High confidence. Sourced from Kalshi’s fee filing, rulebook v1.29, API documentation and live series metadata as of 2026-09-07. These are the constraints an edge has to clear before it is an edge. Part of Kalshi Research.
Fees
Taker fee per fill, rounded up to the cent:
fee = ceil(0.07 x contracts x P x (1 - P))
No settlement fee. Index series use a 0.035 multiplier instead of 0.07.
The curve is the whole story. It peaks at the midpoint and collapses at the extremes:
| YES price | Taker fee per contract | As share of stake |
|---|---|---|
| 5c | 0.33c | 6.65% |
| 25c | 1.31c | 5.25% |
| 50c | 1.75c | 3.50% |
| 75c | 1.31c | 1.75% |
| 95c | 0.33c | 0.35% |
A round trip at the midpoint costs about 3.5c against a 1c tick. That single fact rules out most naive strategies: anything crossing the spread twice needs to be right by more than three ticks just to break even.
Because the fee is charged on notional but scales with P(1-P), the cheap side
subsidizes the expensive side. A taker buying a 5c longshot pays 6.65% of stake;
the maker facing them pays either nothing or a quarter as much.
Maker fees depend on the series, not the exchange
The detail most external write-ups miss. Every series carries a fee_type:
fee_type | Series count | Maker pays |
|---|---|---|
quadratic | 13,644 | nothing |
quadratic_with_maker_fees | 159 | 0.0175 x P x (1 - P), a quarter of taker |
quadratic_with_combo_maker_fees | combos | half of taker |
Makers pay nothing on roughly 98.6% of series, including daily temperature ladders, rain, 15-minute and daily crypto, mention markets and most politics. The 159 fee-charging series are the ones carrying the volume: 107 sports series plus the headline macro ladders (CPI, Fed, U-3, GDP).
The set is not static. GET /series/fee_changes exists because batches migrate
between fee types, most recently 2026-09-03.
Resolving a fee type is a two-hop lookup. The markets payload carries no series ticker at all, only an event ticker:
GET /markets/{ticker} -> event_ticker
GET /events/{event_ticker} -> series_ticker
GET /series/{series_ticker} -> fee_type, fee_multiplier
Read both fields. fee_multiplier is where the 0.07 against 0.035 index
distinction lives.
The measured consequence
Independent analysis of 72.1 million Kalshi fills finds makers earn about +1.12% of notional and takers lose about the same. Contracts under 10c lose more than 60% of stake to the taker side. Every category anchored to an external price prices efficiently net of taker fees.
The conclusion is narrow and important: a retail edge has to be maker-side. Not because making is clever, but because fee incidence hands the maker the taker’s cost.
Two caveats on that number, both from the source papers. Makers lost about 2.0% in 2021 to 2023; the gap flipped only after professional liquidity providers arrived in late 2024. And the maker premium may simply be fair compensation for a 33% return standard deviation.
Position and rate limits
- Position accountability is $25,000 per strike per member on most current contract certifications. Designated market makers get ten times that. Kalshi can demand information, freeze increases and liquidate under Rule 5.18, and aggregates accounts under common control under Rule 5.19(f).
- API rate limits are token buckets. Basic is 200 read and 100 write tokens per second with most requests costing 10, so roughly 20 reads and 10 order actions per second. Advanced at 300/300 is a permanent self-upgrade once at least one of the last hundred orders came through the API. Higher tiers are earned daily on trailing-30-day share of exchange volume.
Sharding
The exchange is sharded and balances are local to each shard:
| Shard | Contents |
|---|---|
| 0 | default |
| 1 | combos (KXMVE*) |
| 2 | crypto and commodities, the 15-minute series |
| 3 | tennis and baseball, basketball added 2026-09-10 |
Collateral has to be pre-positioned per shard, and write buckets are per shard too. When crypto moved to shard 2 in August 2026, bots that had not noticed silently stopped filling for hours. A shard reassignment is a silent outage, not an error.
Order mechanics
YES and NO are one security with two views of the same book. Buying NO is selling
YES. You cannot hold both sides of one market, and buying the opposite side
closes your position. Single-market YES + NO < $1 arbitrage is structurally
impossible, whatever the screenshots claim.
Available controls: good_till_canceled with optional expiration,
immediate_or_cancel, fill_or_kill, post_only, reduce_only,
self_trade_prevention, cancel_order_on_pause, fractional contracts to 0.01,
and subaccounts 0 through 63.
Ticks are 1c, 0.5c or 0.2c at the center, and $0.0001 below 1c and above 99c on multivariate markets. A fixed 1c price grid is wrong for a meaningful slice of the exchange.
Order groups accept a delta limit that auto-cancels remaining resting orders if position moves by that limit inside a rolling 15-second window. That is an exchange-side kill switch, and it keeps working when your process does not.
There is no native stop-loss.
Lifecycle traps
- After
close_time, every order operation is rejected withMARKET_INACTIVE, including cancels. Orders resting into close cannot be pulled; they settle. - During an exchange pause nothing can be cancelled unless the order was
submitted with
cancel_order_on_pauseset. - Scheduled maintenance runs Thursdays 03:00 to 05:00 ET.
- The live API data window is about three months. Older data needs the
/historicalendpoints. GET /marketsis roughly 99% combo shards. Use/events?with_nested_markets=trueormve_filter=exclude.
Discretionary settlement
Rule 5.11 lets Kalshi cancel fills outside fair value plus or minus 20c within 15 minutes of the trade. It was exercised on 2026-08-28 and 2026-09-06. Rule 7.1 allows a 24-hour outcome review and Rule 7.2 allows source-agency substitution. Rule 6.3(e), added 2026-04-28, settles at last traded price if the subject dies or violence interrupts the event. Rule 6.3(c) reserves sole discretion to interpret.
The practical implication for any position ledger: a fill is not final for 15 minutes, and an outcome is not final for 24 hours. Documented discretionary outcomes include an Oscars viewership market, a YES settlement on the syllable “de” in “John Deere”, a last-price settlement on Khamenei, and an all-NO board at a Sanders event. Indeterminate outcomes can split 10/90.
Settlement sources decide thin edges
- 15-minute and hourly crypto settle on the 60-second average of the CF Benchmarks BRTI, available over a 5Hz websocket. Not the chart.
- Daily temperature settles on the NWS Daily Climate Report for one named ASOS station in Local Standard Time, relayed by The Weather Company since 2026-08-14, at 0.1C precision with 6-hour maximum groups. Hourly weather settles on METAR instead.
- Macro ladders settle on the exact BLS or BEA series and rounding named in the contract rules.
- Shutdown and policy markets read a fixed morning snapshot, 10:00 to 11:00 ET.
- Mention markets settle on official transcripts, with root-word and closed-press rules.
Incentive programs
All revocable, all under CFTC scrutiny since the 2026-08-12 advisory, and all closed to market-maker-agreement holders, Kalshi affiliates and FCM or IB customers.
- Liquidity Incentive Program. Pays 1,000 per market per day from
random once-per-second snapshots. A snapshot counts only when both sides
rest at least the Target Size, which runs 100 to 20,000 contracts and 300 to
1,000 on live markets. Score is
size x 0.5^(ticks from reference). - Volume Incentive Program. Caps at $0.005 per contract on fills between 3c and 97c, through 2027-09-01.
- Liquidity Provider Program. Auctions designated slots up to $50,000 per series per week, to members holding a market-maker agreement.
- Market Maker Program. Requires two-sided quotes 98% of every hour across covered products, in exchange for reduced fees, rebates, revenue share, disconnect-cancel and ten times position limits.
Two numbers here were measured directly against GET /incentive_programs and both
correct figures that circulate publicly. The live pool is **171,775 often quoted. And the program
sunsets 2027-01-01.
The scoring formula is a direct instruction about how to quote. 0.5^ticks means
a quote two ticks off reference scores a quarter of one at reference, and the
Target Size gate is binary: quoting 99 contracts against a Target Size of 100
scores zero, not a small number.
Liquidity reality
Sports are about 80% of contract volume, and parlays are more than 80% of incremental football volume. Median resting depth at the touch is roughly $4 of notional. 84.8% of markets have zero volume over 24 hours. The top 200 markets carry 62.7% of volume. A few thousand dollars moves a niche binary 10 to 15c. Liquidity is worst between 01:00 and 07:00 ET.
Counterparties on the deep books are professional: Susquehanna arrived as the first institutional maker in April 2024 and quotes roughly 75,000 contracts per side at 3 to 4c on Fed markets at 98% availability. Jump, DRW, Akuna, Wintermute and FanDuel are also named.
Collateral earns about 3.25% annually while it sits. On a one-to-three percent gross edge that yield is not a rounding error; it is the hurdle any months-long locked position has to beat.
Prohibited practices
Rule 5.17 and active surveillance cover wash trades, prearranged trades, self-matching between subaccounts, volume inflation, spoofing, and multiple accounts to dodge RFQ profiling. Trading on material non-public information, by people who can influence the outcome, or on one’s own candidacy is prohibited and enforced: a $172,000 fine with a three-year ban in one case, a lifetime ban in another, roughly 200 probes a year, and accounts frozen before withdrawal.
Self-matching between subaccounts is worth flagging for any multi-strategy deployment. Two strategies quoting the same market from one member account can constitute a wash trade.