Who Can See Kalshi's RFQ Surface

A correction to something widely repeated, including in my own earlier notes. Part of Kalshi Research.

The claim

Kalshi prices combo contracts, its parlays, through a request-for-quote system rather than a resting order book. Retail buys the YES side; someone else quotes the other side.

The commonly repeated version of this is that the RFQ surface is invisible and inaccessible to ordinary accounts, restricted to approved institutional market makers. Every write-up I found traces that to press reporting rather than to the API. Kalshi’s own documentation names no eligibility gate: the RFQ page says a request “is broadcast to all makers,” and the create-quote endpoint documents no permission requirement. The only access language in the spec concerns futures commission merchant members and subaccounts.

Nobody in the sourced literature had tested it.

What a normal account actually sees

Authenticated with a standard API key on a retail account, GET /communications/rfqs returns 200 requests, 95 of them open, in a two-hour window on a Sunday afternoon.

So the visibility half of the claim is false. Whether the same account can answer a request is a separate question, and it cannot be tested read-only: submitting a quote is a live order that can be accepted and confirmed into a fill. That test was deliberately not run.

The records are richer than expected

Every request carries its full leg list with sides:

{
  "id": "9e5429d8-...",
  "market_ticker": "KXMVECROSSCATEGORY-SHARD1-S2026F7B165D6601-...",
  "mve_collection_ticker": "KXMVECROSSCATEGORY-SHARD1-R",
  "target_cost_dollars": "2...",
  "status": "closed",
  "creator_id": "",
  "mve_selected_legs": [
    { "market_ticker": "KXMLBTEAMTOTAL-26SEP071410CHCMIL-MIL3", "side": "yes" },
    { "market_ticker": "KXMLBTEAMTOTAL-26SEP071710WSHSD-SD4", "side": "yes" },
    { "market_ticker": "KXMLBTEAMTOTAL-26SEP072205TORATH-TOR4", "side": "yes" },
    { "market_ticker": "KXNCAAFSPREAD-26SEP07SMUFSU-SMU5", "side": "no" }
  ]
}

That is a complete, priceable object. You can compute a joint probability from the legs’ own live book prices and compare it to what the requester paid, for every request, without ever quoting.

Shape of the flow

From that 200-record snapshot, 17:32 to 19:43 ET on a Sunday:

Open / closed	95 / 105
Median target cost	$10.00
Target cost range	$0.03to$170
Median contract count	232
Leg count, median	6
Leg count, max	79

Leg counts run from 2 to 79 with a long tail past 30 that is lottery-ticket shaped. Legs concentrate in baseball, Champions League, college football and home-run props. Every record in the snapshot belonged to a single cross-category collection.

creator_id was empty on all 200. The requester-profiling channel that incumbent makers reportedly use is not exposed here.

Why this matters

The measured overpayment on Kalshi combos is real and large: 3.3c per combo equal-weighted across 8.93 million settled combos, corroborated by independent reporting on retail combo losses. The reason that is not a strategy is not secrecy. It is that RFQ is a sealed-bid auction where responders cannot see each other’s quotes, so a new entrant wins precisely the requests where their correlation model is most generous relative to the incumbents’. The population average is not what a marginal quoter earns.

That gap between the population number and the winner’s realized edge is measurable without submitting anything. Poll the open request list, price each one against its legs’ live books, and record what the requester paid against your model. Restrict the analysis afterward to the subset your quote would have won. Quoting the population is exactly the error that manufactures the headline number.

Three other things would still bite even with access. The measurement window that produced 3.3c ran entirely inside a no-maker-fee regime that ended 2026-08-20. Combo collateral is local to shard 1 and must be pre-positioned. And Rule 5.11 bust authority was exercised on combo markets twice in the fortnight before this was written.

Method note

Read-only throughout: authenticated GET requests and a websocket subscription. No quote, RFQ, order, accept or confirm call was made. Figures are one snapshot on 2026-09-07 and should not be read as a stable rate.