • Time‑bounded evidence: AI providers are restricted to information available up to the market’s deadline.
  • Source‑checking: The oracle checks the pre‑listed resolution sources first. It extracts the numbers/facts that resolve the question.
  • Explicit computation: Shows the math or threshold logic (e.g., totals, comparisons), then decides YES/NO—or returns ambiguous only if truly unverifiable.
  • Consensus mode (optional): Multiple providers can be queried; if they agree, we finalize with higher confidence. If they conflict, we prefer authoritative sources and fall back safely.
  • Finalization: When decisive, the on‑chain module commits the result; when ambiguous/non‑resolvable, a dispute path is enabled.

​
Why This Oracle Gets Markets Right

Purpose‑built for prediction markets where clarity, deadlines, and fairness matter most.
  • Evidence‑first prompting: Requires checking the listed sources, extracting exact values, and doing the calculation explicitly—not just summarizing news.
  • Time‑bounded lookups: Date filters prevent leakage from after‑deadline information.
  • Authoritative‑source precedence: Official databases and governing bodies outrank general reporting. Conflicts use a clear precedence policy.
  • Deterministic rubric: Criteria define what counts, how to count it, and how to compare it—before any decision is made.
  • Multi‑provider optionality: Uses multiple AI providers and can require agreement to reduce single‑model bias.

​
Operational Flow

  • Before end time
    • Run criteria generation for a batch of markets.
    • Mark resolvable vs. non‑resolvable and store resolution sources.
  • At/after end time
    • Execute settlement analysis with time‑bounded evidence.
    • If decisive, finalize on‑chain; otherwise, enable refunds.
  • Admin safety
    • Sensitive write actions are gated; emergency refund flows are supported.
    • Many test endpoints are disabled in production.

​
Non‑Technical Pitch

  • Fair and transparent: Decisions cite sources and show the arithmetic behind the outcome.
  • Deadline‑honoring: Uses information available before the market closes—no hindsight bias.
  • Robust and pragmatic: Human‑like reasoning that behaves like a careful adjudicator, not a chatbot.
  • Multi‑model resilience: Optional consensus reduces brittleness; disagreements trigger safe fallbacks.
  • On‑chain finality with safety nets: When clear, we settle; when unclear, refunds are enabled fast.

​
Oracle Roadmap

  • Ongoing upgrades to criteria templates, source‑precedence policies, and time‑bounded retrieval.
  • Expanding support for more market types: token/coin price markets, sports, politics, social/media events, on‑chain protocol events, and custom rubric‑defined questions.
  • Deeper integration with Perplexity and Grok 4, plus additional MCP connectors for domain‑specific evidence.
  • Path to decentralization: optional operator network reaching consensus prior to finalization.