Research is a work in progress. Estimated release: 1 December 2026. Our goal is to make it a search engine for trading setups in recorded crypto markets.

edgedepthEARLY ACCESSSign in
Menu
Terminal
Recent moves
ResearchResearch hubResearch WorkbenchInvestigate a moveLiquidation mapsPublished reportsDaily movesEvent archiveSaved studiesLive alertsMarketsData coverage
LearnGuidesReading libraryWorked examplesGuided lessonsFAQ
DevelopersAPI documentationMCP setupConnect an AI agentOpen source
Pricing

Paste a chart

Confirm what your AI reads from a chart, then test the conditions against recorded history.

Clip placeholder: take 2, about 60 s
THE RECIPEANY MCP CLIENT
  1. 01Paste the screenshotDrop any chart into Claude (or another MCP client with the EdgeDepth connector) and ask: "What followed when these conditions held?"
  2. 02The AI reads the imageConfirm the symbol, time and proposed conditions. Image interpretation can be wrong.
  3. 03EdgeDepth verifiesThe agent calls snapshot_at for the exact moment, then run_scan for every other time the market looked like that. From there on, every number is deterministic.
Connect a client first →EdgeDepth never guesses from pixels. The image stays with your AI client; only exact symbols, times, and clauses reach the research engine.

What comes back

  • The moment itself: every registry feature as-of that bucket, from snapshot_at. What the market actually was, not what the chart implies.
  • The base rate: how often that condition is even true. A pattern that holds 99% of all hours carries no information, whatever the screenshot suggests.
  • Forward outcomes over EVERY occurrence: return, best excursion, and drawdown at 30m, 1h, 4h and 24h, with denominators. Never a hindsight-selected list: outcome filters are refused by design (OUTCOME_IN_PREDICATE).
  • A reproducibility key. Same key, same bytes: anyone can rerun the exact document and get the exact result.
  • If you ask whether it is worth trading: a proposed trade test with a stop, exit and costs. Once you approve it, the answer leads with payoff, meaning whether the wins outweighed the losses after fees and slippage. How often a move followed never answers that.

Prompts that work

  • "Here is a chart. Confirm its conditions, count how often they held, then compare outcomes with the unconditional reference."
  • "Read the symbol and timestamp off this screenshot, snapshot that exact moment, and find every similar moment in the corpus. What followed, over all of them?"
  • "Someone claims this pattern front-runs pumps. Test it: occurrences, forward distribution, and the baseline for everything else."
  • "AKE/USDT, Binance Futures, 2 Sep 2026, UTC. What happened before this move, and is there an edge in trading it?"

Limits

One screenshot is a selected example. Compare all matching minutes with the unconditional reference, including counts and missing outcomes. This describes history; it does not establish predictive value.