Research, replay and access, explained.

EdgeDepth is a market research platform for independent systematic crypto traders and small teams. Test a setup against recorded history, compare what followed with a baseline, inspect opposing examples and revisit the evidence.

Use the Research Workbench, your assistant through MCP, or the REST API. Live order flow, tick replay, events and lessons support the investigation. Research currently covers Binance USDT-M; live coverage includes Binance USDT-M and Hyperliquid. Replay depends on coverage and plan access.

Ask questions that can be expressed using the supported readings and operators: price moves, candle patterns, funding, open interest, liquidations and available order-flow measurements. Review the translated thresholds, markets and dates before running a study.

The Workbench supports editable conditions and ordered sequences within its grammar. You can also investigate a historical move. Setup studies measure what followed; outcome-first discovery is a separate retrospective workflow. Neither turns hindsight into a predictive test. See the query vocabulary.

Yes. The terminal is C++ compiled to WebAssembly and draws through WebGL2, so it runs in an ordinary browser tab: no install, no download, and the free tier needs no card.

The terminal covers what it does, and the terminal at a glance walks through the panels.

Free includes the live terminal, limited replay and research allowance, and free lessons. You can also read existing public reports without consuming research allowance.

Pro adds 90 days of rolling tick replay across supported recorded markets, expanded hosted features and a larger research allowance. Research adds deeper available replay history and the full research workflow, including REST API and MCP access.

Search history and tick replay are separate: a study can reach the full verified research record where its measurements exist. New scans are metered by scope and can cost more than one credit. Pricing lists current allowances, features and billing terms.

No. Reading an existing public report is unlimited and consumes no research allowance. Creating a study or materially changing one can require account-metered computation.

A scan can cost more than one unit depending on scope. Cache hits, unchanged ETag revalidations and cursor continuation do not incur a new scan charge. Check credits and limits and pricing for the current allowances and access.

Yes. The open-source terminal and gateway can run on public Binance market data without an EdgeDepth account. The terminal is AGPL-3.0 and the gateway is MIT licensed. Local recording and replay tools are available for your own data.

Self-hosting does not include the hosted research corpus, server-derived analytical layers or hosted services. See the open-source guide for the supported features and setup.

No. Every reading is named in plain words first and its technical id second, and the numbers are always counts: how many times something happened, out of how many chances it had. You can read a result without knowing how the reading behind it is computed.

When you do want the background, each idea has a short guide: order flow, VPIN, book imbalance, funding, open interest and liquidations. The rest are here.

No. An EdgeDepth alert reports that a supported saved condition occurred again. It returns you to the definition and research context; it does not prescribe an entry, exit or position size.

Monitoring is a web action that requires confirmation. The saved study retains its historical dates; the live evaluator applies its supported conditions to future closed market minutes. Monitoring does not automatically validate a strategy. How live alerts work.

A setup study reports whole-result counts and forward outcomes over its matching population. The visible page of example rows is not the denominator. Future outcome fields cannot be inserted into an ordinary setup predicate.

That does not eliminate selection bias. Choosing thresholds after seeing a move, repeatedly trying questions or selecting a favourable period can still distort a conclusion. Outcome-first discovery explicitly starts with a past outcome and is exploratory. Check a separate period without silently changing the definition. Read about selection bias.

No. These are descriptive event studies. A favourable historical rate does not establish net profitability, causality or robustness. Costs, fills, overlapping observations, repeated experimentation and changing market conditions still matter.

Inspect counterexamples and concentration, then consider a separate period or forward observation. An unsupported or inconclusive belief is a useful result too. Read a published study to see the evidence in context.

Missing required measurements can exclude an observation from the eligible population. Missing forward windows are reported separately from completed outcomes; they are not counted as zero returns or failures.

Coverage differs by market, date and feature. Earlier candle history does not imply a recorded order book, and exchange liquidation feeds can omit events. Modeled levels are estimates. Check the result exclusions and data coverage before interpreting a rate.

The result identity records the exact query, measurement versions and dataset revision. Keep those details with the result so you can tell what was tested and which record it used.

Reading a published report returns its pinned evidence. Running a query against newer data or revised definitions can produce a different identity and result; it is not a promise that a fresh scan will never change. Reproducibility details.

Conditions is a shelf of 26 curated conditions: named market states such as an extreme funding rate, a one-sided tape, or open interest accelerating into liquidations.

Each one has a page per market showing how often it happened there, what usually followed it with the count beside every rate, and links to inspect occurrences where replay coverage and access allow. It is a starting shelf, not a recommendation list: a condition that fires constantly and leads nowhere reads exactly like that, which is the point of counting it. Base rates explains how to read those numbers.

For the markets we publish, yes. Every market page shows what that market’s readings look like at the most recent minute in the record, which of the tracked states were true at that recorded minute, and what usually followed each of them, with the count beside every rate.

The chart underneath covers the last month: the candles, the moments that fired (with replay where recorded coverage and access allow), and the liquidation layers. Every reading also carries its percentile, so an ordinary value is easy to tell apart from an unusual one. Percentiles explains that chip.

One page per UTC day. The daily brief lists what fired across the tracked markets, and for each one what usually followed the previous times it fired on that market, with the count beside every rate and a link to inspect the historical moment, with replay subject to coverage and access.

It is assembled from the same counts as the rest of the site: no commentary, no opinions, nothing written by a model. Read it on the site, or leave your email and get it once a day. The typical-worst-drawdown figure beside each median is explained in MFE and MAE.

The same-scope unconditional reference shows what happened across eligible observations in the selected markets and dates, when available. It includes the setup matches. It is not a matched control group and does not isolate the effect of one condition.

Compare rates with their counts, completed windows and missing outcomes. Adjacent observations may overlap. Comparing two separately detected studies can change their anchors and populations, so a difference does not establish causality. How research works.

Yes. REST and MCP expose the same research engine as the browser. A compatible assistant can help define a question, request approval for the exact study and return counts, forward outcomes and reproducibility metadata. Compatible hosts can also show an inline result comparison and distribution.

Use the web app to save studies and confirm monitoring of supported definitions. MCP research tools do not change alerts or execute trades. Developers explains the options; API and MCP documentation covers setup and access.

Yes. The web Workbench supports saved studies and access to their definitions and evidence. Session run history also lets you reopen a run while working. Reopening a definition proposes a study; it does not by itself authorize a new scan.

If you want forward observation, confirm monitoring for a supported definition through the web workflow. Saving and alert management are separate from MCP research tools. Open saved studies.

A research match can link to its historical moment, but playback depends on the recorded streams and your plan. Research derived from earlier candle history may have no corresponding tick or order-book recording.

Use the result coverage and replay-availability information. Named archived events have their own access rules, and a permanent event page does not imply universal free replay. Replay overview.

For hosted EdgeDepth research and replay, no. The supported history is recorded server-side, so you can investigate a market you were not watching at the time.

That does not mean every market, stream or date is complete. Check coverage and replay access for the requested window. If you self-host the open-source terminal, local recording and replay use the data you supply.

Candle replay redraws bars from historical prices. Where recorded streams are available, EdgeDepth rebuilds the market sequence that produced those bars: the individual trades, the order-book changes as they arrived, the liquidity, the liquidations and the open interest.

You are not watching a recording of a chart. You are watching the market that drew it. Replay has the detail.

Research: searches can use the full verified record, starting from 15 July 2025 where the required measurements exist. Each study states its markets and dates. Coverage varies by feature and instrument; inspect coverage for your question.

Tick replay: Pro covers a rolling 90-day window. Research reaches available recorded replay history. Free has a smaller allowance, and named archived events have their own access rules. Searchable history does not guarantee a replayable book at the same date. Compare plans.

Live coverage includes 847 Binance USDT-M crypto and TradFi perpetuals, plus every Hyperliquid perpetual. Current searchable research covers Binance USDT-M. Check markets and data coverage for the symbols, dates and measurements you need.

Binance contracts tracking stocks, indices and commodities are perpetual derivatives, not direct equities or primary-exchange feeds. Live Hyperliquid coverage should not be read as a promise of searchable Hyperliquid or cross-exchange history.

Yes. Guided lessons can pause before a decisive moment, ask you to read the market, and let you place a paper trade before the recorded outcome is revealed. No money moves, and the market you are reading actually happened.

No. The free and public lessons cost nothing, and the full catalog is included with Pro: there are no per-lesson purchases.

Interactive questions and paper trading in replay work today. Saved quiz scores and progress across the catalog are still to come. Lessons shows how one runs.

No. EdgeDepth is market data and education. Nothing on this site or in the product is a recommendation to buy or sell anything, and no result is promised. Perpetual futures are leveraged instruments and carry a substantial risk of loss.

Counts describe what the record contains. They are observed history, never a forecast, and what you do with them is your decision. The disclaimer is the formal version.

Downtime-credit terms are not finalized yet, so we are not going to imply that one exists. Any credit policy will be published once the billing behavior and commercial terms are approved. Until then, the terms are the only commitment we have made.