Trust

Trust Center

Everything a serious reader needs to decide whether to trust Orrery — what we are, what we aren't, where the data comes from, how the signals are computed, what every resolution-source type means, and how to reach a human.

The short version

no advice · no execution
  • Public data only. Polymarket Gamma, CLOB, and Data APIs. No private feeds, no insider information, no privileged sources.
  • No brokerage, no custody. Orrery never holds funds, never signs transactions, never asks for a private key. Watchlist / Alerts / Portfolio are all read-only.
  • Not financial advice. Every signal carries Evidence / Backtest / Action labels — and Action is always a verification step, never a trade recommendation.
  • Honest about misses. Backtests show weak / mixed verdicts up-front. Signals that didn't pay off get the same surface as the ones that did.
  • Minimal tracking. Plausible (privacy-first analytics) and Google Analytics 4. No third-party ad networks, no wallet linking without your action.

Methodology

reviewable rules · no ML

Orrery's signals come from a small set of deterministic rules over public Polymarket data — no machine learning, no proprietary scoring. Every rule input is reviewable: trade feeds, price changes, liquidity, spread, expiry windows.

Signal kinds

  • Flow. Clusters of aligned trades on the same side of a market within a short window. Forward-only — the public CLOB price feed can't replay event-time flow history.
  • Momentum. 24-hour probability move ≥ a configurable threshold, weighted by liquidity. Backtested on 1-month CLOB price history; see the live verdict on /backtest.
  • Divergence. Short-window reversal pattern: positive 1h move after a negative 6h move (or vice-versa), filtered to liquid markets with at least one cluster of trades. Also backtested.
  • Resolution risk. Combines proximity to expiry, UMA dispute state, and extracted resolution-source type. Forward-only — event- time risks can't be replayed from price history.
  • News lag. Markets with thin trade history that haven't re-priced after a likely external trigger. News ingestion is pending so the kind currently fires conservatively.

Calibration tiers

  • Evidence. Live confidence in the rule firing — clustered trades, aligned price move, tight spread. Score 0..100, mapped to Low / Medium / High.
  • Backtest. Per-kind historical performance: Strong / Mixed / Weak / Forward- only. Today: momentum=Mixed (positive win rate, thin average move), divergence=Mixed (low sample), flow / resolution_risk / news_lag=Forward-only.
  • Sample. How many markets and firings the backtest claim rests on. High = ≥ 10 markets and ≥ 50 firings; Medium = ≥ 3 markets and ≥ 25 firings; Low otherwise. Forward-only kinds are explicitly labelled — no implied edge.
  • Action. Always a verification step. Watch only / Inspect timeline / Verify source / Create alert / Compare. Never a buy or sell call.

Edge Score

The composite ranking on /opportunities. Strongest live signal evidence (up to +45) · multi-kind bonus when several kinds agree (+6 each, capped at +12) · log-scaled liquidity (up to +15) · tight-spread factor (up to +10) · resolution-risk penalty (−10). Always presented with a Confidence reading next to it so the user knows whether the score itself is believable.

Resolution-source types

14 types

Every Polymarket market resolves against an external source — an exchange print, a government release, a sports federation site, a tracker. Orrery extracts that source from the market description and classifies it into one of these canonical types so the resolution-risk surface is consistent across markets.

  • Exchange priceexchange_price

    Resolution reads a price from a centralized crypto exchange (Binance, Coinbase, Kraken, etc.) at a specific instant.

    e.g. Binance BTC/USDT 1d close on April 30 ≥ $100,000.

  • Commodity price feed / futures datacommodity_price_feed

    Resolution reads a futures or commodity price from a public feed (Pyth, CME settlement) — these are programmatic and verifiable.

    e.g. WTI Crude futures price published by Pyth on April 30.

  • Official government sourceofficial_government

    Resolution depends on a government release: an agency report, executive order, election certification, or central-bank decision.

    e.g. BLS jobs report headline for May 2026 above 200k.

  • Official statisticsofficial_statistics

    Resolution reads a number from an official statistical agency — BLS, BEA, Eurostat, IMF, etc.

    e.g. U.S. CPI YoY for April 2026 above 3.2%.

  • Official sports resultsports_official_result

    Resolution depends on the official result of a sports match or tournament — federation site, league, or governing body.

    e.g. Serie A standings on the official Lega Serie A website.

  • Court recordcourt_record

    Resolution depends on a court filing, ruling, sentencing, or other docket-visible event.

    e.g. Filing visible on PACER for case XYZ on or before June 1.

  • Company filingcompany_filing

    Resolution depends on an SEC filing, earnings release, S-1, 8-K, or similar corporate disclosure.

    e.g. Company files an 8-K with the SEC by June 30.

  • News consensusnews_consensus

    Resolution depends on consensus reporting across credible newsrooms — typically used for events that can be confirmed but don't have a single oracle source.

    e.g. Three or more outlets report the deal closed before July.

  • Social media postsocial_media_post

    Resolution depends on a specific social-media post or set of posts (typically X / Twitter) being publicly visible.

    e.g. Elon tweets between 220 and 239 times in a 24h window.

  • Public-figure statementpublic_figure_statement

    Resolution depends on a named individual making a specific statement — often via interview, press conference, or official channel.

    e.g. Public figure publicly endorses candidate X by Election Day.

  • Tracker / countertracker_or_counter

    Resolution reads a count or running total from a public tracker — flight tracker, satellite ship counter, polling aggregator.

    e.g. Number of cargo ships passing Strait of Hormuz tracked by xtracker.

  • Leaderboard / benchmarkleaderboard_or_benchmark

    Resolution reads a position on a public leaderboard / benchmark site — model rankings, GitHub stars, etc.

    e.g. Hugging Face leaderboard top-1 by end of quarter.

  • Ambiguous wordingambiguous

    The resolution rule's wording is ambiguous — Orrery could not extract a single deterministic source. Treat with extra care.

    e.g. Resolves based on 'general consensus' without naming a source.

  • Source not classifiedunknown

    No resolution source could be classified from the market description. Open the market and read the rule directly.

    e.g.

Where the data comes from

  • Polymarket Gamma API. Markets, events, categories, end dates, descriptions, resolution rules. Cached server-side for 60 seconds.
  • Polymarket CLOB API. Order book, 1-month minute price history, spreads. Used for backtests and the timeline charts on market pages.
  • Polymarket Data API. Trade feed, wallet positions, holders. Powers the whale feed, smart-money flow on /brief, and wallet alert evaluation.
  • Pyth, Binance, Coinbase, atptour.com, etc. Referenced as resolution sources extracted from market rules — Orrery surfaces and links to them but never aggregates these feeds directly.
  • Anthropic Claude. Powers the “Why did it move?” summary on market pages and the AI Copilot. Prompts cite live data; Orrery never fabricates sources or numbers, and the AI block always carries an “explains, never forecasts” disclaimer.

Every page footer shows when its data was last fetched. No opaque scoring — every metric is reviewable.

Trust Center | Orrery