The Microstructure of 15-Minute Crypto Prediction Markets: CF Benchmarks BRTI Basis, Path-Averaged Settlement Lock-In, and Maker vs. Taker Execution on Kalshi

1. The Kalshi 15-Minute Crypto Universe & The Strike Continuity Identity

By querying the Kalshi Series API (category=Crypto) and paginating through settled contracts via the Kalshi Trade API v2 Markets Endpoint, we cataloged 14 registered 15-minute cryptocurrency binary series, of which 10 series have active settled markets as of October 18, 2026. Unlike longer-dated daily or weekly brackets, every single-asset 15-minute contract is a pure directional roll contract governed by an Asian-style 60-second arithmetic settlement average.

Per Kalshi's official rules_primary contract specification (Kalshi Markets API): "If the simple average of the sixty seconds of CF Benchmarks' BRTI before [close_time] is at least the simple average of the sixty seconds of CF Benchmarks' BRTI before [open_time], then the market resolves to Yes." Let BRTI(s) denote the 1-second CF Benchmarks Real-Time Index value at Unix second s. For a 15-minute contract opening at Topen and expiring at Tclose = Topen + 900:

Because contract k expires at the exact timestamp Tclose, k = Topen, k+1 where contract k + 1 opens, the 60-second averaging window that determines expiration_valuek is identical to the 60-second averaging window that determines floor_strikek+1. Across our 2,000-market KXBTC15M sample, this Strike Continuity Identity (floor_strikek+1 ≡ expiration_valuek) holds in 99.85% of consecutive contracts (99.90% on KXETH15M and KXSOL15M, and 100.00% on KXXRP15M, KXDOGE15M, and KXBNB15M). This identity allows researchers and execution engines to reconstruct exact historical BRTI settlement prints even on legacy logs that recorded only subsequent strike prices.

1.2 Intra-Candle Orderbook Liquidity & Spread Lifecycle (15,000 One-Minute Bars)

Using the Kalshi Batch Candlesticks API (period_interval=1) across 1,000 settled KXBTC15M contracts, we measured the top-of-book bid-ask spread (yes_ask − yes_bid) and contract volume for each elapsed minute m ∈ {1, …, 15}. Liquidity follows a distinct U-shaped spread curve and monotonic volume ramp: mean bid-ask spread opens at 2.15¢ in Minute 1 (138,679 contracts), compresses to 1.68¢–1.77¢ between Minutes 2 and 9 (median spread 1.00¢ from Minute 5 through Minute 13), and then widens to 2.46¢ in Minute 14 (where contract volume peaks at 286,706 contracts/minute as the 60-second settlement window begins) and 6.33¢ in Minute 15 (215,969 contracts) as automated market makers pull quotes during the final lock-in seconds.

2. Single-Exchange Coinbase Spot vs. Synthetic Multi-Exchange CF Benchmarks BRTI

A foundational flaw in retail Kalshi trading tools—and in pre-v11 versions of Strike15—is computing strike distance and settlement direction exclusively from instantaneous Coinbase BTC-USD spot. In reality, CF Benchmarks' Bitcoin Real-Time Index (BRTI) aggregates orderbook depth across multiple constituent spot exchanges including Coinbase (BTC-USD), Kraken (XBTUSD), Bitstamp (btcusd), and Gemini (btcusd), and Kalshi settles on the 60-second simple average of that multi-exchange index rather than a final-second spot print.

Variance Attribution of Basis Error: Decomposing the high-resolution 1-second tick reconstruction (Coinbase Trades API, Kraken Trades API, and Bitstamp Transactions API) demonstrates two orthogonal error sources: (1) Temporal Averaging Error (~72% of MAE variance), where using an instantaneous print at Tclose instead of the 60-second simple average over [Tclose − 59s, Tclose] accounts for the drop from $11.68 MAE to $3.26–$3.29 MAE; and (2) Single-Exchange Venue Basis (~28% of MAE variance and 68% of outcome flips), where even a 60-second averaged Coinbase-only feed still misclassifies 6.52% of settlements (16.67% of close calls) due to localized Coinbase orderflow spikes, whereas combining Coinbase + Kraken + Bitstamp cuts outcome misclassifications to 2.17% (and 0.00% on Kraken 1s 60s average).

3. Mathematical & Empirical Analysis of the 60-Second Settlement Window ("Minute-14 Lock-In")

During the final 60 seconds before contract expiration (t ∈ [14m00s, 15m00s]), let τ ∈ [0, 60] denote the number of seconds already elapsed inside the CF Benchmarks settlement window, and let u = 60 − τ denote the seconds remaining until expiration. At time t = 14m + τ, the final settlement price Ssettle decomposes into a deterministic locked-in partial average S̄[0,τ] and a stochastic remaining path average S̄(τ,60]:

Notice the super-linear (1 − τ/60)3/2 / √3 collapse of remaining settlement volatility relative to 1-minute spot volatility σ1m:

• At t = 14m00s (τ = 0s locked in): σeff(0) = σ1m / √3 ≈ 0.5774 σ1m (42.3% lower volatility than a point-expiry option over the final minute).
• At t = 14m15s (τ = 15s, 25% locked in): σeff(15) = σ1m (0.75)3/2 / √3 ≈ 0.3750 σ1m (62.5% lower volatility).
• At t = 14m30s (τ = 30s, 50% locked in): σeff(30) = σ1m (0.50)3/2 / √3 ≈ 0.2041 σ1m (79.6% lower volatility — nearly 5× tighter than a 1-minute spot move).
• At t = 14m45s (τ = 45s, 75% locked in): σeff(45) = σ1m (0.25)3/2 / √3 ≈ 0.0722 σ1m (92.8% lower volatility — 13.9× tighter than a 1-minute spot move).

4. Why High-Win-Rate Taker Bots Bleed: Execution Isolation (C1 → C4) & Maker vs. Taker Fees

A central paradox in binary prediction markets is how a directional signal achieving a 70.66%–75.48% win rate can still lose money net of execution costs. On Kalshi, transaction fees follow a parabolic variance-weighted formula on contract count C = ⌊Stake / P⌋ and contract price P ∈ (0.01, 0.99) (Kalshi Fee Schedule):

In our forensic audit of the historical July 2026 live trading dataset (strike15-rescue.json, N = 634 live trades, $25,365.00 total stake) and our 1,000-market October 2026 backtest (N = 310 baseline signals at $100 stake), we isolated four compounding execution leaks in the legacy v10 architecture:

1. Mid-Candle Auto-Exit Spread & Double Taker Fee Bleed (kalshi-exit.js): Settling a winning binary contract at expiration pays $1.00 with $0 settlement fee and 0¢ bid-ask spread. By contrast, closing a position mid-candle via an Immediate-or-Cancel (IOC) order with a −15¢ aggressive limit buffer crosses the bid-ask spread, incurs 2¢–5¢ of book-walking slippage, pays a second 7.0% taker fee, and caps winning payouts at 60% of maximum gain while leaving losing trades uncapped at −100%. Across 118 early-exited live trades in July 2026, premature profit-capping forfeited −$918.03 in gross profit (−$805.59 net counterfactual drag) and paid $57.91 in extra exit taker fees.
2. Aggressive +10¢ Entry IOC Slippage Buffer & Unlogged Fill Telemetry (kalshi-trade.js): Legacy kalshi-trade.js sized contracts from the unbuffered quote (Math.floor(betSize / (priceInCents / 100))) and appended a +10¢ IOC limit buffer (Math.min(priceInCents + 10, 99)), permitting fills up to 10¢ worse than the EV gate's maximum acceptable price and causing budget overruns of +12.5% to +33.3%. Furthermore, because Kalshi API v2 returns taker_fill_cost and fill_count_fp rather than average_fill_price, fillMult was null on 100% of the 634 live trades—hiding −$376.92 of real account drag per 1¢ of adverse fill slippage across ~37,692 contracts.
3. Multi-Leg Intra-Candle Stacking: Firing multiple correlated bets into the same 15-minute candle accounted for 177 extra trades ($8,785 stake) that had a 68.36% win rate and generated −$337.35 Gross (91.7% of all gross losses) / −$474.50 True Net (−5.40% ROI), led by Leg #2 in a candle (n = 151, 66.23% WR, −$534.72 True Net, −7.60% ROI).
4. Taker (7.0%) vs. Passive Maker (1.75%) Execution: Holding the exact 310 baseline signals fixed on the 1,000-market October 2026 dataset and stepping from C1 (v10 IOC Taker + Auto-Exit) to C4 (v11 Passive Maker Limit at Ask − 2¢ + Hold to Expiration) flips net P&L by +$2,075.33 (+6.73% of notional stake):

5. Multi-Strategy Backtest Results Across 1,000 KXBTC15M Markets & 883 July Intervals

We evaluated four quantitative signal architectures across the 1,000-market October 2026 KXBTC15M candlestick dataset ($100 standardized stake per trade, exact Kalshi fee schedule, strict chronological split-half out-of-sample testing) and cross-validated on the 883 reconciled July 2026 intervals:

• Strategy A — Synthetic BRTI Volatility-Adjusted Black-Scholes Digital Fair Value Model: Estimates fair probability pYES(t) = Φ([ln(SBRTI(t)/K) + μ̂(t)] / [σ̂EWMA(t) √(14.333 − m)]) using EWMA 1-minute realized volatility (λ = 0.94), a multi-exchange lead-lag drift term μ̂(t), and the exact path-averaged variance horizon τeff = 14.333 − m. Calibrated within 0.10 pp to 1.69 pp across all probability deciles between 50% and 90% (N = 5,000 evaluations), generating +$1,520.68 net P&L (+1.72% ROI, 889 trades, 78.97% WR) as a Passive Maker.
• Strategy B — Minute 14 "Settlement Window Lock-In" (t = 14m00s – 14m30s): Exploits the (1 − τ/60)3/2 / √3 variance collapse inside the active 60-second BRTI averaging window. In October 2026 (1,000 markets), Minute 14 Taker earns +10.34% ROI (+$617.77, 88.33% WR, Profit Factor 1.874, +10.99% 1st Half / +9.69% 2nd Half) and Minute 14 Maker earns +9.63% ROI (+$815.14, 90.59% WR, Profit Factor 2.021, Brier 0.0946). In July 2026 (883 markets at 14m15s when 25% of settlement is locked in), Taker earns +20.74% ROI (59W–3L, 95.16% WR, Profit Factor 5.174) and Maker earns +23.53% ROI (58W–3L, 95.08% WR, Profit Factor 5.755). By contrast, trading one minute earlier at Minute 13 (t = 13m00s, before the 60s window opens) loses −16.14% ROI—proving the edge is structural to the 60-second settlement averaging window.
• Strategy D — Volatility-Scaled Sigma Bucket (|z| ≥ 0.85σ) + Multi-Exchange Confirmation (Minutes 3 & 5): Requires constituent exchanges (Coinbase and Bitstamp) to agree on strike direction with cross-exchange basis dispersion ≤ 3.5 bps (which lifts leader persistence by +9.85 pp to +18.18 pp over disagreement intervals). Decomposing by read point proves that Minute 3 Maker (+5.96% ROI, +$818.70, 87.68% WR, +6.86% 1st Half / +5.06% 2nd Half), Minute 5 Maker (+4.70% ROI, +$692.41, 87.16% WR), and Minute 3 Taker (+4.05% ROI, 84.62% WR) drive all early-candle alpha, whereas Minutes 7 and 10 are negative EV (−1.81% and −0.63% Maker; −0.58% and −7.65% Taker) because market makers price mid-contract Brownian diffusion efficiently.

6. Cross-Venue Price Discovery: Kalshi vs. Polymarket Across 4,404 Paired Book Snapshots

Across 4,404 settlement-verified paired orderbook snapshots sampled simultaneously on Kalshi and Polymarket for the exact same 15-minute interval iv and elapsed minute m ∈ {3, 5, 7, 10, 12, 14}, the two venues disagree on the Leader's ask price by >4¢ (≥5¢) in 2,301 pairs (52.25%) and by ≥4¢ in 2,643 pairs (60.01%), with an overall mean absolute gap of 7.89¢ (and 13.35¢ when |gap| > 4¢).

Two-Leg Cross-Venue Arbitrage Verification (N = 3,767 Non-Basis Full-Book Pairs): When evaluating simultaneous two-leg arbitrage locks (buying Leader YES on the cheaper venue and Leader NO on the dearer venue, excluding 61 near-strike oracle-split pairs where |gap| > 40¢), 3,069 of 3,767 pairs (81.47%) lock a gross profit (combined cost < 100¢) with 0 lost locks upon official settlement verification (100.0%), and 2,469 pairs (65.54%) lock a net profit after deducting both Kalshi taker fees and a 2¢ Polymarket cost estimate, with a median net lock of +6.60¢ (mean +8.34¢) per contract pair.

7. Strike15 v11 Reference Architecture & Live Suite Access

Every empirical finding in this paper is directly implemented in the Strike15 v11 Production Suite (strike15.com/terminal):

• 4-Exchange CF Benchmarks BRTI Synthetic Proxy (/.netlify/functions/brti-proxy): Polls Coinbase, Kraken, Bitstamp, and Gemini in parallel, computing the outlier-trimmed multi-exchange composite, cross-exchange dispersion in basis points, and the exact second-by-second running 60-second settlement average during Minute 14.
• Minute-14 Settlement Lock-In Engine & Read-Point Gate: Restricts early directional entries to Minutes 3 and 5 with multi-exchange confirmation (|z| ≥ 0.85σ, dispersion ≤ 3.5 bps), disables negative-EV mid-candle reads (min7, min10), and activates the Minute-14 Lock-In calculator at t = 840s elapsed.
• Passive Maker / Tight-Taker v2 Order Execution (kalshi-trade.js & kalshi-order-status.js): Sizes contracts strictly from worst-case limit price (eliminating budget overruns), caps slippage to preserve positive EV, defaults to resting Maker limit orders (cutting Kalshi fees by 75% from 7.0% to 1.75%) held to expiration ($0 exit fee), and computes exact realized fill prices from taker_fill_cost + maker_fill_cost divided by fill_count.
• Real-Time Polymarket Cross-Venue Divergence Gate (/.netlify/functions/polymarket-quote): Streams live Polymarket CLOB orderbook quotes alongside Kalshi orderbook depth, highlighting PM_Ask − K_Ask > 4¢ leader opportunities (+15.24% Taker / +18.07% Maker ROI) and blocking trades when Kalshi is overpriced relative to Polymarket.

Cite This Working Paper