wos-rfq

The client of the Wolverine tick-quote add-on. The add-on stores private::RealFutureQuote at granularity 0, one row per tick, under the DATED contract code. This package is pure Python and works from any host with Python 3.9 or newer — and inside the Wolverine ring images, on the image's own binding.

pip install wos-rfq          # also installs wos-py (the native pycaitlyn binding) >= 0.0.8
wos-rfq --help

Endpoints

The add-on has no login. The client talks to two endpoints:

• The storage master, raw TCP — the binary protocol, the default and fast route for reads. Pass --master HOST:PORT, or set WOS_RFQ_MASTER.
• The add-on's web. Pass --web http://HOST:PORT, or set WOS_RFQ_WEB. It carries the upload API and a JSON read API (/rfq/api/structs, /ticks, /bars, /main) for inspection and debugging. Reads use it only when asked: --via web (--via master is the default; --via auto = web when it answers). JSON is much slower than the binary route; no binding is needed for it.

Commands

The same features are available from Python as wos_rfq.master, wos_rfq.reads, wos_rfq.ticks, wos_rfq.convert and wos_rfq.web (ReadWeb is the JSON read client).

Bars — the sampler's rules

wos_rfq.ticks.fold (and read --bars, read_bars, the web's /bars) builds by default the bars the ring's own sampler stores for the contract (global::SampleQuote), traded minute for traded minute:

1. OHLC from traded ticks only — a tick whose cumulative volume advanced over the previous tick of its trade day. A tick that only moved the book adds no price; it is still counted in ticks and kept in the vectors (tick_time_tags, prices, tick_volumes, bid1, ask1, …). traded counts the traded ticks of the bar.
2. The pre-open auction (a tick before the trade day's first section) belongs to the first bar, its volume included.
3. A tick between two sections, or after the close, belongs to the last bar of the preceding section — never forward.
4. No trade, no bar — the sampler stores no bar for a slot in which nothing traded. quiet_bars=True (--quiet-bars, /bars?quiet_bars=1) returns a dense grid instead: those slots too, traded 0, prices empty.

volume/turnover are sums of per-tick deltas of the cumulative fields over every tick of the bar. outside="drop" (a tick outside every section is in no bar) and outside="next" (to the first bar of the following section) remain as explicit choices; rules 1 and 4 apply in every mode. Bar bounds are the binding's tradeday_sections / sampler_tm_bound on the singularity's rows.

A day the singularity has no row for fails with one stable prefix, no singularity sections for <market> <commodity> on <day>:, followed by the case, read from the same master: holiday/weekend, a gap on the global ring (the neighbours <d1>, <d2> answer; use their sections as the manual override …) or a day not yet reached.

Bindings

The storage-master route needs a pycaitlyn binding; it is imported at the first connect, not at import wos_rfq:

• the wos-py wheel (>= 0.0.8) — the declared dependency, for every pip user;
• a ring image's own binding — the SWIG build the ring's workers import. wos_rfq detects it by its surface and does the raw ring frames, the delete body and the error body in pure Python.

wos_rfq.binding() says which one is in use. Without either, import wos_rfq and the JSON read route still work; a storage-master call stops with one line naming what is missing (exit 1 from the command).

Inside a ring image install with --no-deps:

pip install --no-deps wos-rfq

The image already has its binding; installing the wos-py dependency there would put a second pycaitlyn in the venv and, for any process that does not put the image's directory first, shadow the one the ring's workers depend on. The ring images built by the regular pipeline ship this package that way.