Get started
Advanced perps workflows
Use the existing tools for a trading question, then inspect source time, coverage and calculation limits.
Trading and research clocks
Compare all-in dollar costs by position size. The cost workflow supports BTC/ETH on Hyperliquid and Lighter. It uses the same base quantity for entry and exit and reports fee, funding-state and dollar-conversion assumptions. Lighter standard-account latency and fast/API fees differ. Missing inputs withhold totals.
Fresh order-book reads use get_perps with mode: trading, or the trade workflow below. The default maximum source age is three seconds. Each adapter has a two-second read deadline; network transport and the assistant add their own time. A one-second process cache rechecks source age on every hit. A missing, future-dated or stale clock produces an unavailable estimate.
This fresh path currently supports Hyperliquid and its builder dexes. Stored cross-venue funding, OI and liquidation research has separate timestamps and coverage. A quick HTTP response does not make an hourly reading live. A book read cannot certify the age of a venue’s funding or oracle.
Prompt pack version 2026-10-19.1. These calls have schema and numerical regression checks. Availability still depends on the source. Download the JSON pack.
Compare entry, holding and exit costs by size
Compare a $1,000, $10,000 and $100,000 BTC long on Hyperliquid and Lighter. Explain the dollar cost to enter, hold for 24 hours at the stated rate, and close the same quantity against the current opposite book. Show fees and visible price impact once, fee tier, source time, funding state and missing costs. Keep price P&L separate. An unchanged-book exit and constant future funding are scenarios.
{
"workflow": "trade",
"market": "BTC",
"parameters": {
"analysis": "cost",
"sizesUsd": [
1000,
10000,
100000
],
"side": "long",
"holdingHours": 24,
"maxAgeSeconds": 3
}
}Check the answer: Source-clocked books within three seconds; entry and exit capacity; explicit fee and USDC conversion assumptions. Incompatible funding states or more than one second of book skew withhold a net venue rank.
Inspect observed BTC crowding conditions
Check these explicit BTC crowding rules using the last completed four-hour stored window. Join funding snapshot percentiles, actual linear OI quantity, observed account shares and recorded liquidations. Show missing coverage. Keep the result descriptive; do not turn a candidate into a squeeze probability or call these hourly inputs executable quotes.
{
"workflow": "screen",
"market": "BTC",
"venue": "hyperliquid",
"parameters": {
"analysis": "crowding",
"windowHours": 4,
"fundingTailPercentile": 0.95,
"minQuantityChangeFraction": 0.05,
"minObservedSideShare": 0.6,
"minPositionCoveragePerSide": 0.5,
"minObservedAccounts": 100,
"maxAgeSeconds": 7200
}
}This reads the recipe again and saves that response publicly. It does not share a private account or turn an earlier demo response into a current quote.
Check the answer: Same instrument at both OI endpoints; at least 72 prior funding readings; position coverage, account count and liquidation-recording gates; no missing input can qualify.
Research BTC price and taker-flow disagreement
Check the last completed four-hour stored BTC window for at least $10 million of net taker flow while price is within 0.5% or moves against the flow. Require qualified recorded hours and recent observed depth. Explain the source clocks and missing data. A disagreement can suggest absorption; it does not prove fills or anyone's intent.
{
"workflow": "screen",
"market": "BTC",
"venue": "hyperliquid",
"parameters": {
"analysis": "divergence",
"windowHours": 4,
"minCvdUsd": 10000000,
"priceFlatFraction": 0.005,
"maxAgeSeconds": 7200
}
}This reads the recipe again and saves that response publicly. It does not share a private account or turn an earlier demo response into a current quote.
Check the answer: Recorded-window taker coverage and matching price endpoints; missing flow is unavailable rather than zero; depth is a research snapshot.
Measure a BTC/ETH hedge relationship
Use 30 days of synchronized closed Hyperliquid BTC and ETH candles to estimate a dollar hedge for a hypothetical $10,000 BTC exposure. Show sample size, missing bars, full-sample beta, the earlier segment's frozen beta and residual volatility in the later segment. Treat this as descriptive research and disclose costs, basis risk and uncertified point-in-time selection.
{
"workflow": "study",
"parameters": {
"analysis": "hedge",
"asset": {
"venue": "hyperliquid",
"symbol": "BTC"
},
"hedge": {
"venue": "hyperliquid",
"symbol": "ETH"
},
"stepSeconds": 3600,
"notionalUsd": 10000
}
}This reads the recipe again and saves that response publicly. It does not share a private account or turn an earlier demo response into a current quote.
Check the answer: Closed bars timestamped at close; matching timestamps; no missing return treated as zero; held-out hedge uses only the earlier beta; no strategy-profit claim.
Stress a hypothetical linear portfolio
Run this hypothetical simultaneous BTC/ETH and collateral shock, with the supplied costs and haircuts. Explain exposure, equity changes and residual risk. These are example positions and assumptions. Do not infer actual holdings, current venue margin rules or an exact liquidation price.
{
"workflow": "risk",
"parameters": {
"equityUsd": 5000,
"positions": [
{
"market": "BTC",
"payoff": "linear",
"signedNotionalUsd": 10000,
"fundingCostUsd": 20,
"exitCostUsd": 5
},
{
"market": "ETH",
"payoff": "linear",
"signedNotionalUsd": -5000,
"fundingCostUsd": 10,
"exitCostUsd": 5
}
],
"shocks": {
"BTC": -0.1,
"ETH": -0.15
},
"collateral": [
{
"asset": "EXAMPLE_COLLATERAL",
"valueUsd": 1000,
"shock": -0.1,
"haircutFraction": 0,
"stressedHaircutFraction": 0
}
]
}
}Check the answer: This example produces $290 of position loss including costs, $100 of collateral loss, and $4,610 ending equity. It does not compute exact venue liquidation.
Price a visible-book entry
Estimate a $10,000 BTC buy on Hyperliquid from a book no older than three seconds. Give source time, visible capacity and VWAP. If the book is missing or stale, say unavailable. Treat the result as a snapshot estimate.
{
"workflow": "trade",
"market": "BTC",
"parameters": {
"venue": "hyperliquid",
"notionalUsd": 10000,
"side": "buy",
"maxAgeSeconds": 3
}
}This reads the recipe again and saves that response publicly. It does not share a private account or turn an earlier demo response into a current quote.
Check the answer: Fresh source time; sufficient visible depth; no hourly mark fallback; no fill guarantee.
Read a funding cash-flow scenario
Compare the funding cash-flow scenario for a $10,000 BTC short across covered venues. Show actual intervals, rate states and observation times. Leave an unknown settlement time unknown.
{
"workflow": "funding",
"market": "BTC",
"parameters": {
"analysis": "calendar",
"notionalUsd": 10000,
"side": "short"
}
}This reads the recipe again and saves that response publicly. It does not share a private account or turn an earlier demo response into a current quote.
Check the answer: Scenario dollars at stored rates; next settlement only where the schedule is known; not expected profit.
Study settled funding persistence
Analyze the last 30 days of stored BTC settlements on Hyperliquid. Show percentiles, sign flips, run durations, coverage gaps and 95% moving-block bootstrap confidence bands using 24-settlement blocks. Explain stationarity and block-size assumptions. Separate recorded-series uncertainty from future funding or wallet payments.
{
"workflow": "funding",
"parameters": {
"venue": "hyperliquid",
"symbol": "BTC",
"confidenceBlockObservations": 24
}
}This reads the recipe again and saves that response publicly. It does not share a private account or turn an earlier demo response into a current quote.
Check the answer: Settlement-level observations; blocks do not cross gaps; insufficient samples and ambiguous schedule changes are refused. Confidence bands do not predict future APR.
Check current contract size and margin rules
Read current public BTC contract metadata on Hyperliquid. Show native size precision, lot size, variable price precision, maintenance tiers and the maintenance-margin scenario for $10,000 notional. Cite the metadata hash and fetch time. Keep native update time and account feasibility unknown when not supplied.
{
"workflow": "contracts",
"market": "BTC",
"parameters": {
"venue": "hyperliquid",
"liveRules": true,
"notionalUsd": 10000
}
}This reads the recipe again and saves that response publicly. It does not share a private account or turn an earlier demo response into a current quote.
Check the answer: Exact venue and native symbol; fetched time is not a source update time. A public maintenance scenario does not certify account margin, collateral conversion or a liquidation price.
Investigate a qualified position build
Show five markets passing the position-build screen. Explain the passing window and thresholds, comparable open-interest change, funding state and source time. Keep directional interpretations conditional.
{
"view": "builds",
"limit": 5
}This reads the recipe again and saves that response publicly. It does not share a private account or turn an earlier demo response into a current quote.
Check the answer: Passing windows filtered before ranking and pagination; open interest is not directional inflow.
Explain observed account moves
Explain the largest observed changes in BTC positions on Hyperliquid over four hours. Separate opens, additions, reductions, closures and flips. Give account-book completeness and stream gaps before any cohort conclusion.
{
"view": "position_changes",
"market": "BTC",
"window": "4h",
"limit": 5
}Check the answer: Observed account scope only; missing accounts and undisclosed off-venue hedges stay unknown.
Research stock perps while stocks are closed
Which stock-perp moves agree across venues while the underlying stocks are closed? Show funding, liquidity, source clocks and historical opening-session grading. Do not describe a stale stock reference as a live oracle.
{
"view": "weekend",
"limit": 5
}This reads the recipe again and saves that response publicly. It does not share a private account or turn an earlier demo response into a current quote.
Check the answer: Underlying session and reference age disclosed; historical weekend grades do not predict the next open.
Save conditions and acknowledged comparison baselines
Prospective cost studies freeze a private specification and five-minute sample cohort before outcomes arrive. Late and missing samples remain in the cohort. Website writes require your signed-in account; the read-only MCP cannot amend a study. These are quote-cost observations, not trading returns.
The signed-in perps desk saves compound conditions against exact venue and symbol identities. All conditions must pass; missing or stale observations stay unknown. The existing queue samples about every five minutes, with longer intervals possible as the monitored universe grows. This is a research monitor, not an immediate trading alert.
Read a new report and mark it received to advance your comparison baseline. A failed source read, unacknowledged response or inbox delivery does not advance it. Alerts have persistent account-inbox receipts, capped retries, cooldowns and condition invalidation. An inbox receipt does not establish email delivery or a human read.
Replay an edition known to the platform at a cutoff
The replay workflow accepts a past at and knownAt cutoff, with at ≤ knownAt. It selects an immutable normalized edition captured by the platform before that cutoff, including its own registry, source revisions and venue read failures. Prospective editions are retained for 90 days; a cutoff before capture began returns unavailable.
platformKnownAtCertified concerns platform capture only. Original vendor first-publication times, old executable books and pre-capture universe completeness remain uncertified, so pointInTimeCertified stays false. An hourly bucket is not a native trade timestamp.
Propose a recipe
A useful contribution includes the exact tool call, a public source, expected units, time and coverage requirements, a normal example and an example that must return unavailable. Keep private account data, keys and positions out of a public submission. Recipe changes need provenance and regression review before joining the versioned pack.
Public tools remain read-only. Saved monitors and acknowledgements use the authenticated website. Private exchange accounts and external notification channels require their own authorization, data adapters and delivery verification.