See who moves the market — tick by tick, wallet by wallet
EventFlow Data captures full L2 order books, a wallet-attributed trade tape, on-chain settlement, and market-moving news — and turns them into research-grade analytics: toxicity and skill scores, markout and absorption metrics, and latency-honest event signals.
Archived, and the definition below is superseded — see the notice at the top of this page. Adverse selection is not profit.
Every wallet on the tape, scored
Every taker trade in the pipeline carries wallet attribution. That lets EventFlow characterize the participants behind price moves — statistically, transparently, and from public data alone.
Toxicity score
Counterparty-conditional signed markout at 10-second, 60-second, and 10-minute horizons, taken as a rolling 30-day median — complemented by a market-wide price-drift score computed over every trade on the tape at the same horizons. It answers one question: when this wallet trades, does the price move against whoever took the other side?
Skill score
A Gómez-Cram-style sign-randomization test: each of a wallet's realized bets has its direction re-flipped by a fair coin across 10,000 Monte-Carlo randomizations, building a luck-only baseline. The output is a skill percentile with a z-score effect size — separating genuine foresight from a lucky streak.
Cash-flow PnL engine
Realized PnL for any public wallet, reconstructed purely from raw cash flows — buys, sells, splits, merges, redemptions, and rewards — using weighted-average cost basis, the same avgPrice convention the venue itself reports. Conservation-identity reconciliation gates flag any wallet whose flows don't add up before a number is shown.
Behavior flags & activity stats
Pattern classifiers over public activity: fresh-whale (new wallet, large concentrated bets), bot (high-frequency small clips), and farmer (one-sided deep-in-the-money accumulation) — plus category-concentration (HHI) and hour-of-day entropy statistics.
Wallet analytics are statistical descriptions of publicly observable activity. They are research metrics — not an endorsement of any wallet, and not a suggestion to follow or copy anyone's positions.
The order book, measured — not just displayed
Every trade is replayed against event-driven L2 history, producing per-trade and per-market metrics that quantify liquidity, informed flow, and hidden-size behavior.
Markout & realized spread
Per-trade markout in basis points at multiple horizons, computed against the best bid/ask prevailing at trade time, with realized-spread metrics on captured quote fills — the standard toolkit for measuring adverse selection and effective liquidity.
Absorption & iceberg candidates
Replays the trade tape against order-book history to flag same-price replenishment and price pinning — places where displayed size keeps refilling as it is hit. Output is always a candidate with the underlying evidence attached, never a claim of proof.
Depth & flow features
Analysis-ready features from the live book and tape: depth imbalance, microprice, cumulative volume delta, and large-trade detection — the structure behind the price, precomputed for research.
Fill-intensity estimation
A descriptive model of historical fill rates: expected fill intensity (λ) bucketed by distance from the touch and prevailing spread state. A measurement of liquidity — not an execution tool.
Headlines, read against the market
Multi-source news ingestion meets the order book: LLM pipelines classify what matters, read settlement clauses, and score every signal against the price that was actually available when the signal could have been delivered.
Multi-source ingestion & classification
RSS, Telegram, and X connectors — RSS polled with conditional-GET caching — feed an LLM classification cascade that scores each item for materiality, relevance, and confidence against specific markets.
Settlement-clause deep read
For flagged events, an LLM reads the market's actual resolution text alongside its current probability and sibling markets, and emits a fair-probability estimate with a written thesis — a model output for research, not advice.
Latency-honest evaluation
Every signal is evaluated against the market state at a realistic delivery time — never against prices a reader could not have seen. Wallet-flow "wave position" analysis shows whether informed money was already positioned before the headline broke.
Full LLM audit trail
Every model call is logged with model, token counts, cost, and latency — so any signal's provenance, and what it cost to produce, can be inspected end to end.
“Resolves YES if the target rate is cut at or before the September FOMC meeting.”
Capture built for replay — analytics built on capture
Every metric above is computed from data the pipeline captures itself: raw messages retained verbatim next to structured records, sequenced for deterministic replay.
L2 order books
Raw messages verbatim plus structured snapshots — bids/asks, best bid/ask, spread, midpoint — sequenced for replay.
Attributed trade tape
Taker wallet, side, price, size, and transaction hash — deduplicated on capture — alongside a live trade feed enriched with trade-time BBO and inferred aggressor side.
Market & event catalog
Titles, categories, status, close times, liquidity, 24h and all-time volume, tick size, and neg-risk flags.
On-chain settlement
Polygon resolution and settlement events, transfers, and payout vectors — the ground truth PnL reconciles against.
Reference & context
Deribit option chains and implied-vol surfaces for crypto-linked markets; RSS, Telegram, and X news feeds.
Research-grade history
Prices-history, 1-minute candles, and 24-hour sparklines, retained alongside the raw streams.
A working research console over the full dataset
A web workbench built on the same pipeline: a live market board with sparklines, token detail with candlestick charts, an order-book ladder, depth chart and live trades, wallet-intelligence leaderboards with drilldowns, watchlists, read-only views of public-wallet positions, and a docs portal — in light or dark theme.
Interface previews are illustrative; figures shown are sample data, not live product metrics.
One dataset, four ways in
Explore visually, query programmatically, or subscribe to live updates — the same standardized, quality-flagged data behind every surface.
Web workbench
Markets board, token detail with candles and depth, wallet-intelligence leaderboards and drilldowns, watchlists, and a docs portal — in light or dark theme.
REST API
Read endpoints for tokens, order books, prices-history, market search, and public-wallet trade and position views. Standard JSON, built for research scripts and internal services.
WebSocket push
A push channel for live dashboards and monitoring — subscribe to market summaries, price changes, and depth updates as they happen.
Docs portal
Integrated documentation covering endpoints, field definitions, and the methodology behind every published metric.
All interfaces are read-only over analytics and public data. There are no order, execution, or account-operation endpoints. Market data and wallet intelligence are live in the workbench and API today; microstructure and news-signal analytics are delivered as research outputs.
Every record explains its source and status
Freshness, gap logs, latency metrics, log trails, and audit records make data trustworthiness explicit — helping teams separate real market moves from data-quality issues.
Freshness, latency, and gap logs
Freshness and latency metrics plus explicit gap logs make it clear when data is complete — and when it is not.
Raw + structured dual retention
Raw messages are retained verbatim alongside structured snapshots, so any derived number can be traced to its source.
Sequenced capture, deterministic replay
Per-worker sequence numbers on captured streams make order-book history replayable, exactly as received.
Health checks & observability
Health endpoints, Prometheus metrics, Grafana dashboards, and audit trails with correlation IDs cover the pipeline end to end.
Read-only by design
The platform analyzes publicly observable data and exposes it through read-only surfaces. This site collects only essential contact, access, and cookie-preference information — never private keys, seed phrases, account or asset credentials, bank cards, or sensitive identity materials.
Minimal data collection
Only the minimum needed for contact, access, and preferences.
Explicit cookie consent
Analytics cookies run only after consent and can be withdrawn anytime.
HTTPS and security headers
Encrypted transport site-wide with security response headers.
Contact-form anti-spam
Form submissions include anti-abuse and duplicate-submit protection.
Configurable log retention
Security logs are retained by policy, with configurable periods.
No trading access
No exchange accounts, no keys, no funds, no transaction capability — the product cannot transact on any venue.
See it on live public data, then decide
Book a 30-minute walkthrough: wallet toxicity and skill scoring, markout and absorption analytics, news-signal evaluation, and the capture pipeline underneath — then choose how to integrate.
This form is part of an archived page. It is disabled and sends nothing. To reach us, use support.
Frequently asked questions
An analytics platform for public prediction-market data. It captures L2 order books, a wallet-attributed trade tape, on-chain settlement, and news feeds, and turns them into research analytics — wallet toxicity and skill scores, cash-flow PnL, markout and absorption metrics, and LLM-read event signals — delivered through a web workbench, REST API, and WebSocket.
No. EventFlow Data has no trading functionality of any kind: it cannot place, route, or execute orders; it offers no market making or automation; it does not connect to or operate exchange accounts; and it does not custody funds or keys. It is a read-only analytics product over publicly observable data.
A measure of adverse selection. For each wallet we compute counterparty-conditional signed markout — how far the price moved against the wallet's counterparties — at 10-second, 60-second, and 10-minute horizons, then take a rolling 30-day median. A high score means the market tends to move against whoever trades opposite that wallet. Where a wallet has no fills against captured quotes, a market-wide signed price-drift score over all of its tape trades provides the same lens.
PnL tells you what a wallet made; the skill score tells you whether that result is distinguishable from luck. We run a sign-randomization test in the style of Gómez-Cram: each realized bet's direction is re-flipped by a fair coin across 10,000 Monte-Carlo randomizations to build a luck-only distribution, and the wallet's real result is placed on it as a percentile with a z-score effect size.
No — and it is deliberately labeled that way. The system replays trades against L2 order-book history and flags same-price replenishment and pinning patterns as candidates, each with the underlying evidence attached. Hidden size cannot be proven from public data; we surface evidence, not conclusions.
No. They are model-generated research outputs. Each item is scored for materiality, relevance, and confidence; every fair-probability estimate ships with its written reasoning and is evaluated against the market state at a realistic delivery time. Nothing on the platform is a recommendation to buy or sell anything.
From publicly accessible sources: venue market-data APIs and WebSocket feeds, public Polygon on-chain records, public news and social feeds (RSS, Telegram, X), and Deribit reference data for crypto-linked markets. The Data Sources page in the footer lists sources, update behavior, and limitations.
Yes. The pipeline retains raw messages verbatim alongside structured snapshots, plus prices-history and 1-minute candles. Sequence numbers on captured streams make order-book history replayable for research.
The landing page does not collect private keys, seed phrases, account or asset credentials, bank cards, or sensitive identity materials.
Necessary cookies keep the site running; analytics cookies are enabled only after consent and can be changed anytime in the footer's Cookie Settings.