Using dp_exchange_webull

EXPERIMENTAL. Not run in production. Pin three-part. Maturity is per endpoint — read capabilities/0, not this banner.

Everything general is in dp_exchange_core's usage rules. This file is only what is specific to Webull.

BREAKING — a refusal now carries the venue's status as well as its words

# before
{:refused, {:venue_error, "signature mismatch"}}

# after
{:refused, {:venue_error, 401, "signature mismatch"}}

This package refuses on 400, 401 and 403, and all three used to look identical to a caller even though their remedies are opposite:

An unrecognised body now keeps its status too — {:venue_error, 403} rather than a bare :refused. Thin, but it is the difference between "the venue rejected this and here is which kind" and "something went wrong".

This is the shape dp_exchange_coinbase, dp_exchange_robinhood and dp_exchange_schwab already used; this venue was the outlier.

If you pattern-match refusal reasons, check that clause before upgrading. A caller matching {:venue_error, message} will now fall through to whatever catch-all follows it — silently, with nothing raising. That exact trap cost a consumer a canonical error mapping on dp_exchange_gemini and is worth ten seconds of grep here.

BREAKING — Quote and OrderBook no longer carry :timestamp

They carry :venue_time (the venue's own, nil where the venue publishes none) and :observed_at (when this package read it, always present) — the shape Core.Types.TopOfBook has always had. Requires dp_exchange_core ~> 0.2.1.

# before
quote.timestamp

# after
quote.venue_time  # may be nil — the venue did not date this
quote.observed_at # always present

Why it had to break. :timestamp was documented as the venue's own, "never invented", and two packages in this family could not keep that promise: the frames they decode carry no venue time at all. With one field their only options were to lie or to drop real data, and they lied. Now they can say nil and mean it.

What to do with nil. Whatever you would have done with a wrong answer, but knowingly. The consumer who decided this design stores venue_time as their time-series point time where it is present, and where it is nil stores observed_at and records that they did — so a mis-bucketed value is attributable rather than invisible. That decision was not expressible before, because there was no way to see which kind of time you had.

Trade, Fill, Balance and OrderBookDelta are unchanged — they keep a single :timestamp, because every one of them is built from a venue-supplied time and fails closed without it.

When is venue_time nil on this venue? Whenever the venue did not state one — and until 0.4.37 this section said it never is. It read: "Every Quote and OrderBook this package builds parses a time the venue sent … and fails closed when it cannot. A nil branch for this venue is dead code."

That was true and stopped being true in 0.4.36 and 0.4.37, when the REST quote, the REST order book, and the MQTT quote and book stopped discarding real prices and real levels over a time the venue had not stated. Core.Types.Quote and Core.Types.OrderBook do not enforce venue_time, and observed_at is what states freshness. Write the nil branch.

Nothing is ever substituted into that field: it is the venue's own instant or it is nil, never this package's clock. Other venues in the family return nil too, which is why the field is nullable.

Full reasoning and the options that were weighed: dp_exchange_core issue #31.

Credentials are required for market data

There is no anonymous endpoint on this venue. Every OpenAPI call is signed, including the ones that look public, so get_price/2 needs credentials that the same call on Coinbase or Gemini does not:

{:ok, quote} = DpExchange.Webull.get_price("BTC-USD", credentials: %{
  app_key: "…", app_secret: "…"
})

capabilities/0 declares credential_benefit: :required — the first venue in the family to, and not the only one (dp_exchange_robinhood and dp_exchange_schwab declare it too). Branch on that rather than assuming market data is free; the alternative is finding out from a 401.

You keep the credentials. This package signs one request with them and holds nothing.

It is not only market data, and the fake enforces all of it

Every account and order call — get_accounts/2, get_balances/2, get_transfers/2, get_transactions/2, get_positions/1, quantization/2, place_order/3, place_orders/3, preview_order/3, replace_order/4, cancel_order/3, get_order/3, get_orders/2 — and every options, watchlist, fundamentals, news and screener call — get_option_chain/2, get_option_expirations/2, list_watchlists/1, get_watchlist/2, create_watchlist/3, update_watchlist/2, delete_watchlist/2, get_financials/3, get_corporate_events/1, get_filings/2, get_news/1, get_screener/2 — refuses the same way market data does without a credential: {:error, {:missing_credentials, :webull}}, never {:ok, _}. Not {:refused, _} either — a missing local credential never reaches the venue at all (Core.Venue's :refused is the venue's own permanent word about a request it received), so this is Auth.headers/2's own return value, not an invented one.

A blank credential counts as a missing one. An :app_key or :app_secret that is "", or only whitespace, is refused locally with {:error, {:missing_credentials, :webull}} — it is never signed with. This matters because the usual way a credential goes missing is not a nil: it is a .env line reading NAME= with nothing after it, and System.get_env/1 hands that back as "". Signing with it produced a well-formed request the venue refused for a reason naming signatures, which points at the signing code rather than at the credential.

DpExchange.Webull.Fake enforces this too, and it did not before. Before this fix, Fake.get_accounts/2, Fake.get_balances/2, Fake.get_transfers/2, Fake.get_transactions/2, Fake.place_order/3, Fake.place_orders/3, Fake.preview_order/3, Fake.replace_order/4, Fake.cancel_order/3, Fake.get_order/3, Fake.get_orders/2, Fake.get_positions/1 and Fake.quantization/2 answered {:ok, _} regardless of whether credentials were supplied at all — several ignored the argument outright, and two checked only opts[:account_id], a different question. A consuming test suite that called any of these without credentials and asserted success was passing against behaviour the real venue does not have. If your own tests do this, they need updating; that is the point of the fix, not a regression in it.

market_status/1 and get_fees/2 are the exceptions, and stay exceptions. market_status/1 answers {:error, :not_supported} regardless of credentials — this venue publishes no reachable market-status endpoint, so there is nothing to gate. get_fees/2 answers {:ok, _} with no credential at all: it builds no request and returns a crypto spread rate captured from Webull's own published pricing (source: :published_rate), so there is nothing here for a credential to gate either. A 2026-09-06 sweep gated get_fees/2 behind a credential anyway, reasoning that its real path "had never run through Auth.headers/2" — true, and the reason there is nothing to gate, not a reason to add a check. That broke a real consumer resolving fees before any account is attached, and was reverted 2026-09-07. If your code passes no credential to get_fees/2, that is correct usage, not a workaround.

The same gap survived one round longer on the options, watchlist, fundamentals, news and screener callbacks, and is now closed as well: Fake.get_option_chain/2, Fake.get_option_expirations/2, Fake.list_watchlists/1, Fake.get_watchlist/2, Fake.create_watchlist/3, Fake.update_watchlist/2, Fake.delete_watchlist/2, Fake.get_financials/3, Fake.get_corporate_events/1, Fake.get_filings/2, Fake.get_news/1 and Fake.get_screener/2 all answered {:ok, _} with no credentials. Each call's own argument validation runs before the credential check, matching the real Rest order (it refuses before it builds a request) — so Fake.get_corporate_events() with neither a credential nor a :symbol is {:error, :symbol_required}, and an unknown screener or a non-statement kind is refused the same way the real path refuses it, credential or not. The fake calls Rest's own validators for these rather than copying the rules, so the two cannot drift: Fake.preview_order/3, Fake.get_financials/3, Fake.get_corporate_events/1 (which now also honours :kind), Fake.get_news/1 (which also takes one symbol as a string), Fake.get_screener/2 (week52_high_low requires :rank_type), Fake.update_watchlist/2, Fake.get_volume_profile/3, Fake.get_historical_prices/4 (a :limit over 1200 is refused), and the cash-activity calls (a range across two calendar years is refused, and get_transactions/2 honours :activity_types). A fake order placed under a caller-chosen client_order_id can now be read back and cancelled under it. An order placed with no client_order_id is issued a new one in the venue's format (32 lower-case hex characters, as Rest generates), and Fake.place_orders/3 rows carry the client_order_id that get_order/3 and cancel_order/3 take, with the system order_id on an accepted row and error_code and message on a refused one, as order-batch-place.md documents. A watchlist made with Fake.create_watchlist/3 gets its own id, is read back by get_watchlist/2 with the symbols it was created with, appears in list_watchlists/1 under its name (and its renames), and reads as no rows ({:error, :unexpected_response_shape}) once deleted; wl-1 is the canned one before you create any. These were missed by the earlier sweep because dp_exchange_core's assertion 17 checks a fixed list of nine callback names that predates this surface and does not include any of them — passing that assertion is not evidence that the rest of your fake gates credentials.

Start it, and it brings its own rate limiter

children = [{DpExchange.Webull, credentials: my_credentials()}]

Passing credentials at start is what lets the package replay your subscriptions after a reconnect — see below. Without them, a reconnect cannot re-subscribe.

You never need to pass :limiter yourself. The supervision tree above starts and names its own rate limiter, and every call this package makes on your behalf — the reconnect replay, the 60-second blind resubscribe, and subscribe/2/unsubscribe/2/update_symbols/2 called with no :limiter in opts — is wired to it automatically. If you do pass one explicitly, it wins.

A shard's socket crash costs one shard, never your whole subscription

Each shard's MQTT-over-WebSocket connection is a linked child of Feed — not a supervised sibling you can restart independently. Feed traps exits, so a shard's socket dying abnormally does not take Feed down with it: only that shard's symbols drop out of coverage/1 and coverage_by_kind/1 (cleared immediately, not left to report :stream for a connection that no longer exists), you get a :link_down Core.Notice naming the crashed shard, and this package reopens it on its own — you never need to call subscribe/2 again. The first crash reopens at once. A shard that crashes again within a minute of its last crash waits first, 1 s doubling to 60 s, so a socket that connects and dies cannot spin through the venue's five-connections-per-key budget. The same goes for a socket that reaches the broker's CONNACK and drops within 30 s: it counts as a failed attempt toward the reconnect backoff instead of resetting it.

Your app_secret and access_token will not appear in Feed's crash log. Feed keeps what you passed at start (or on a later subscribe/2) so it can replay your subscriptions after a reconnect — see above — and a crash of Feed logs its state via OTP's default crash report, which is where you would see it, because a crash report prints unredacted Logger metadata otherwise. The credential map is wrapped in a struct before it ever reaches state, so the crash line reads credentials: #DpExchange.Webull.Credentials<...> rather than the key pair itself. That now holds for the credentials you pass to subscribe/2, unsubscribe/2 and update_symbols/2 as well, which used to sit unwrapped in in-flight reconcile state and in the messages that reopen a shard. This does not extend to app_key on a live shard's socket — app_key is sent as a plaintext header (x-app-key) on every signed request this venue accepts by design, so it carries none of the confidentiality app_secret does, and a shard's own crash report still shows it.

A specific cause of that per-shard crash is now a clean reconnect instead — see dp-exchange-core issue #27. Webull sometimes closes a shard's socket with a WebSocket close frame that carries prose instead of an RFC 6455 status code (observed: "bye-bye!!!"). Every published version of websockex through 0.5.1 (the current Hex release) raises on that frame and kills the socket process outright — before this package's own reconnect logic runs at all — which is indistinguishable from any other crash from where you sit, except that it can repeat every few seconds under sustained conditions, holding a shard's coverage down well below its symbol count even though nothing alarms (each crash is caught and reopened, so supervision "works"). This package now vendors a fix ahead of upstream — see DpExchange.Webull.Vendor.WebSockex's own moduledoc if you want the mechanism — so this specific cause no longer crashes the shard: it is an ordinary :link_down / reconnect-and-resubscribe cycle, the same as any other disconnect reason. If you were seeing elevated shard-crash :link_down notices or a Webull coverage ceiling well below your symbol count before this fix, that is the most likely explanation; if it persists after upgrading, the venue is closing the connection for a different, still-open reason — see this package's own CHANGELOG entry for what was and was not established about why.

The same vendored fork gives the opening handshake a deadline (2026-09-27): socket_connect_timeout plus socket_recv_timeout. Upstream bounds each read of the upgrade response separately, so a peer that trickled the response held a shard's open, or its reconnect, indefinitely. That kept Feed waiting inside a call on open, and kept the shard delivering nothing on reconnect. A handshake that misses the deadline now fails with %WebSockex.ConnError{original: :timeout}, the error a plain read timeout gives, and takes the ordinary failed-open or reconnect-with-backoff path.

The WebSocket verifies the venue's TLS certificate (2026-09-27). websockex's own default is not to (verify: :verify_none), and until then this package connected with that default, accepting any certificate. It now uses the operating system's trust store and HTTPS hostname rules. On a host with no system CA bundle the connection fails with a TLS error rather than connecting unverified. Pass your own :ssl_options to override.

What still costs you your whole subscription: Feed itself crashing — a bug outside the per-shard crash path, or anything that kills the Feed pid directly. DpExchange.Webull.Supervisor restarts Feed under :one_for_one, but from the static opts your supervision tree started it with; every subscribe/2, update_symbols/2 and subscribe_notices/1 call you made afterward is gone. Nothing inside this package can replay those calls — it never held onto the functions or the process that made them. If your consumer needs to survive a Feed restart unattended, monitor the Feed pid (or the DpExchange.Webull pid it sits under) yourself and re-issue subscribe/2 on :DOWN.

Subscribing is two protocols, and you see neither

Market data arrives over MQTT on a WebSocket. Subscriptions are HTTP calls. They are joined by a session identifier this package generates and gives to both.

:ok = DpExchange.Webull.subscribe(["BTC-USD"], credentials: creds, to: self())

to: names who receives the stream and defaults to the caller. It must be a pid or a registered name (an atom); anything else is {:error, {:invalid_subscriber, to}} and nothing is subscribed. subscribe_notices/1 takes to: under the same rule.

The venue does not restore subscriptions after a reconnect. This package replays them for you, which is the whole reason you never have to notice a reconnect. That replay uses the credentials you supplied — at start, or on the subscribe call.

What one subscription delivers, and where its volume comes from

A subscribe asks the venue for its SNAPSHOT, QUOTE and TICK topics. Each is forwarded to you as its own type: %DpExchange.Core.Types.Quote{} (a traded price), %DpExchange.Core.Types.TopOfBook{} (bid/ask) and, whenever the venue sends a tick, %DpExchange.Core.Types.Trade{} (one print). Match on the struct, not on having subscribed once. A handler that only matches %Quote{} silently drops every other message rather than erroring.

receive do
  {:dp_exchange, :webull, %DpExchange.Core.Types.Quote{} = q} -> handle_price(q)
  {:dp_exchange, :webull, %DpExchange.Core.Types.TopOfBook{} = t} -> handle_book(t)
end

A streamed TopOfBook now carries bid_size and ask_size. Until 0.4.34 they were always nil, on the strength of a comment saying this venue published no depth at the top. Its own protobuf schema declares size on every level and the decoder had been reading it off the wire and discarding it. A nil there still means what Core.Types.TopOfBook says it means — "not published", never "none available" — so a level the venue sends without a size is nil, and one that states "0" is a zero.

Volume comes by two routes, and each says which quantity it is.

• The snapshot's volume arrives on every streamed Quote that states one, with volume_window: :running_total: the session's cumulative total. Difference consecutive values for an interval's volume, and treat a decrease as a reset. Never sum them.
• The tick arrives as a Trade whose quantity is one print: sum these. id is nil, because the tape carries no per-print identifier.

TICK is requested by default again (dp_exchange_webull issue #7). The streaming page lists it for "Stocks, Futures and Crypto". A consumer on ~325 us-crypto symbols measured none on 2026-09-29 (dp-exchange-core issue #40), from a request the venue accepted with TICK in it. So asking costs nothing, and every tick that does arrive is delivered. capabilities/0 declares streamable: [:quotes, :top_of_book]. :trades joins it once a run shows ticks arriving, because a declaration states what was measured. Until then a Trade that does arrive is delivered to you but not counted: coverage/1 and coverage_by_kind/1 never report :trades, because Core requires every reported kind to be a declared one and coverage/1 to be exactly their union. Do not read an empty :trades as "no ticks arrived"; a Trade on your mailbox is the evidence.

Measured on 2026-10-02, on 0.4.99: both routes are empty for crypto. A consumer subscribed with TICK and saw the request accepted and quotes flowing. Its first Webull quote carried no volume, so the snapshot leaves field 8 empty. About 6 minutes in, its 1-minute candles for BTC-USD and ETH-USD read volume_avg 0 and OBV 0, so no tick had arrived for either. Gemini's BTC-USD, built from Gemini's :trades, read real volume on the same consumer at the same moment. Webull crypto delivers no traded volume on its stream. The REST snapshot and the bars read a volume field whenever a response carries one, but their spec documents none for crypto. For volume on a pair Webull trades, use another venue's :trades for the same pair. That choice belongs to the consumer. This package never puts another venue's volume under Webull's name.

Coverage means delivering, not accepted

On this venue there are three different moments: you asked, the HTTP subscribe returned 200, and data is arriving. coverage/1 reports only the third. A 200 on the subscribe does not mean the stream is flowing.

coverage/1 folds all three kinds above into one :stream per symbol. coverage_by_kind/1 splits them apart — a symbol can show :quotes healthy while :top_of_book has gone dark for it, or the reverse, and coverage/1 alone cannot tell you which. Only the declared kinds (:quotes, :top_of_book) appear; see the note on :trades above:

DpExchange.Webull.coverage_by_kind(credentials: creds)
#=> %{
#=>   quotes: %{"BTC-USD" => :stream},
#=>   top_of_book: %{"BTC-USD" => :stream}
#=> }

The venue's connection budget shapes what you can ask for

This package shards across at most five connections — the venue's own ceiling — because a single MQTT session caps at 100 subscribed tickers. It never exposes sockets, so you cannot cause a sixth: five shards of 100 is 500 symbols, and a larger universe on this venue needs a second App Key rather than a bigger number here. The one-minute retention is why reconnect backoff matters: reconnecting immediately after hitting the limit fails until the venue ages the old sessions out.

If two instances ever shared a session id, the venue would disconnect whichever connected first — each instance looking healthy in isolation. The id is generated per shard, inside this package, so this cannot happen.

UAT has REST but no stream

{:ok, quote} = DpExchange.Webull.get_price("BTC-USD", environment: :uat, credentials: creds)

environment: :uat gives authenticated REST against test data. There is no UAT broker — the hostname does not resolve — so:

DpExchange.Webull.subscribe(["BTC-USD"], environment: :uat)
#=> {:error, {:streaming_unavailable, :uat}}

It refuses rather than falling back to production, because a consumer testing against UAT that received production prices would be reading real market data believing it was fake.

Ask first if you need to branch:

if DpExchange.Webull.streaming?(environment: :uat), do: …

Production and UAT can run side by side — supervisor, feed and limiter names all derive from the environment, so the two neither collide nor share a rate-limit bucket.

:production is the default and a typo raises. environment: :uatt is an ArgumentError, not a quiet fallback: meaning UAT and getting production sends a real order to a real broker.

live?/1 answers the money question directly, resolving opts the same way every call here does, so a caller can confirm before a money-moving call rather than trust a default:

if DpExchange.Webull.live?(opts) do
  # confirm with a human before place_order/3 goes out
end

There is no aggregate trade volume on crypto — but there is on stocks

Not on the bars, and not on the REST crypto snapshot's Quote.volume. On a crypto symbol from get_price/2 that field is nil, never 0, because zero would look like a real measurement of no trading.

The streamed snapshot is passed through, not assumed away. A Quote from the stream carries the Snapshot frame's own volume (field 8) whenever the frame states one, with volume_window: :running_total: the session's cumulative total, to be differenced, never summed. Treat a decrease as a reset. When a frame leaves the field empty, volume and volume_window are both nil. This package has not measured whether crypto frames fill it in. Read volume_window, not the venue, to know what a Quote.volume means: every quote that carries a volume now says which quantity it is.

The stock snapshot is different: get_price/2 with category: "US_STOCK" or "US_ETF" carries a real volume, the day's aggregate rather than the last trade's size. capabilities/0 says reports_trade_volume: true because that path is real — a single package-wide boolean cannot say "true for equities, false for crypto," and false would be an under-declaration of a venue capability this package already reaches. Branch on opts[:category], not on the boolean alone, if the distinction matters to you: nil on a crypto call, a real aggregate on a stock or ETF call. Bars are the same split, not a blanket "no volume": the stock, option, futures and event bar endpoints each document a required volume on their own bar row, and this package now carries it through; only the crypto bar endpoint documents none, and volume stays nil there — never 0, for the same reason as the crypto snapshot. get_volume_profile/3 remains the separate equity endpoint that splits traded volume by price and side.

That is a different claim from Trade.quantity — the streamed tape and get_trades/2 both report one print's own size, which the venue does publish. This package does not sum sizes into an aggregate figure of its own; that would be this package's arithmetic wearing the venue's name.

Eleven declared candle widths

1m 5m 15m 30m 1h 2h 4h 1d 1w 1M 1y — the full capabilities/0 historical_timeframes list, as of 2026-09-07. Two under-declarations were corrected to get here. Until 2026-09-06 this declared only the first eight, the crypto and event-contract default generalised to the whole venue, which was false for the equity, option and futures bars that have always served 1w and 1M too. 1y was then withheld for a real reason that stopped being true: dp_exchange_core 0.1.48's Timeframe.nameable/0 had no entry for it and Capabilities.new/1 raised on it. Core 0.1.57 names it, and the workaround outlived the gap by three releases — nothing fails when a documented workaround goes stale, which is exactly why a served width stayed hidden from consumers reading capabilities/0.

The crypto and event-contract bar endpoints serve 1m 5m 15m 30m 1h 2h 4h 1d and nothing wider. 1w is left off those endpoints deliberately: a weekly boundary depends on which weekday the venue starts its week, Core.Timeframe models no alignment rule for it, and a bar whose boundary cannot be verified is a bar that should not be stored. That reasoning is about a continuous, boundary-less crypto week — it does not apply to equities, which trade on a fixed calendar week the venue's own bar aligns to.

The equity, option and futures bar endpoints take three widths beyond those eight — 1w, 1M and 1y — because that is the timespan vocabulary those endpoints publish (Rest.get_stock_bars/5, tested against it). Core.Capabilities has one flat list for the whole package with no per-asset-class shape, so the declaration is the union every active path reaches rather than the crypto subset, and it is now derived from Rest.wide_timeframes/0 rather than a hand-maintained subtraction from it. Asking a crypto or event-contract category for 1w, 1M or 1y is still an error, never the nearest width it does serve — branch on opts[:category], not on the flat list alone, if you need to know which endpoint a width actually reaches.

Equity/ETF bars are adjusted at daily and above, not below

get_historical_prices/4 against US_STOCK/US_ETF: daily and longer widths are forward-adjusted for splits, minute widths are not — the vendor's own rule, and these are two different series, not the same one at two resolutions. adjusted?/1 answers it for a width before you fetch it, so stitching a daily series onto a minute one across a split does not join an adjusted half to an unadjusted one silently:

DpExchange.Webull.adjusted?("1d")  #=> true
DpExchange.Webull.adjusted?("1m")  #=> false

get_orders/2 follows pagination_key to the end, bounded

/trading/orders/open-orders/list and /trading/orders/historical-orders/list both document pagination_key — the venue's own reference pages, not just its endpoint inventory — and this package walks it the same way get_symbols/2 walks the instrument catalogue's: to the end, or {:error, {:too_many_pages, n}} on a bound, or {:error, :pagination_key_did_not_advance} on a repeated key. get_orders/2 returns the whole set the venue holds for the request, not a prefix.

opts[:limit] is no longer read. Neither endpoint documents a page_size parameter — sending one asked the venue a field it does not define, and did not actually bound the walk (there was no walk). history: true accepts opts[:since]/opts[:until] (DateTimes) instead, sent as the documented start_time/end_time; omitted, the venue answers the last 7 days, its own default.

get_symbols/1 lists everything; list_instruments/1 says which of it you can trade

get_symbols/1 is Core's contract, unfiltered: every symbol this venue lists, tradable or not. Some of what it lists is "NT" (Non-Tradable) or "CO" (Liquidate only — only a CLOSING order matches) on the venue's own instrument catalogue, and get_symbols/1 has no way to say so — this is exactly why a subscribe to one of those symbols can come back INVALID_SYMBOL even though get_symbols/1 just handed it to you.

list_instruments/1 reads the same catalogue and answers Core.Instruments instead of bare strings, each with a status: :tradable for "OC", and :unknown for "CO", "NT", and anything absent or unrecognised. Filter on status == :tradable before you build an order book or a strategy universe from a symbol you only got from get_symbols/1:

{:ok, instruments} = DpExchange.Webull.list_instruments(credentials: creds)
tradable = for i <- instruments, i.status == :tradable, do: i.symbol

:unknown is not the same claim as :delisted — a "NT" row is still listed, just not currently tradable, and this package does not say the venue removed it. list_instruments/1 only serves US_CRYPTO; US_STOCK/US_ETF refuse by name ({:error, {:unsupported_instrument_category, category}}) rather than being served with an invented instrument type or quote currency — see DpExchange.Webull.Rest.list_instruments/2.

:tradable does not promise a quote: {:refused, :no_quote}

The catalogue and the quote endpoint can disagree. On 2026-10-02, 51 of the ~325 USD pairs list_instruments/1 marked :tradable answered every get_price/2 with the venue's literal "null" price, and never ticked on the stream. Examples are MATIC-USD, FTM-USD, FTT-USD, KLAY-USD and MIOTA-USD, several of them delisted or renamed upstream (issue #8). get_price/2 returns {:refused, :no_quote} for these. That is permanent by contract, so don't retry it on a timer, and treat the pair as one with no market here. It used to come back as {:error, {:invalid_decimal, :price, "null"}}, which looked like a parser fault and, as an :error, invited endless retries. This package leaves the catalogue's "OC" alone, because that status is the venue's word and the refusal is its other word.

Many crypto quotes over REST: get_prices/2, twenty per request

The crypto snapshot endpoint takes "up to 20 symbols per request". get_price/2 asks for one, so polling many pairs with it spends the endpoint's whole 1/s budget at 60 symbols a minute. get_prices/2 sends them twenty at a time and answers a result for every symbol:

{:ok, results} = DpExchange.Webull.get_prices(symbols, credentials: creds)
for {symbol, {:ok, quote}} <- results, do: handle_price(symbol, quote)

Each result has get_price/2's own shape. A {:refused, :no_quote} is permanent (see above). {:error, :not_in_response} means the venue returned no row for that symbol. A chunk the venue rejects because of one symbol it calls invalid is halved and retried down to that symbol, so the rest of the chunk still gets its quotes. A chunk whose request failed gives each of its own symbols that error and no others. US_CRYPTO only.

Timestamps come from the venue, or the call fails

The local clock is never substituted — an undated bar stamped with your own clock is indistinguishable from a real one, which is how a gap becomes invisible. That guarantee is the whole of this section and has never changed.

What an undated value does depends on the TYPE, because the contract differs:

• A bar, a volume-profile interval or a trade is refused, with {:error, :missing_venue_timestamp}. Core.Types.Candle and VolumeProfile enforce :opened_at and Trade enforces :timestamp, so a bar at an invented minute or a print that cannot be placed in time is not something this package can report at all.
• A quote or an order book is delivered with venue_time: nil. Neither type enforces it, and refusing threw away a real guarded price, or a whole set of book levels, over an optional field. This changed in 0.4.36 and 0.4.37; before then both were refused.

Units. An epoch is read as seconds up to ten digits and milliseconds up to thirteen; one longer (microseconds, nanoseconds) is not a time this package can read and is refused ({:error, :invalid_unix_time} on REST, an undated frame on the stream) rather than becoming a DateTime tens of thousands of years out. The streaming schema states no unit or format for timestamp/time at all, so the same reading there is unmeasured.

An order's created_at is the venue's place_time_at (ISO8601 UTC, per order-detail.md), and updated_at is filled_time_at ("time of the last executed trade") only for a :filled or :partially_filled order; for any other status it is nil, since the order changed after its last fill and the venue states no time for that. Both are nil where the venue sent none or sent something that is not ISO8601. Unmeasured against a live order.

get_symbols/1, and the symbols on positions and orders, are split into BASE-QUOTE only for crypto. An equity, option, future or event row keeps the ticker or code the venue named (GBTC is not G-BTC).

A volume profile's price maps hold Decimals only. A level whose size is not a number refuses the profile ({:error, {:invalid_decimal, :buy_detail | :sell_detail, value}}) instead of storing nil where VolumeProfile.point_of_control/1 adds sizes.

An order book is sorted, so hd(bids) is the best bid

get_order_book/2 guarantees it as of 0.4.39 — bids descending, asks ascending — and it is Core.Types.OrderBook's contract rather than this package's convenience. You do not need to sort what you receive, and you should not assume the venue's row order means anything.

A level whose price this package cannot read is dropped rather than passed on with a nil price, because level/0 is {Decimal.t(), Decimal.t()} and hd(bids) must be able to answer with a real number. A level with a readable price and no readable size refuses the whole book - {:error, {:missing_required_field, :size}}, or {:error, {:invalid_decimal, :size, value}} for text that is not a number - because level/0 is a pair of Decimals, and dropping the level would publish a hole in the book as "no liquidity at that price". (Earlier releases passed {price, nil} through, a pair OrderBook.new/1 refuses.)

A row the venue did not identify is dropped, not given an empty id

As of 0.4.38, get_screener/2 and list_watchlists/1 drop a row that cannot supply its own identifying field rather than publishing it with "".

An empty string is worse than the nil it replaced: a nil is detectable and "" is a value, so a consumer keying coverage by symbol used to get a live entry named "".

get_news/1 no longer has "rows" to drop. It reads Server-Sent Events, one generated reply per call, not a JSON list of independently-published stories each naming their own id — see the news paragraph above. The old fallback on that JSON shape, before the endpoint was corrected, was an absent id falling back to the ticker symbol, which looks like an id and collides for every story about that symbol; that whole failure mode is gone with the JSON-list assumption it depended on. What replaces it: a stream with no conversation id in its meta event is {:error, {:missing_required_field, :id}} rather than a fabricated one, since Types.NewsItem.id is required.

So a shorter list is not an error, and the venue's own ranking is preserved. A dropped screener row leaves a gap in rank rather than renumbering the survivors, because rank is the position the venue returned the row in and closing the gap would re-rank the list.

An order list is the exception: it is refused, not thinned. get_orders/2 answers {:error, {:unreadable_orders, ids}} when any group in it cannot be read as one Order, most often a multi-leg combo (OCO, OTO, bracket), which this package does not represent. ids are the groups' client_order_ids where the venue gave one. Dropping them would hand you a list of working orders that is missing some, which reads as "nothing working there"; place against that and you may duplicate an order.

A :degraded notice carries its severity from the frame — read severity, not only kind

The venue publishes telemetry on its notice topic, and most of it says nothing is wrong. A keepalive reporting drop: 0 arrives as severity: :info. A notice is :warning only when the venue sent words, or reported a non-zero drop — absence of an adverse indicator is not evidence of degradation.

Every frame used to be :warning, which on one live host meant 184 WARN an hour announcing that nothing had been dropped, crowding out 1,331 genuinely serious warnings underneath. If you map Notice.severity straight to a log level — the honest reading of a field whose own docs call it "not a log level — a call to action" — this is the distinction you want.

If you route on kind alone you will still see :degraded for both, and that is a known limit rather than an oversight: Core.Notice's kinds are a deliberately closed set with no slot for "the venue reported on itself and nothing is wrong", and widening a shared vocabulary is not one venue package's call.

A slow subscriber gets dropped, and told — it does not get an unbounded mailbox

If your process falls far enough behind that its mailbox reaches 10,000 queued messages, this package stops sending it events and emits a Core.Notice:

%Core.Notice{
  kind: :degraded,
  severity: :warning,
  message: "subscriber #PID<0.123.0> is 10000 messages behind, past the 10000 bound — ...",
  details: %{subscriber: "#PID<0.123.0>", queue_len: 10_000, bound: 10_000, dropping: :newest}
}

and a second one, severity: :info, once you have drained below half the bound (5,000 by default). Not merely back under it: a mailbox hovering at the bound used to flip on every message, with a notice each time, into the same full mailbox. Now one stall costs one pair. Those two notices bracket exactly the window you have to reconcile from the pull endpoints — that is what the pair is for, and why the recovery notice exists at all rather than just the alarm.

Three things are worth knowing about the shape of this:

• The newest is dropped, not the oldest. A sender cannot remove a message from your mailbox; only you can. So what happens is that nothing further is added while you are over the bound. It is also the better trade: a quote that arrives while you are ten thousand messages behind is stale by the time you would read it, and the frames it would have displaced are no fresher.
• Only you are affected. Another subscriber keeping up keeps receiving everything. One slow consumer is never allowed to become an outage for the others.
• coverage/1 does not change. It reports what the venue delivered to this package, not what this package forwarded to you. A symbol whose frames are being dropped for your backlog is still arriving, and reporting it as :not_covered would point you at the venue when the backlog is yours.

Notices themselves are never dropped, whatever your queue looks like — the notice telling you that you are being dropped must not be the first casualty of it.

Raise or lower the bound at start:

{DpExchange.Webull, max_queue_len: 50_000}

Any positive integer. It must be an integer — a string or a float raises at init/1 rather than quietly falling back to the default, because a back-pressure setting you believe you configured and did not is worse than not having configured one.

If you are seeing these notices, the fix is on your side. Consume in a process that does nothing else, or hand the payload straight to a queue or an ETS table and do the work elsewhere. The bound is not a tuning knob for throughput; it is the line past which this package stops writing into memory you are not reading.

Telemetry: attach one handler, see every venue

This package emits the :telemetry events DpExchange.Core.Telemetry documents. Attach to the names, not to anything venue-specific:

:telemetry.attach_many(
  "dp-exchange-metrics",
  DpExchange.Core.Telemetry.event_prefixes(),
  &MyApp.Metrics.handle/4,
  nil
)

:provider in the metadata is how you tell venues apart — one handler covers the whole family.

These names were documented long before anything emitted them. If you attached a handler before this version and saw nothing, that was not your venue being quiet: :telemetry.attach/4 against a name nobody emits succeeds, so an empty panel was indistinguishable from no traffic. Every one of them fires now.

Four things to know before you build on them:

• Telemetry is the metrics channel; Core.Notice is the action channel. Telemetry is aggregate and lossy by design. Alarm on notices. Graph telemetry. Alarming on a gauge documented as droppable is how a missed sample becomes a page at 3am.
• :stop fires for every outcome, success or error. That is deliberate: a latency panel built on successes alone shows a venue getting faster exactly as it starts failing, because the slow calls are the ones dropping out of the sample. :exception is a separate event from a :stop carrying an error — an error result is the venue answering badly, an exception is this package failing to ask.
• retry_after_ms is always milliseconds, including where the venue's own header is in seconds. A panel summing a mixture of the two is wrong by a factor of a thousand without ever looking wrong.
• :bytes may be absent from [:dp_exchange, :link, :event], and absent is not zero. A polling route has no frame, so there is no byte count that means what a socket's does. Treat a missing :bytes as "this route cannot measure that", not as no data.

:endpoint metadata is the request URL with the query string removed. Deliberate: telemetry metadata reaches logs, aggregators and third-party exporters, and the query string is the one part of a URL that can carry a token.

Every capability claim is dated, and says what it was measured against

capabilities/0 carries two provenance fields, and they are there to be read:

caps = DpExchange.Webull.capabilities()
caps.measured_at       # ~D[...] — when someone last checked
caps.measured_against  # what they checked: which documents, which endpoints, live or not

measured_against is the one that matters when you are deciding how much to trust a claim. It distinguishes a figure measured live against the venue's API from one read off a documentation page and never probed — and this package's own declarations contain both kinds, stated per claim rather than as one blanket sentence.

A claim read from documentation is not worse, it is different: it is what the vendor says it will do, which is the right basis for a ceiling you must not exceed and a weaker basis for behaviour you are about to depend on. If your decision turns on the difference, read the field rather than assuming.

measured_at going stale is a real thing that happens — venues change, and a declaration is a claim about a real venue. script/check_doc_sources.sh reports the age on its weekly run so it does not quietly get old, but nothing forces it: if you are relying on a capability in a way you cannot afford to have wrong, check the date and, where it matters, verify against the venue yourself.

What this package does not do yet

Read capabilities/0, not this paragraph. As of 2026-09-01 the order path, balances, accounts, transfers and order book are all implemented, along with options, futures, event contracts, fundamentals, screeners, news and watchlists. get_trade_history/2 and get_market_overview/1 are not — both still answer {:error, :not_supported}, as not-yet-ported rather than as an absence at the venue. What remains :unsupported is listed there and per endpoint.

The money-movement callbacks are the ones that will not arrive: Webull's published API moves no money. There is no payment-method endpoint at either scope, no bank registration, no crypto network list, no allowlist and no transfer between accounts — funding happens in Webull's own applications, which need a person.

Five markets, five sets of endpoints, five parameter sets

opts[:category] routes: US_CRYPTO, US_STOCK/US_ETF, US_OPTION, US_FUTURES, US_EVENT. They are not one endpoint with a filter — each is its own path with its own parameters, and this package sends only the ones that endpoint's page documents.

That matters because the sets differ in ways that look arbitrary and are not: futures bars take no real_time_required and no time range; the futures tape and depth take no session filter; the option snapshot takes no extended-hours block. A parameter an endpoint does not know is at best ignored and at worst a refusal, and neither tells you which happened.

The default depends on which endpoint you called. get_price/2, get_top_of_book/2, get_historical_prices/4 and get_symbols/1 default to US_CRYPTO, which is what this package served before it widened. get_order_book/2, get_trades/2, get_volume_profile/3 and get_fundamental/3 default to US_STOCK, because this venue publishes no crypto depth, tape or footprint endpoint for them to have defaulted to. Changing either default would silently re-route existing callers onto a different market.

Event contracts have two prices and four books

get_trades/2 and get_order_book/2 refuse US_EVENT, and that is deliberate. An event tick carries a yes_price and a no_price and a side of yes/no; Types.Trade has one price and a side of :buy/:sell. An event book returns four arrays; Types.OrderBook has two sides.

Use get_event_trades/2 and get_event_order_book/2, which return the venue's own rows. The nearest mapping would produce a number that looks right and belongs to the other instrument of a two-instrument market.

The venue notes that in a binary market a yes bid at X equals a no ask at 1−X. That identity is the venue's; this package does not derive one side from the other, because a derived level cannot be told from a quoted one.

Event bars do not say which side they are. The venue's schema names open, close, high, low and no side. Reconcile against get_event_trades/2, which does name both.

Options: a chain is rebuilt, and greeks do not exist here

get_option_chain/2 turns the venue's flat contract list into expiry × strike, both sides. A contract this package cannot address — no readable expiry, strike or right — is refused, naming the keys the venue actually sent. A dropped row leaves a chain with a hole in it that looks complete.

A strike listed with one side keeps a nil on the other. :underlying_price is nil: this endpoint lists contracts and does not quote the underlying, and pairing it with a price fetched separately is two observations at two times.

get_option_greeks/2 returns {:error, :not_supported} and it is the venue, not the package. No Webull endpoint publishes them, and computing them would need a rate and a volatility surface it does not publish either.

Fundamentals: twenty-three endpoints, one shape

get_fundamental/3 reaches any of them; fundamental_kinds/0 lists them. Every one takes symbol and category; :type and :count reach only the endpoints that document them and are dropped elsewhere.

get_financials/3 takes the contract's :balance_sheet, :income, :cash_flow or :indicators. A fundamentals kind that is real but is not a statement is refused — :company_profile exists and answering get_financials/3 with it would put a profile in a statement's shape.

fiscal_period is the venue's code translated through the venue's own legend (0=FY, 1=Q1 …). The raw integer stays in line_items, and a code outside the legend leaves the label nil rather than inventing one.

:indicators answers one statement per fiscal period, oldest first. The venue sends one object whose metrics each list their own {fiscal_year, fiscal_period, value} readings; those are regrouped by period, with that period's metrics (the venue's names, string values) in line_items beside fiscal_year, the raw fiscal_period and currency. period_end is nil for them — the venue publishes no end date for an indicator.

A dividend's amount is a cash amount only for a CASH_DIVIDEND (or a row naming no div_type). Stock, scrip and the other types leave it nil; details holds the venue's row, div_type included.

get_corporate_events/1 needs :symbol — these calendars are per issuer, not market-wide — and without :kind reads both calendars, which is two requests.

get_news/1 is generated, not reported — and answers one item, not a list of stories. The vendor's own description is "invokes LLM to generate news summaries", and the endpoint itself answers Server-Sent Events (event:message\ndata:{"type":"text","message":...} chunks), not a JSON list — this package concatenates the chunks into one summary and returns it as a single-element list, :id taken from the stream's own conversation id (convId, falling back to sessionId). source names the venue rather than a wire; if you quote summary, you are quoting a paraphrase, not the publisher's text. Before this was corrected, every call returned {:error, {:undecodable_response, :webull}} — the generic JSON decoder this package's other POST calls share cannot parse an SSE body, so the endpoint never actually worked.

A screener's rank is the position the venue returned the row in. Nothing is re-ranked: two venues' "top movers" answer different questions, and so do one venue's under two sorts.

Watchlists: three absences and a boolean

symbols is nil on a list_watchlists/1 row — that endpoint names watchlists and does not read membership, and [] would say the watchlist is empty. name is nil on get_watchlist/2, because the membership endpoint does not return it.

update_watchlist/2 refuses opts[:symbols]. This venue's update endpoint touches properties only; add_watchlist_instruments/3 and remove_watchlist_instruments/3 are the membership writes, and silently skipping the option would leave you believing the list changed.

Every watchlist write answers {"success": …} rather than an error status, so a false is a 200 that did nothing — reported here as {:refused, :watchlist_write_rejected}.

Creating with members is two requests. Where the add fails, the watchlist exists and is empty: {:error, {:watchlist_created_without_members, id, reason}} carries its id so you can deal with it.

A stop order needs :stop_price; a trailing stop needs :trailing_stop_step

place_order/3's request map takes the same field names for every instrument type, and which ones matter depends on :order_type:

A field a given order type does not use is dropped rather than sent — a :stop request built from a :limit template (with :price left over) will not carry a stray limit_price the venue's schema for STOP_LOSS does not have. Read the fields back the same way: get_order/3 and the struct place_order/3 hands back both carry :stop_price on a stop or stop-limit order, nil on anything else.

Which pairs of :order_type and :time_in_force this venue actually accepts differs by instrument — crypto's list is five pairs (MARKET/IOC, LIMIT and STOP_LOSS_LIMIT at DAY or GTC — no plain STOP_LOSS, no TRAILING_STOP_LOSS); equity, option and futures orders add STOP_LOSS and (equity only) TRAILING_STOP_LOSS; an event contract takes LIMIT only, at any of five time-in-force values. A pair outside the matrix is refused before the request is sent, naming both halves of what was wrong, rather than being sent and rejected by the venue.

Batch orders: fifty, equities, and one request

place_orders/3 is not place_order/3 in a loop — the venue accepts the batch as one request. Both of its limits are enforced before the request: 50 orders and equities only. A batch over the cap is refused rather than split, because splitting turns one atomic request into several.

The result is per order. The venue validates each and returns each; a batch where three of five were accepted is the normal shape. The vendor also notes the endpoint is not available to every client, so a refusal can mean the account is not entitled.

The batch endpoint's own matrix is narrower than place_order/3's, and it is checked against, not discovered. order_type is MARKET or LIMIT only — no STOP_LOSS, STOP_LOSS_LIMIT or TRAILING_STOP_LOSS, all real for a single order. time_in_force is DAY, the one member of this endpoint's own enum; a caller who asks for anything else is refused rather than silently switched to DAY. Sizing is QTY only — :amount is refused here, where place_order/3 accepts it for equities. And :support_trading_session (:core, :all or :night) is required per order, with no documented venue default, unlike on place_order/3 where it is optional.

An entry with no :symbol or :side, or a LIMIT entry with no :price, is refused by index before anything is sent: {:error, {:batch_order_rejected, index, {:missing_required_field, field}}}.

place_order/3 answers {:error, {:order_unconfirmed, client_order_id, :unexpected_response_shape}} when the venue accepts (2xx) but its row names no order. The order may be live: settle it with get_order/3 and the id in the tuple.

Tokens: a token that exists is not a token that works

create_token/1 returns one that is PENDING, and verification happens through an SMS code in the Webull app — which needs a person. check_token/2 is the only call that tells PENDING from EXPIRED from INVALID; all three fail identically at the next request and each has a different remedy. A response with no token, or a status outside the venue's four (PENDING, NORMAL, INVALID, EXPIRED), is refused — {:error, {:missing_required_field, field}} / {:error, {:unknown_token_status, status}} — and oauth_token/3 refuses a 2xx with no access_token. NORMAL is never inferred.

oauth_token/3 is one endpoint doing two jobs on a different host, with a form body. opts[:code] exchanges, opts[:refresh_token] refreshes, and exactly one is required. Two expiries come back and they are not the same clock — rt_expires_in is the one that ends the session.

Rows this package will not flatten, and rows that are not yours

An answer is for the symbol you asked, not the first row. get_price/3, get_top_of_book/3, quantization/2 and get_event_order_book/3 answer {:error, :not_in_response} when the venue's rows name other symbols than the one asked for.

get_positions/1 refuses what a Position cannot say: a multi-leg option strategy and any event-contract row answer {:error, {:unsupported_position, symbol, reason}} for the whole list, rather than a position that drops the legs or the yes/no outcome. A SINGLE option row's symbol is the venue's, unverified as the contract symbol.

get_auction_imbalance/2 labels each row with the auction the row names (PRE_OPEN is :opening, PRE_CLOSE is :closing), which outside the publication windows can differ from the one you asked for.

get_option_chain/2 keys strikes by value: "150.0" and "150" are one cell at Decimal.new("150").

get_watchlist/2 takes its id from the venue's reply and refuses a reply that names none.

A write is sent once; a read is retried

Every request is signed afresh for each attempt, so a retry after a timeout or a 5xx is a real second request rather than a replay the venue refuses. That makes a retried write able to take effect twice, and the venue documents no idempotency key that would deduplicate a repeated order. So place_order, place_orders, replace_order, cancel_order, create_token and every watchlist write are sent once. A timeout on one of them means "unknown": read the order or watchlist back before sending it again. Pass :retry_attempts yourself only if you have your own reason to believe a repeat is harmless.

An order's numbers and ids are checked before anything is signed

quantity, amount, and the price fields the order type sends (price, stop_price, trailing_stop_step) must be a finite number above zero, in any of Decimal, integer, float or numeric text: otherwise {:error, {:invalid_number, field, value}}, by index inside a batch. They go out as the venue's strings in full notation ("0.00000001", never "1.0e-8" and never a JSON number). replace_order/4 holds its changes to the same rule, treats a nil value as no change ({:error, :no_order_changes} when nothing is left), and refuses an :order_type or :time_in_force it has no wire name for ({:unsupported_order_type, v} / {:unsupported_time_in_force, v}) instead of dropping it. cancel_order/3, get_order/3 and replace_order/4 refuse a missing or empty client_order_id as {:error, {:invalid_client_order_id, id}}. get_order_book/3 on US_FUTURES takes depth 1 to 10 ({:depth_out_of_range, n, max: 10}), as the venue's page states. The fake refuses all of these identically.

An order's expire_date and event_outcome are checked before a request, in place_order/3 and preview_order/3 (and the fake). A GTD expire_date is a Date or one real yyyy-MM-dd string ({:error, {:invalid_expire_date, v}} otherwise; a missing one is still left missing); event_outcome is :yes, :no, "yes" or "no" ({:error, {:invalid_event_outcome, v}}). Time bounds (:start/:end on get_transfers/2 and get_transactions/2, :since/:until on a history: true get_orders/2) are DateTimes or absent: anything else is {:error, {:invalid_time, key, value}}. A get_positions/2 side is read from the sign of quantity (negative is :short), which the vendor page does not document and which has not been measured against a live short position.

opts[:account_id] must be non-empty text on every account and order call; nil, a number, false and "" are {:error, :account_id_required}, here and in the fake (place_orders/3 included). :activity_types on get_transfers/2 and get_transactions/2 is a non-empty list of text or the venue's own comma-joined string; anything else is {:error, {:invalid_activity_types, value}}.

An Order that place_order/3 hands back is in the contract's vocabulary, not in your spelling of the request: side is :buy or :sell (nil for SHORT, which Order has no word for, as a read-back of one is), and quantity, price and stop_price are Decimals whatever you sent (an integer, a numeric string). The fake builds its orders with the same function. A replace the venue accepted whose read-back then fails is {:error, {:replace_accepted_unread, client_order_id, reason}}: the amendment took effect, so read the order with get_order/3 rather than sending the replace again.

get_positions/1 refuses a row with no readable quantity ({:error, {:missing_required_field, :quantity}}, or {:invalid_decimal, :quantity, value}) for the whole list: the venue's page lists it as required and Position has no nil quantity or side.

The fake narrows what the venue narrows. Fake.get_option_chain/2 and Fake.get_option_expirations/2 honour :expiry and :strike; Fake.get_price/2 and Fake.get_top_of_book/2 refuse a snapshot category the venue has no endpoint for ({:unsupported_snapshot_category, c}) and Fake.get_top_of_book/2 refuses US_EVENT; Fake.get_historical_prices/4 checks the category, width, range and :limit of whichever of the five bars endpoints the category names; Fake.get_trades/2 checks the category and :limit; Fake.get_financials/3 refuses a :count over 20; and Fake.get_news/1 has no headline, as the real stream names none. A request those endpoints would accept for a symbol the fake does not model is {:refused, :not_listed}.

Error shapes that mean "do not act on this answer"

Returned by calls that previously answered {:ok, _} carrying a value you could not act on. A consumer matching only {:ok, _} needs no change; one that enumerates error reasons should know them.

{:error, {:undecodable_response, :webull}} — the venue answered 2xx with a body this package could not decode. The realistic cause is not malformed JSON from the venue; it is a 2xx that never reached the venue, such as a captive portal or a CDN maintenance page answering 200 text/html. Worth retrying: nothing about the request was wrong. This package used to substitute an empty object for such a body, which then decoded into a well-formed value with every field nil and was returned as success.

{:error, :unexpected_response_shape} from get_balances/2 — a balance row the venue did not attribute to an asset refuses the whole reply. An amount you cannot name an asset for cannot be sized, booked or reconciled against, and dropping the row silently would read as "you hold none of that asset", a different and more dangerous claim than "this response could not be read". Not retryable on its own.

A Balance's own balance field may still be nil, and that is a different statement: the venue named the asset and did not state a quantity for it. Read that as unknown, never as zero.

A subscribe that is still in progress answers before it times out

subscribe/2, unsubscribe/2 and update_symbols/2 wait for the venue's HTTP subscribe, or for a connecting shard's CONNACK. Either can be slower than the call's own 15-second timeout, and a call that times out EXITS the calling process. So after about ten seconds the call is answered anyway, with one of these:

• {:error, {:reconcile_pending, ms}}: the HTTP subscribe is still running.
• {:error, {:connack_timeout, ms}}: the shard has not finished connecting.

This holds wherever the call has got to, including after the package moved your symbols to another shard because the first was full or the venue refused one: the ten seconds run from when your call arrived, not from the latest internal retry.

A list that is not made of non-empty strings is refused at once with {:error, {:invalid_symbols, offenders}} (at most five named; the whole value if it was not a list), and nothing is subscribed or removed.

Neither needs a retry. Your symbols are already recorded. The subscribe finishes or the shard connects, and your symbols go out with it. coverage/1 tells you when they start arriving.

Every negative here is audited

docs/reference/webull/negative-claims.md lists each one with the source and date consulted. Three were wrong, all the same mistake: a true statement about the stock endpoint restated as a claim about the venue. It also records why the vendor's pages have to be rendered to be read — their parameter tables are built in JavaScript, and an inventory captured without them looks finished and cannot be implemented from.

priority_symbols: — keep the stream on the pairs you trade

The account holds the us-crypto stream on one session at a time. A consumer measured this on 0.4.110: each of three shards streamed one contiguous minute in three, never two at once. By default the permission rotates, so every pair streams about a third of the time.

To keep the pairs you actually trade on the stream continuously, name them at start:

{DpExchange.Webull, credentials: creds, priority_symbols: ~w(BTC-USD ETH-USD SOL-USD)}

Those symbols, up to one session's capacity (100), are planned into one shard. That shard keeps the stream: it is re-asserted the moment it goes quiet, and no other shard is ever re-asserted against it. The price is that the other shards do not rotate. With one holder, any of them streaming would mean the priority shard is not. Plan to cover those pairs from REST (get_prices/2, twenty per request). Priority symbols past one session's capacity are placed with the rest. A value that is not a list of strings fails at start.

The guarantee holds while shard 0 really is the priority shard. If none of the priority symbols is subscribed, shard 0 is an ordinary shard and every shard rotates as without the option. While the priority shard is disconnected, the others are repaired and rotated as usual, and the priority shard takes the stream back when it reconnects. Subscribing a priority symbol later re-plans shard 0 so the priority symbols come first. The ordinary symbols it held move to the other shards, and those re-subscribes take the stream briefly.

Shard 0's unused room is not wasted, and only overflow uses it. Ordinary symbols are planned into shards 1 to 4 first. Only when those are full does what would otherwise be refused as capacity_exceeded fill the room left in shard 0 after the priority symbols. So three priority symbols and 450 others fit, where they used to be refused below the 500-symbol ceiling. This costs the priority guarantee nothing, because the permission is per session: an ordinary symbol in shard 0 streams on the session that already holds it, and a priority symbol always takes its place first. Such a symbol therefore streams continuously while the total is high, unlike the symbols on the other shards, which fall back to REST.

A shard reopens its own socket only once it has stopped delivering

Each shard reasserts its symbols on a timer, but only once it has gone quiet: a shard where any of its symbols arrived within the last interval (60 s) is left alone. Each completed re-assert grabs the us-crypto permission from a sibling session, and on 2026-10-08 a consumer measured that costing a streaming shard 30 s or more of silence every minute (issue #10). The account cannot stream every shard at once (issue #11, measured across 0.4.107–0.4.109): re-asserting every quiet shard made two shards flap, each losing half its minutes, and never re-asserting while a sibling streamed left them with no stream at all. So the permission is rotated. While any shard streams, each tick re-asserts exactly one quiet shard, the one longest since it last streamed or last had a turn, and announces it with an :info :coverage_change naming the shard, its session_id and its symbols. A sibling may go quiet in turn, and that is the rotation, not a fault. When no shard streams at all, every quiet shard is re-asserted at once. Venue notices carry session_id in details. If one shard's blind resubscribe fails twelve times in a row and none of that shard's symbols has arrived for five minutes, this package reopens that shard's socket on a fresh session — the same recovery an explicit INVALID_SESSION triggers. A :link_down notice says so, naming the shard.

Both conditions are required, and the second one is the important half. A reconcile is a subscription-management call: it can keep failing while the transport underneath streams everything it is already subscribed to. Reopening in that state loses live data and gains nothing.

That is not a theory. Version 0.4.25 escalated on the failure count alone. On a live fleet three shards hit the limit, reopened, and failed the identical reconcile on the very next tick and every tick after — a fresh session changed nothing — while this venue's coverage fell from 226 distinct symbols to about 160, with two other streaming venues flat across the same windows. The stuck-but-subscribed sockets had been delivering; tearing them down is what stopped it.

So while a shard is still delivering, the failures are counted and reported and nothing is torn down. The log line says which case you are in.

Why five minutes of silence. A shard carries on the order of a hundred symbols. The question is not "has this symbol been quiet" — an illiquid pair can go minutes without a print — but "has every symbol on this shard been quiet at once", which a live shard does not do. Erring long costs a few more minutes of a stale subscription set; erring short costs live data. Both are tunable: resubscribe_failure_limit: and stale_delivery_ms: on start_link/1.

A success resets the failure count, so only consecutive failures count toward the limit.

A shard that never OPENED is a different case, and retries on its own backoff

Everything above is about a shard whose socket is up and has stopped delivering. A shard whose socket never opened at all — a handshake that timed out, a venue refusing connections — cannot be judged that way: it has never delivered anything, so there is no last-delivery instant to age, and "never worked" is not "stopped working".

Such a shard is retried on its own schedule: one second, doubling, capped at a minute, until it opens or the slot is filled by a resubscribe of its own. A :coverage_change warning naming the shard fires once when it first fails to open, and a matching :info fires once when it opens — so a shard going dark at boot is something you hear about rather than something you find by grepping. open_retry_base_ms: and open_retry_max_ms: on start_link/1 tune it.

Until 0.4.30 this case was not handled at all. A failed open logged one WARN line and returned, leaving the shard untracked — not merely un-reopened, absent — and its symbols were orphaned for the life of the process. Reported from a real boot under load: three of four shards timed out their handshake, and stayed dark for 28 minutes across zero reopen attempts until the node was restarted, with the feed alive and healthy-looking throughout.

{:reconcile_timeout, ms, :no_venue_response} says which of two opposite actions applies

A reconcile that never came back now reports {:reconcile_timeout, 60_000, :no_venue_response} rather than {:reconcile_timeout, 60_000}. The third element is a statement of fact: the deadline arrived and nothing answered.

That matters because the two causes call for opposite responses. A venue refusing — because another session holds the permission, say — must be waited out, since reconnecting claims a further session against the same ceiling. A wedged socket of ours must be reconnected, which is the only thing that helps. A venue that answers with a refusal reaches you as that refusal and never as this tuple, so the two are now distinguishable without either being inferred.

You do not have to act on it: the escalation above is this package taking the reconnect decision itself, on the one signal it has and you do not.

Streaming calls answer; they never exit your process

subscribe/2, unsubscribe/2, update_symbols/2 and subscribe_notices/1 return {:error, :feed_not_started} when this venue's feed is not running (unsubscribe/2 answers :ok: there is nothing to stop). They return {:error, :feed_timeout} when the feed is too busy to answer within its call budget, and {:error, {:feed_exited, reason}} when it dies while answering. coverage/1 and coverage_by_kind/1 answer an empty map in all three cases, which means "not observed". Treat :feed_timeout as transient and retry. Treat the other two as a feed your supervision tree has to bring back.

:websocket_url in application config replaces the default socket endpoint

config :dp_exchange_webull, websocket_url: "wss://your-proxy.example/..." sends every socket this package opens, unless a call passes its own :url, to that address instead of the venue. Unset, the venue is the default. It exists so a test suite never dials the live venue by accident, and it is equally usable for a proxy.

A print that cannot be delivered says so

Trades come from the tick topic, requested by default, which the vendor lists for stocks, futures and crypto. A tick this package cannot turn into a Trade (undecodable, or missing a readable time, price or volume) raises a :data_quality Notice once per reason per connection, with the raw field strings in details (dropped, symbol, time, basic_timestamp, price, volume). So does a message on a topic this package does not decode (details.dropped == {:unrecognised_topic, topic}). No Trade and no such notice means the venue sent no ticks for those symbols. A notice means it did, and says why each kind was dropped (dp-exchange-core issue #40). A tick whose own time is unreadable falls back to the frame's timestamp before being refused.

The same holds for the other two topics. A snapshot with no readable price (details.dropped == {:snapshot, {:invalid_decimal, :price}}, or {:snapshot, :undecodable}), a quote book that states no readable level ({:quote, :no_readable_level} or {:quote, :undecodable}) and an MQTT packet type this package does not handle ({:unhandled_packet, type}) each raise one :data_quality Notice per reason per connection, with the venue's strings in details. The reason never contains the venue's offending value, and a connection reports at most 64 distinct reasons, so a venue sending something different every message cannot make the notices, or this package's memory, grow with it. Before this, an unreadable snapshot or book was dropped in silence.

A streamed TopOfBook carries a bid_size only with a readable bid, and an ask_size only with a readable ask. A level whose price is "null" but whose size is a number used to arrive as bid: nil, bid_size: 5.

Credential refusals. A CONNACK of 3, 4, 5, 103 or 104 is a :credentials_rejected Notice. 4 (bad user name or password) and 5 (not authorised) are MQTT's own codes, which the venue's table does not list; they used to be a bare :link_down.

MQTT acknowledgements. The QoS a session is delivered at is the venue's choice, not this package's, and the documentation states none. A QoS 1 or 2 message is acknowledged (PUBACK, PUBREC, PUBCOMP) and delivered as any other. Unacknowledged, a broker's in-flight window fills and delivery stops on a connection that still looks healthy.

Duplicates by QoS (MQTT 3.1.1 §4.3). QoS 2 is delivered once: a retransmitted PUBLISH (DUP set, or a packet id reused before its PUBREL) is acknowledged again but not handed to you a second time. The packet ids held for that are bounded at 65,535 and dropped on disconnect. QoS 1 is at-least-once by the spec: a repeat of an acknowledged id is a new publication whatever DUP says, so it is delivered again. For a quote feed the newest value wins and a repeat is harmless; the venue's own venue_time is on every struct if you need to tell a repeat from a new value.