Shared by every crypto get_alpha / get_exposure endpoint:
period — bar size, {n}min / {n}h / {n}d ({n}m is read as minutes). Documented values are 5min, 15min, 1h, 4h, 8h, 1d; indicators with a coarser minimum say so in their block. A missing or unparseable period is a 403 {"error": "period is required"} (a 400 on unusual_movement and liquidation).
start_date / end_date — YYYY-MM-DD (UTC). end_date defaults to today; start_date defaults to end_date − 365 days. A range longer than 365 days is silently clamped (start_date moved to end_date − 365 days). A malformed date is not validated and ends in a 500 — always send YYYY-MM-DD.
symbol — Binance USDT-M perpetual, e.g. BTCUSDT; BTC, btc, BTC/USDT, BTC-USDT are also accepted (a bare token gets USDT appended). The get_symbols endpoints list the covered symbols.
Only closed bars are returned — the still-forming last bar is dropped.
Response is {"data": {"timestamp": [...], "alpha": [...], ...}} — parallel arrays, one element per bar. timestamp is Unix seconds (UTC), the bar label.
For history beyond 365 days, send one request per year and concatenate.
stat object — returned by indicators whose block lists it. A logistic model fitted on the symbol's past year of this indicator versus the next-24h price move:
Field
Type
Description
up_prob
float
Probability of a 24h up move given the latest indicator value (0–1)
avg_up_return
float
Mean 24h return over past up moves (decimal, 0.02 = +2%)
avg_down_return
float
Mean 24h return over past down moves (decimal, negative)
Binance USDT-M perpetual last price; tokens not listed there fall back to Binance spot, then BingX perpetual. change_24h normally comes from the same venue as price (the two are cached separately, 5 s each, so they can briefly differ)
data is keyed by token (BTC, 1INCH, …). Each token maps indicator name → {param: value} ("-" when the indicator has no parameter). "" = insufficient data. Besides the indicators:
Key
Description
statistics
up_prob (prob. of a 24h up move), exp_value (expected return), avg_up_return, avg_down_return, return_ratio, is_data_sufficient
price
{"-": 70000} — current price
price_change
{"15min": ..., "1h": ..., "24h": ...} — % change
market_cap
{"-": 1234567890} — USD market cap
market_cap_percentile
{"-": 85.3} — percentile among all listed coins
funding_rate
{"binance": -0.01, ...} — per exchange
oi_imbalance
{"-": 0.12} — OI imbalance (full detail: /oi_imbalance/get_overview_data)
fields — one entry per indicator: id, name (key used in data), param (parameter name or null), name_en, name_zh.
note — keyed by indicator id: a list of value bands {min, max, note: {en, zh}, direction: {en, zh}, color, background_color} used to label a value (e.g. max: -3 → "Overly Bearish").
Errors
Shared errors only.
Notes
Use this first for any multi-coin screen or ranking — one request covers every symbol.
Always check statistics.is_data_sufficient before using statistics.
Always fetch fresh; do not reuse an earlier response.
2020-01-01, or the contract's listing date if later — every interval, 1min included
Update
The latest closed 1-minute bar is usually available within the next minute; in rare degraded cases the newest bars can lag about 5–10 minutes. Server cache: 30 s (1min), 300 s (1h / 4h / 1d), 60 s (every other period); only 5 s when the latest closed bar is not in yet
Source
Binance USDⓈ-M futures
Coverage. Every Binance USDⓈ-M perpetual whose symbol ends in USDT — 740 contracts as of October 2026, including tokenized stocks, commodities and other TradFi USDT perpetuals. All 740 have their full 1-minute history stored server-side, from 2020-01-01 or the listing date. A newly listed perpetual becomes queryable about 5–10 minutes after listing.
Not covered: contracts quoted in USDC, USD1, U or BTC (e.g. BTCUSDC, ETHBTC) and quarterly delivery contracts — these return the 400 unknown symbol or no data error below. Delisted contracts are not guaranteed: one may answer with whatever history is still stored or with that same 400 (those delisted before October 2026 have no 1min–4min history).
Parameters
symbolstringqueryRequired
Binance USDⓈ-M USDT perpetual. A bare token gets USDT appended; a bare token that only exists with a multiplier prefix resolves to it (PEPE → 1000PEPEUSDT)
AllowedBTCUSDT; also BTC, btc, BTC/USDT, BTC-USDT
periodstringqueryRequired
Bar size. Any whole-minute size is accepted, not only the listed examples
Defaultend_date − 365 days (period ≥ 5min); end_date − 30 days (period < 5min)AllowedYYYY-MM-DD (UTC)
end_datestringquery
Last day, inclusive (bars through the end of that UTC day)
Default nowAllowedYYYY-MM-DD (UTC)
Per-request window: period ≥ 5min → at most 365 days; a longer range is silently clamped (start_date moved to end_date − 365 days, no error). Period < 5min → at most 30 days; a longer range is a 400.
A bare JSON array (not wrapped in {"data": ...}), ascending by time.
Field
Type
Unit
Description
time
float
Unix seconds (UTC)
Bar open time
open / high / low / close
float
quote currency (USDT)
Prices
volume
float | null
base asset
Traded volume; null when the source has no volume for that bar
Errors
Status
Body
When
403
{"error": "symbol is required"}
symbol missing
403
{"error": "period is required"}
period missing or not parseable
400
{"error": "minimum supported period is 1min"}
period below one minute
400
{"error": "periods below 5min must be a whole number of minutes"}
e.g. 90s
400
{"error": "date range exceeds max 30 days for periods below 5min"}
Period < 5min and range > 30 days
400
{"error": "invalid date format, expected YYYY-MM-DD"}
Malformed start_date / end_date
400
{"error": "unknown symbol or no data: <symbol>"}
Symbol not covered (see Not covered) or not a real contract, e.g. BTCUSDC, ETHBTC, a typo. <symbol> echoes the value you sent. Not transient — do not retry
Notes
A window entirely before the listing date of a covered contract returns 200 with [], not an error.
At 1min every bar returned is closed. A longer period can end in a partial bar when end_date is today — it is built from the bars closed so far (at 07:09 UTC the 07:001h bar holds 07:00–07:05); drop it if you need closed bars only.
For history beyond one window, send one request per window and concatenate (e.g. 3 years of 5min = 3 requests; 1 year of 1min = 13 requests of ≤30 days).
All 740 USDT perps have full 1-minute history on disk, so a deep request is answered from stored data — there is no slow first fetch.
2024-12-27 (earliest 1d bar returned; market-wide series, no symbol dimension)
Update
Every 5 minutes
Source
Blave — one market-wide series (composite of Blave's crypto indicator overviews, no symbol dimension)
Parameters
No symbol: alpha is a single market-wide series, not a per-coin one. BTCUSDT is used internally only to align the bars; it is not returned and does not change alpha. See Crypto indicator conventions.
{"data": {<condition_id>: {...}}}, keyed by condition id (a JSON string key, e.g. "12"; send it back as the integer condition_id). Only this account's own conditions.
Field
Type
Description
filters
object[]
The condition's filters, all ANDed: indicator_id (int), indicator_name ({"en", "zh"}), param (string; "-" = the indicator's default), operator_id (int), operator_symbol (string, e.g. >), value (threshold)
note
string
The name the account saved the condition under (a condition you save is named here)
name_en / name_zh
string | null
Set only on official / community conditions; on your own conditions expect empty or null and read note instead
description_en / description_zh
string | null
Set only on official / community conditions; may be empty or null
One per distinct filter indicator, in filter order: en / zh (display name, with the parameter in parentheses when there is one), name (indicator key), type, url; plus value_format, value_decimals, polarity when the indicator defines them
symbols
object[]
Tokens matching every filter
symbols[].symbol
string
Token without USDT (e.g. BTC)
symbols[].id
int
Token id (0 when unknown)
symbols[].user_selected
bool
Token is on the account's watchlist
symbols[].price
float
Latest price in USDT (same feed as /price; 0 when unavailable)
symbols[].24h_chg
float
24h change as a fraction (0.025 = +2.5%)
symbols[].alphas
float[]
The token's value for each filter indicator, parallel to alpha_names
symbols[].alpha_notes
object[]
Per indicator: indicator_id, color, direction, note (plus background_color when a rule matched; colornull and empty strings when none did)
Errors
Status
Body
When
400
{"error": "Missing parameter: condition_id is required."}
Missing
400
{"error": "Invalid parameter: condition_id must be an integer."}
Not an integer
404
{"error": "Invalid condition_id: <id> not found for user."}
Trailing one-year daily window — a call returned 366 daily points, 2025-09-16 → 2026-09-16, each sector seen starting at 0.0. There are no parameters, and each rebuild recomputes only the trailing 365 days, so no longer history exists
Update
Built by a scheduled job, ≈ hourly — it reruns 13 min after the previous run ends, and recent rebuilds landed 50–60 min apart; served from the job's output with no extra cache
{"data": {"alpha": {<sector>: {"data": [...], "name_en": ..., "name_zh": ...}}, "timestamp": [...]}}. timestamp sits next to alpha at the top of data, not inside each sector; it is the shared time axis for every sector series.
Field
Type
Description
timestamp
float[]
Unix seconds (UTC), one per point — 366 daily points in the observed response
alpha.<sector>.data
float[]
The sector series, same length and order as timestamp (366); the first element is 0.0
Built by a scheduled job, ≈ hourly — it reruns 13 min after the previous run ends, and recent rebuilds landed 50–60 min apart; served from the job's output with no extra cache
Z-score of the timeframe-summed OI change (or volume) vs its 30-day mean
stat
object
See stat object
Errors
Status
Body
When
403
{"error": "symbol is required"} / {"error": "period is required"}
Missing parameter
403
{"error": "Invalid score_type"}
score_type not allowed
Notes
timeframe is not validated server-side. One value outside the list was tried, 2h, and it returned 200 with data. That is a single observation — it does not establish that any {n}h works. Stay on the listed values unless you verify the one you need.
black = squeeze on (Bollinger Bands inside Keltner Channels), grey = squeeze off
stat
object
See stat object
Errors
Status
Body
When
403
{"error": "symbol is required"}
Missing parameter
Notes
The series is padded: BTCUSDT rows exist from 2020-01-01, but alpha is 0.0 until 2020-02-09. A request starting earlier than the Data from date returns those padded rows — they are filler, not signal. Drop leading zeros, or start at the Data from date.
Blave — one market-wide series (all coins these traders hold, no symbol dimension)
Parameters
No symbol: alpha is a single market-wide series, not a per-coin one. BTCUSDT is used internally only to align the bars; it is not returned and does not change alpha. See Crypto indicator conventions.
{"data": {"timestamp": [...], "alpha": [...]}}. No stat.
Field
Type
Description
timestamp
float[]
Unix seconds (UTC)
alpha
float[]
Net exposure of Blave's top traders (top 10% by account assets) — the median, across those traders, of each trader's net notional position ÷ own assets, ×100 (positive = net long). A raw ratio, not a z-score
(short − long liquidations, in coins) summed over timeframe ÷ its 30-day rolling std; mean not subtracted. > 0 = more shorts than longs liquidated, < 0 = more longs
stat
object
See stat object
Errors
Status
Body
When
400
{"error": "symbol is required"} / {"error": "period is required"}
Missing parameter
Notes
The series is padded: BTCUSDT rows exist from 2020-01-01, but alpha is 0.0 until 2023-01-01. A request starting earlier than the Data from date returns those padded rows — they are filler, not signal. Drop leading zeros, or start at the Data from date. Symbols other than BTC/ETH turn non-zero later still, so check per symbol.
Estimated allocation of Binance open interest to each bucket (model estimate, not real positions)
cumsum
float[]
USD
Cumulative liquidation exposure across buckets
liquidation
object
USD
{"24h": {"buy_liq": [...], "sell_liq": [...]}} — actual Binance force-order liquidations over the last 24 h per bucket, each divided by a fixed 0.3 Binance-share assumption: buy_liq = short liquidations, sell_liq = long liquidations
Computed per request: the current map is rebuilt (server cache 5 s) and compared with saved map snapshots from ~1 h, ~8 h and ~24 h ago; snapshots are taken every 30 min for every symbol by a scheduled job, plus on each map request
Blave (Binance, Bybit, Gate.io, OKX, HTX, Bitfinex liquidation feeds; USD notional converted at collection time, so exchanges can be summed)
One coin's forced liquidations across every exchange feed Blave collects: four rolling windows with the per-exchange split, an hourly 24-point series, and each feed's coverage notes. This is real liquidation flow (like get_alpha's input), not the model estimate behind get_map.
Parameters
symbolstringqueryRequired
Coin. The quote suffix is stripped; the response symbol is the normalised coin name
Summed across exchanges; long = long positions liquidated (price fell), short = shorts liquidated (price rose)
windows.<h>.long_pct / short_pct
float / null
ratio 0–1
Share of the window total; null when the window total is 0
windows.<h>.covered_hours
float
h
How many hours of the window actually have buckets (< h right after a collector restart)
windows.<h>.by_exchange
object
USD
{exchange: {total_liq_usd, long_liq_usd, short_liq_usd}}. An exchange with no event in the window has no key — read exchanges[] to tell "feed alive, this coin quiet" from "feed absent"
exchanges[]
object[]
—
One row per feed (six today, present even with no event), sorted by 24 h total desc, then roster order
exchanges[].exchange
string
—
binance, bybit, gate, okx, htx, bitfinex
exchanges[].listed
bool / null
—
Whether that exchange lists a USDT-margined perpetual for the coin; null when the contract list could not be fetched (Bitfinex always null)
exchanges[].last_event_at
string / null
ISO 8601 UTC
Latest bucket seen from that feed (any coin); null if never seen
exchanges[].price_basis
string
—
Price used for USD conversion: trade_avg, bankruptcy, unverified
false when the coin may have been cut from a full per-bucket detail list (feeds keep the top 100 coins per 5-minute bucket): windows can then under-count. Top-50 coins are almost always true
series.bucket_seconds
int
s
Always 3600
series.points
object[]
—
Always 24 entries, oldest → newest, on clock hours; the last is the current hour up to updated_at (the snapshot is rebuilt every 5 min, so it can trail the request by a few minutes). Each {ts, long_liq_usd, short_liq_usd}; hours with no event are 0, never missing
windows["24"] is the same rolling frame as the exchange matrix, so the same coin shows the same 24 h number there. series is on clock hours, so Σ points ≠ windows["24"] by design: take totals from windows, timing from series.
Errors
Status
Body
When
400
{"error": "symbol is required"}
symbol missing, blank, longer than 32 chars, or nothing left after the quote suffix is stripped (e.g. symbol=USDT)
404
{"error": "<SYMBOL> is not a collected symbol"}
No exchange lists the coin and no feed reported it in the last 24 h — an answer, not a transient
503
{"error": "liquidation data unavailable"}
Upstream buckets did not answer — retry later; never read as "0 liquidations"
Notes
Snapshot only: no start_date / end_date, and nothing older than 24 h. For history use get_alpha (a normalised long/short imbalance, not USD amounts).
Symbol handling differs from the other liquidation endpoints: get_alpha / get_map need the Binance perp form (BTCUSDT); here BTC and BTCUSDT are the same coin.
Rolling window, 5-minute aligned (default last 24 h)
Update
Every 5 minutes (server cache 5 min)
Source
Blave (Binance, Bybit, Gate.io, OKX, HTX, Bitfinex liquidation feeds; USD notional converted at collection time, so exchanges can be summed)
Whole-market forced liquidations, aggregated per exchange and as a coin × exchange matrix — the market-wide companion to /liquidation/get_coin. Same rolling 5-minute-aligned frame, so a coin's 24 h total is the same number on both.
Parameters
hoursintquery
Window length. Outside the range → 400 (it is not clamped); a non-integer value silently falls back to the default
Default24Allowed 1–168
top_nintquery
How many coins to list individually; the rest go to others. Same range / non-integer rules as hours
Default10Allowed 1–50
Example
Python
params = {"hours": 24, "top_n": 10}data = requests.get(f"{BASE_URL}/liquidation/get_exchanges", headers=headers, params=params, timeout=30).json()["data"]print(data["total"]["total_liq_usd"], data["total"]["short_pct"])print([(e["exchange"], e["total_liq_usd"], e["coverage"]) for e in data["exchanges"]])# Live capture 2026-09-22 (abridged with ...; anonymous studio twin /studio/charts/liquidation/exchanges,# same inner shape — the API-key response wraps it in "data" and allows top_n up to 50):# {"exchanges": [{"exchange": "binance", "total_liq_usd": 390250999.16, "long_liq_usd": 69367041.38,# "short_liq_usd": 320883957.78, "long_pct": 0.1777, "short_pct": 0.8223, "events": 46200,# "last_event_at": "2026-09-22T05:50:00+00:00", "price_basis": "trade_avg",# "coverage": "sampled_1s", "time_basis": "event"}, ... 6 rows],# "coins": [{"token": "BTC", "token_id": 1, "total_liq_usd": 397165128.6, "long_liq_usd": 68908479.16,# "short_liq_usd": 328256649.44,# "by_exchange": {"binance": {"long_liq_usd": 17994806.07, "short_liq_usd": 173732353.71,# "total_liq_usd": 191727159.78}, ... 6 exchanges}}, ... 10 coins],# "others": {"coin_count": 835, "total_liq_usd": 82658628.27, "long_liq_usd": 34622797.57,# "short_liq_usd": 48035830.7, "by_exchange": {...}},# "total": {"total_liq_usd": 743631889.64, "long_liq_usd": 175240866.53, "short_liq_usd": 568391023.1,# "long_pct": 0.2357, "short_pct": 0.7643},# "buckets": 1846, "covered_hours": 24.0, "window_hours": 24,# "window_start": "2026-09-21T05:55:00+00:00", "window_end": "2026-09-22T05:55:00+00:00",# "updated_at": "2026-09-22T05:50:00+00:00"}
One row per feed, sorted by total desc: exchange, total_liq_usd, long_liq_usd, short_liq_usd, long_pct, short_pct, events, last_event_at. Every feed is always listed: amounts null = no data received from that feed in the window (sorted last), 0.0 = connected but no liquidations
exchanges[].price_basis / .coverage / .time_basis
string
—
How comparable that row is: trade_avg / bankruptcy / unverified; full, full_unstated, sampled_1s, aggregated_1s, sampled_undisclosed; event. Keep them when re-publishing the numbers — a sampled feed is not a full one
coins[]
object[]
—
The top top_n by cross-exchange total: token, token_id (0 when the symbol has no CoinMarketCap id), total_liq_usd, long_liq_usd, short_liq_usd, by_exchange{name: {total/long/short_liq_usd}}. by_exchange is sparse — a feed with nothing for that coin has no key; read it as 0 when that feed's exchanges[] total is not null
others
object
—
Everything top_n cut off, same fields plus coin_count, so coins[] + others adds back to each exchange's total
{"error": "hours must be between 1 and 168"} / {"error": "top_n must be between 1 and 50"}
Out-of-range parameter
Notes
No data at all (feeds down, or the data source unreachable) is still a 200 with the same shape and every amount null, buckets: 0 — never a table of zeros. Check total.total_liq_usd before using the numbers.
Totals only — no time series. Hourly shape for one coin is /liquidation/get_coin; a normalised long-history signal is /liquidation/get_alpha.
The window is rolling, not a calendar day, and it is the same frame /liquidation/get_coin uses for windows["24"].
Every collected coin × every long/short-ratio source, as of now. Ten sources: Binance, OKX and Gate.io each publish three kinds — all accounts, top-trader accounts, top-trader positions — and Bybit publishes accounts only. Rows are ordered by Binance open-interest notional. Only coins that resolve to a CoinMarketCap crypto are listed; the API-key response is the full table (full: true), the anonymous studio twin only the rows of its 30-coin whitelist.
Parameters
None.
Example
Python
data = requests.get(f"{BASE_URL}/long_short_ratio/get_table", headers=headers, timeout=30).json()["data"]keys = [s["key"] for s in data["sources"] ifnot s["stale"]]btc = next(c for c in data["coins"] if c["token"] == "BTC")print({k: btc[k] for k in keys if btc[k] is notNone})# Live capture 2026-09-22 via the anonymous studio twin (/studio/charts/long_short_ratio/table) —# same inner shape; the API-key response wraps it in "data" and has full: true with every row.# {"coins": [{"binance_account": 0.8818, "binance_top_account": 1.0149, "binance_top_position": 2.2666,# "bybit_account": 1.0375, "gate_account": 0.7294, "gate_top_account": 0.6944,# "gate_top_position": 1.6436, "okx_account": 0.9084, "okx_top_account": 0.8655,# "okx_top_position": 1.0222, "token": "BTC", "token_id": 1}, ...],# "sources": [{"exchange": "binance", "key": "binance_account", "last_at": "2026-09-22T05:50:00+00:00",# "stale": false, "type": "account"}, ...,# {"exchange": "bybit", "key": "bybit_account", "last_at": "2026-09-22T05:50:00+00:00",# "stale": false, "type": "account"}],# "summary": {"binance_long_majority": 436, "binance_tokens": 509},# "tokens_shown": 30, "tokens_total": 611, "full": false, "updated_at": "2026-09-22T05:50:00+00:00"}
Collection start (first stored bucket, measured on BTC 2026-09-22)
Every source is now older than the 7 days this dataset serves, so none of them limits a query — the dates matter only if you are told a source "has no history".
That feed's newest sample is over 1 hour old (measured against now). Top-level updated_at is the newest source and says nothing about any single one — judge a source by its own last_at / stale
summary.binance_tokens / .binance_long_majority
int
—
Coins with a Binance account ratio / of those, how many are above 1. Computed over the whole table, not the rows shown
tokens_shown / tokens_total
int
—
Rows returned / rows in the table. Equal on an API key
full
bool
—
true on an API key; false on the anonymous studio twin
updated_at
string
ISO 8601 UTC
Newest bucket across sources
"Top trader" is a different population at each exchange — do not compare the levels across exchanges.
Exchange
Who counts as a top trader (official wording)
Binance
"the top 20% users with the highest margin balance"
OKX
"the top 5% of traders with the largest open position value"
Gate.io
Never published — the docs only name the fields
So binance_top_account above okx_top_account says nothing about which venue's big traders are more long. Compare each source against its own history instead.
Errors
Status
Body
When
503
{"error": "long short ratio data unavailable"}
No fresh snapshot — retry later; never read as an empty market
Notes
A cross-section, not history. For a series use /long_short_ratio/get_coin (7 days) — there is no long-history endpoint for this dataset.
Granularity is 5 minutes, but each source publishes on its own clock — do not describe it as "updates every 5 minutes".
{value, ts} — that source's newest 5-minute sample; null when the source does not list the coin
series.bucket_seconds / .days
int
s / d
3600 / 7
series.timestamp
int[]
epoch s
168 slots, oldest → newest, one per hour
series.<source key>
float[] / null
ratio
Last sample of each hour; null for an hour with no sample. The whole field is a single null (not an array) for a source that does not list the coin
series.price
float[] / null
USD
Binance perp close on the same frame. A single null (with price_symbol / price_multiplier also null) when the coin has no Binance perp
series.price_symbol / .price_multiplier
string / int
—
Which perp the price came from and its contract multiplier (1000PEPEUSDT → 1000; price ÷ price_multiplier = price per coin)
series.provisional_from
int / null
epoch s
First slot that can still change (sources finalise up to ~10 min late); null when no source has data
sources[]
object[]
—
The full roster, same fields as the table plus listed
sources[].listed
bool
—
false = that exchange has no such feed for this coin; its latest and series entries are null
symbol / token_id / updated_at
—
—
As in the table
Errors
Status
Body
When
400
{"error": "symbol is required"}
symbol missing, blank, over 32 chars, or nothing left after the quote suffix is stripped
404
{"error": "<SYMBOL> is not a collected symbol"}
No source collects the coin — an answer, not a transient
503
{"error": "long short ratio data unavailable"}
A listed source could not be read. null only ever means "that venue does not list it"
Notes
The cross-exchange "top trader" caveat on /long_short_ratio/get_table applies here too.
7 days is the whole history this endpoint serves; there is no start_date / end_date.
About the "anonymous studio twin" in every capture on this page's raw-data endpoints: the studio pages (/studio/charts/...) serve the same payload from the same builder without a key, which is where these examples were taken. They are not a substitute for the API-key endpoints: the tables are cut to 30 coins (full: false), and a coin outside the studio whitelist answers 403 {"error_code": "ERR001", ..., "anon_whitelist": [...]} instead of data. The API-key endpoints documented here have no whitelist and no row cut.
Every collected coin × five exchanges, in USD notional. Basis: USDT-margined perpetuals only, USD notional, one side. An exchange that reports both sides is halved (exchanges[].side_factor, Gate.io = 0.5). Because every value is already USD notional, multiplied contracts (1000PEPE and friends) add across exchanges with no rescaling.
Parameters
None.
Example
Python
data = requests.get(f"{BASE_URL}/oi_imbalance/get_table", headers=headers, timeout=30).json()["data"]crowded = sorted((c for c in data["coins"] if c["oi_mcap"] is notNone), key=lambda c: -c["oi_mcap"])[:10]print(data["total"]["oi"], [(c["token"], round(c["oi_mcap"], 4)) for c in crowded])# Live capture 2026-09-22 (abridged with ...; anonymous studio twin, same inner shape):# {"coins": [{"token": "BTC", "token_id": 1, "oi_total": 18645450917.48, "chg_1h": -0.001775,# "chg_4h": -0.001046, "chg_24h": 0.128725, "market_cap": 1629947967503.241, "oi_mcap": 0.011439,# "by_exchange": {"binance": {"oi": 9417422808.02, "chg_1h": 2e-05, "chg_4h": 0.010203, "chg_24h": 0.067527},# "bingx": {"oi": 1328003926.8, ...},# "bybit": {"oi": 2569097877.96, "chg_1h": -0.008636, "chg_4h": -0.030159, "chg_24h": null},# "gate": {...}, "okx": {"oi": 2575361192.92, ..., "chg_24h": null}}}, ...],# "exchanges": [{"key": "binance", "oi": 24712717530.01, "side_factor": 1.0, "since": null, "full_7d": true,# "last_at": "2026-09-22T05:25:00+00:00", "stale": false}, ... 5 rows],# "total": {"oi": 47428938620.05, "chg_1h": -0.008919, "chg_4h": -0.010508, "chg_24h": 0.073881,# "n_exchanges": 5, "n_full_7d": 3},# "summary": {"oi_mcap_leader": "SAGA", "oi_mcap_leader_value": 1.718336},# "tokens_shown": 30, "tokens_total": 622, "full": false, "updated_at": "2026-09-22T05:30:00+00:00"}
Collection start (first stored bucket, measured on BTC 2026-09-22)
Exchange
Collecting since
binance
2021-12-01
bingx
2025-04-15
gate
2026-08-12
okx
2026-09-21
bybit
2026-09-21
OKX and Bybit are why n_full_7d is 3 of 5 and why their chg_24h was still null on 2026-09-22 — read exchanges[].since / full_7d rather than these dates, which age out.
Decimal change (0.0675 = +6.75 %). Each window counts only the exchanges that have a baseline at that window's start, numerator and denominator over the same set; null when none does
coins[].market_cap
float / null
USD
CoinMarketCap market cap; null when there is no usable one
coins[].oi_mcap
float / null
ratio
oi_total ÷ market_cap; null when market_cap is
coins[].by_exchange
object
—
{exchange: {oi, chg_1h, chg_4h, chg_24h}}, one key per exchange; null when the coin is not listed there. A per-window null means that exchange had no baseline that far back — a feed younger than the window
The feed's first day when it has under 7 days of history (full_7d: false), else null / true. Informational on this table (it has no 7-day window); on /oi_imbalance/get_coin such a feed has no 7 d baseline and is left out of series.total
Highest OI ÷ market cap among the top 100 coins by OI
tokens_shown / tokens_total / full / updated_at
—
—
As in the long/short-ratio table
Not the same number as the OI 失衡 indicator./oi_imbalance/get_overview_data (and the oi_imbalance alpha behind screener conditions and alerts) is a three-exchange aggregate — Binance + OKX + BingX. This table is five exchanges on the basis above. The two oi_total values differ, so a threshold tuned on one does not carry over to the other.
Errors
Status
Body
When
503
{"error": "open interest data unavailable"}
The job has no fresh result (missing or over an hour old) — never a table of zeros
Notes
chg_24h: null on OKX / Bybit in the capture above is the young-feed rule, not a gap in the market: those two started on 2026-09-21, so they had no 24 h baseline yet.
A cross-section, not history. One coin's last 7 days (hourly, all five exchanges) is /oi_imbalance/get_coin; backtest-length history for one coin on one exchange is /oi_imbalance/get_history.
One coin's open interest per exchange, on the same basis as the table: current value and share, four change windows with the exchanges each counted, and a 7-day hourly total line.
The full five-exchange roster: key, oi (USD), share (of oi_total), chg_1h/4h/24h/7d, side_factor, listed, since, full_7d, last_at, stale
exchanges[].listed
bool
—
false = the coin is not listed there; its values are null
oi_total / market_cap / oi_mcap
float
USD / ratio
As in the table
oi_mcap_rank / tokens_total
int / null
—
Rank by oi_mcap in the table job's last round; null when that job has no fresh result
windows.<1h|4h|24h|7d>.chg / .chg_usd
float / null
fraction / USD
Change of the counted exchanges' total
windows.<w>.exchanges
string[]
—
Which exchanges were counted in that window — a young feed is absent, so 24 h / 7 d can count fewer exchanges than 1 h
series.timestamp
int[]
epoch s
168 hourly slots, oldest → newest
series.total
float[] / null
USD
Total OI per slot — only the series.total_exchanges (the full_7d feeds) are in this line, so it is not comparable in level to oi_total. A slot is null when any of those exchanges has no sample that hour; the whole field is null when no exchange is full_7d for the coin
400 / 404 / 503 exactly as /long_short_ratio/get_coin, with {"error": "open interest data unavailable"} on 503.
Notes
windows["1h"]["exchanges"] having five entries while windows["24h"] has three is the young-feed rule, not missing data — read that list before comparing windows.
The OI 失衡 caveat from /oi_imbalance/get_table applies to this oi_total too.
Seven days only. For a longer series to backtest on, use /oi_imbalance/get_history (one exchange, coins rather than USD — a different basis from oi_total).
5-minute resolution — binance 2021-12-01, bybit 2025-08-21, gate 2026-03-28 (each: or the coin's listing date there, if later). Currently listed contracts only
Update
5-minute buckets
Source
Blave (Binance, Bybit, Gate.io open-interest feeds; close from the Binance USDT-M perpetual)
One coin's open interest on one exchange, as a time series you can backtest on — the same basis as the Studio dashboard 未平倉量 card. alpha is one-sided open interest in coins, not USD. Multiplied contracts are converted back to coins (1000PEPE returns a PEPE count). Each bucket is the last reading of its period — open interest is a stock, never summed.
Parameters
symbolstringqueryRequired
Coin, same rule as /oi_imbalance/get_coin
AllowedBTC, BTCUSDT, btc
periodstringqueryRequired
Bucket size. 1w is a 400 (unit not supported); 1m is read as 1min, so it is a 400 for being below 5min
One-sided open interest on oi_exchange, last reading of the bucket
close
float[]
USDT
Binance USDT-M perpetual price — per contract unit, so per 1000 coins for 1000PEPE
exchanges
string[]
—
The exchange actually read, e.g. ["binance"]
alpha × close is not USD open interest on a multiplied contract.close there is the price per contract unit (per 1000 coins for 1000PEPE) while alpha is in single coins — divide close by the multiplier first. For non-multiplied contracts such as BTC, alpha × close ≈ USD open interest.
Gate.io: always send start_date ≥ 2026-03-28 — the default window (end_date − 365 days) starts before Gate's history and is a 400.
Not the same number as /oi_imbalance/get_table / get_coin (five exchanges summed, USD) and not the OI 失衡 indicator (/oi_imbalance/get_overview_data).
Errors
Status
Body
When
400
{"error": "symbol is required"} / {"error": "period is required"}
Missing (a 400 here, not the 403 of the indicator endpoints)
400
{"error": "period must be at least 5min, in min / h / d units"}
period below 5min (including 1m, read as 1min) or in a unit other than min / h / d (1w)
400
{"error": "invalid date format, expected YYYY-MM-DD"}
Malformed start_date / end_date
400
{"error": "start_date must not be after end_date"}
start_date > end_date
400
{"error": "oi_exchange must be one of binance, bybit, gate"}
oi_exchange outside the list
400
{"error": "no <exchange> open interest for <TOKEN> at start_date; this exchange's history may start later — try a later start_date"}
The window starts before that exchange's history (see Data from). Without start_date, oi_exchange=gate always hits this until its history passes 365 days (≈ 2027-03) — for Gate always send start_date ≥ 2026-03-28
404
{"error": "<TOKEN> is not a collected symbol on <exchange>"}
That exchange does not list the coin, the coin has no Binance perpetual, or it is delisted — an answer, not a transient
503
{"error": "<exchange> open interest data unavailable for <TOKEN>"}
Data could not be read right now — retry
Notes
200 with empty arrays means no bucket in that window (e.g. the coin listed after it) — do not cache it as a permanent "no data".
The last bucket keeps moving until its period closes when end_date is today.
For history beyond 365 days, send one request per year and concatenate.
Built by a scheduled job, ≈ every 15 minutes (5-minute source buckets; Gate / OKX land ≈ 10 min late)
Source
Blave (Binance, OKX, Gate.io taker-flow feeds)
Taker (market-order) buy and sell turnover per coin, in USD, across three exchanges: Binance from the 5-minute kline's taker-buy quote volume (sell = the bar's total minus it), OKX from its USD taker volume, Gate.io from taker contracts × the same row's multiplier and mark price. Perpetuals only — no spot — and aggregated from 5-minute bars, not from trade-by-trade prints. Each exchange's own reported turnover is used; volume is never multiplied by a price borrowed from another venue.
Parameters
None.
Example
Python
data = requests.get(f"{BASE_URL}/taker_intensity/get_cvd_table", headers=headers, timeout=30).json()["data"]buying = sorted((c for c in data["coins"] if c["net_24h"] is notNone), key=lambda c: -c["net_24h"])[:10]print(data["total"]["net_24h"], [(c["token"], c["net_24h"]) for c in buying])# Live capture 2026-09-22 (abridged with ...; anonymous studio twin, same inner shape):# {"coins": [{"token": "BTC", "token_id": 1, "buy_24h": 24607245298.03, "sell_24h": 23128502231.73,# "net_1h": -212647361.27, "net_4h": -207128163.51, "net_24h": 1478743066.31,# "by_exchange": {"binance": {"buy_24h": 12663681424.94, "sell_24h": 12135756153.44,# "net_1h": -191133129.94, "net_4h": -217337267.5, "net_24h": 527925271.5},# "gate": {...}, "okx": {...}}}, ...],# "exchanges": [{"key": "binance", "buy_24h": 34978603608.55, "sell_24h": 35112588381.38,# "net_24h": -133984772.74, "last_at": "2026-09-22T05:25:00+00:00", "stale": false}, ... 3 rows],# "total": {"buy_24h": 63543431942.93, "sell_24h": 62491177680.21, "net_24h": 1052254262.8},# "tokens_shown": 30, "tokens_total": 591, "full": false, "updated_at": "2026-09-22T05:25:00+00:00"}
Collection start (first stored bucket, measured on BTC 2026-09-22)
Exchange
Source field
Collecting since
binance
5-minute kline (taker-buy quote volume)
2020-01-01
gate
contract_stats
2026-08-12
okx
taker_volume_usd
2026-09-21
OKX is additionally capped by the venue keeping only 5 days, so it stays full_7d: false until a week after that date. This table carries no since / full_7d (it has no 7-day window) — read them per coin from /taker_intensity/get_cvd_coin's exchanges[], not from these dates.
Taker buy / sell turnover over the rolling 24 h, summed over the coin's non-stale exchanges; null when all of them are stale
coins[].net_1h / .net_4h / .net_24h
float / null
USD
buy − sell over each rolling window (same null rule)
coins[].by_exchange
object
—
{exchange: {buy_24h, sell_24h, net_1h, net_4h, net_24h}}, one key per exchange; null when the coin is not listed there or that exchange's data for it is stale
exchanges[]
object[]
—
binance, okx, gate — whole-market buy_24h, sell_24h, net_24h, last_at, stale
exchanges[].stale
bool
—
That feed's last bar is over an hour old: its windows come back null and it is left out of the totals rather than contributing a sum that silently covers less than the window
total
object
USD
Whole-market buy_24h, sell_24h, net_24h over the fresh exchanges
tokens_shown / tokens_total / full / updated_at
—
—
As in the other raw tables
Errors
Status
Body
When
503
{"error": "cvd data unavailable"}
The job has no fresh result — never zeros
Notes
Not the same as the taker_intensity alpha (/taker_intensity/get_alpha), which is a normalised z-score-like signal with years of history. This is raw USD turnover, now.
Windows are rolling and end at the last closed 5-minute bar, so net_24h is not a calendar day.
The full roster: key, listed, windows (same four, per exchange), since, full_7d, last_at, stale
exchanges[].full_7d
bool
—
false = under 7 days of history for this coin there: its 7 d window is null and it is excluded from the 7 d total and from the series. OKX keeps only 5 days, so it is false until a week after the feed's start
series.timestamp
int[]
epoch s
168 hourly slots, oldest → newest
series.net
float[]
USD
Net taker flow per clock hour, summed over series.exchanges; null for an hour with no data. When series.exchanges is empty every slot is null and cvd stays flat at 0 — the line carries no information then
series.cvd
float[]
USD
Running total of net; cvd[0] = 0 — the line is a shape, not an absolute level
400 / 404 / 503 exactly as /long_short_ratio/get_coin, with {"error": "cvd data unavailable"} on 503.
Notes
series.exchanges being shorter than exchanges[] is the full_7d rule — compare the series against itself, not against windows["24h"], which counts more venues.
cvd[0] = 0 by construction: only the slope and the turning points carry information.
Latest report only (history: /cme_cot/get_history)
Update
Weekly — positions as of Tuesday (Monday when Tuesday is a US holiday), published by CFTC on that week's Friday 15:30 US Eastern; a federal holiday Wed–Fri moves it to the next business day, a government shutdown delays it by weeks; server cache 1 h
Source
U.S. CFTC Commitments of Traders, futures only — Traders in Financial Futures (TFF) + Legacy reports
The latest weekly CFTC row for each of the four CME crypto futures, plus standard + micro summed in coins per asset. Same shape as one row of /cme_cot/get_history, without price.
One per contract: key (btc, micro_btc, eth, micro_eth), name, cftc_code, coin, coin_per_contract (5 / 0.1 / 50 / 0.1), latest (a row, see get_history; null if the contract has no rows)
combined[]
btc_combined, eth_combined: name, coin, components, units, micro_since, note, latest (a combined row, in coins)
report
futures_only
units
Unit of contracts[].latest positions: contracts (× coin_per_contract for coins)
source
CFTC attribution line — show it with any figure you publish
source_updated
When CFTC last updated the dataset
fetched_at
When Blave fetched it (UTC)
stale
true = CFTC was unreachable and the last stored copy is served
Errors
Status
Body
When
503
{"error": "CFTC data unavailable. Please retry shortly."}
btc 2018-04-10, micro_btc 2021-05-04, eth 2021-04-06, micro_eth 2021-12-14 (CME's own BTC futures enter the COT on 2018-04-10, not at their 2017-12 launch)
Update
Weekly — positions as of Tuesday (Monday when Tuesday is a US holiday), published by CFTC on that week's Friday 15:30 US Eastern; a federal holiday Wed–Fri moves it to the next business day, a government shutdown delays it by weeks; server cache 1 h
Source
U.S. CFTC Commitments of Traders, futures only — TFF + Legacy reports; price = Binance spot UTC daily close
One contract's weekly positioning by trader category. Rows are dated by the report date — the day the positions are as of, usually a Tuesday (a Monday when Tuesday is a US holiday). The data became public later: normally that week's Friday 15:30 ET; a federal holiday Wednesday–Friday moves the release to the next business day, and a government shutdown delays it by weeks (the 2018-12-24 … 2019-02-26 reports came out 2019-02-01 … 03-05; the 2025-09-30 … 2026-01-20 reports 2025-11-19 … 2026-01-23 — CFTC Releases 7864-19 and 9138-25). A backtest must not act on a row before its release.
Parameters
contractstringquery
Single contracts are in contracts; *_combined = standard + micro summed in coins
contract: key, name, coin, and cftc_code + coin_per_contract (single contract) or components + micro_since + note (combined).
Each row:
Field
Type
Description
date
string
Report date — usually a Tuesday, a Monday when Tuesday is a US holiday
open_interest / open_interest_change
number
Total open interest and its week-over-week change
tff
object
Groups dealer, asset_manager, leveraged_funds, other_reportables, nonreportable
legacy
object | null
Groups non_commercial, commercial, nonreportable; null if CFTC has no Legacy row for that week
price
float | null
Binance spot BTCUSDT / ETHUSDT UTC daily close on the report date; null when unavailable
included
string[]
Combined contracts only: which contracts this row sums (["btc"] before micro_since)
Each group: long, short, spread (not for nonreportable / commercial), net (= long − short), and long_change, short_change, spread_change, net_change. Units: contracts for a single contract, coins for a combined one. The *_change fields are CFTC's own week-over-week figures, not row-to-row differences — CFTC skips weeks (e.g. government shutdowns), so never diff rows yourself.
Errors
Status
Body
When
400
{"error": "contract must be one of btc, micro_btc, eth, micro_eth, btc_combined, eth_combined"}
Unknown contract
400
{"error": "Invalid start_date, expected YYYY-MM-DD"} (or end_date) / {"error": "start_date must not be after end_date"}
Bad date / reversed range
503
{"error": "CFTC data unavailable. Please retry shortly."}
CFTC unreachable and no stored copy
Notes
*_combined has a break at micro_since: the week micro joins, the totals step up and that week's *_change includes micro's whole first position. Start a combined backtest after micro_since, or use the single btc / eth contract.
Full BTC history is ~440 weekly rows (~0.5 MB) — one request, no paging.
Licence: CFTC data is public domain, but CFTC asks to be acknowledged — keep source next to any figure you show. Not endorsed by the CFTC.
Let Your AI Do Quanton Your Own Computer
Blave Desktop is free, open source and needs no sign-up. Connect Claude Code or Codex: it writes the strategy, runs the backtest and trades it live.
Downloading.Drag it into Applications,open it, and pick your AI.The first launch sets up its runtimeand takes a few minutes.If you like it, star it on GitHub.
Downloading. When you open the installer, Windows first shows “Windows protected your PC”:
Click “More info”.
Click “Run anyway”. Blave installs and opens on its own.
The first launch sets up its runtime and takes a few minutes.
Windows warns because the installer isn't code-signed yet (we've applied for a certificate), so it can't identify the publisher. The source is public on GitHub. See the full steps
