# Venue and strategy selection for a small, automated, US-based trading operation (6–12 month horizon)

**Desk memo. Observation date: 2026-11-08.** Perspectives used: head of trading, quant researcher, market-structure specialist, risk/compliance officer, adversarial reviewer.

**Profile assumed:** one operator, AI coding agents for research, one always-on cloud VM (not colocated), USD 1k–25k starting capital that can scale, a US person resident in the US, and a strong preference for staying legal.

> **How to read this memo.** Every line is tagged with one of these:
> - **[F]** a fact taken from a cited source, with its date where it matters.
> - **[I]** an inference made by the desk.
> - **[E]** an estimate. All return, capacity and probability numbers are estimates.
> - **[U]** unknown or unverified.
>
> **Method limits.** Research used web search plus direct fetches of a handful of primary pages: Polymarket fee and liquidity-rewards docs, Paradigm's challenge page, The Block, and Datawallet's Papertrade explainer. Several facts below come from search-engine summaries of secondary sources and are marked as such. No order-book data was downloaded, no backtest was run and no account was opened. Nothing here is a guarantee of returns, nor legal or tax advice.

---

## 0. Bottom line (adversarial reviewer signed off on this wording)

1. **No venue/strategy pair is clearly attractive in dollar terms at USD 5k.** [I/E] The best realistic outcome at 5k is roughly a few hundred to about two thousand USD per year after costs. The loss branch is of similar size. Nearly all of the value at this size is in learning: building the simulator, the measurement tooling and a live track record that can be scaled.
2. **US legality sets the menu before economics does.** [F/I] Hyperliquid's interface terms exclude US persons [F, per secondary summaries of the ToS]. So does Polymarket's international CLOB, whose US users are directed to the CFTC-regulated Polymarket US app [F]. That removes most of the richest crypto-native reward programs from consideration. For a US person the legally clean venues are **Kalshi**, **Polymarket US** (a DCM reached through FCMs) and **CFTC-regulated crypto futures** such as Coinbase Derivatives' nano perpetual-style futures.
3. **Ranked shortlist:**
   1. **Kalshi selective passive market making** in mid-liquidity, fee-light series. Use maker incentive programs only where they are live, and treat them as a bonus.
   2. **Polymarket US ↔ Kalshi cross-venue relative value**, built as maker-first quoting that leans on the other venue's price, not pure taker arbitrage.
   3. **Regulated cash-and-carry / funding capture** (Coinbase nano perp-style futures against spot). This is a low-return benchmark and capital parking spot, not an edge.
4. **The Paradigm challenge is a good model of PM market making, with three important differences in live markets** (section 3):
   - Live retail flow is not symmetric. Longshot buying is biased.
   - Snipers are humans and bots with news, not one omniscient arbitrageur.
   - Fees and rewards change the P&L identity.

   The Kalshi microstructure evidence fits the model. Makers do much better than takers, but the **average maker still lost money** on Kalshi in the Bürgi–Deng–Whelan sample [F]. So naive quoting is negative-EV, and the edge comes from **choosing where and when not to quote**.
5. **If the operator's alternative is T-bills plus their own time,** the honest result is a modest-EV, high-learning project. Size it so that losing it is irrelevant, and kill it fast if the shadow-phase metrics in section 6 are not met.

---

## 1. Evidence base: key current facts

### 1.1 Prediction markets

| Item | Fact | Source |
|---|---|---|
| Polymarket fee formula | `fee = C × feeRate × p × (1−p)`, taker only. "Makers are never charged fees." | [F] [Polymarket docs – fees](https://docs.polymarket.com/trading/fees), fetched 2026-11-08 |
| Polymarket taker rates by category | Crypto 0.07; Sports 0.05; Finance/Politics/Mentions/Tech 0.04; Economics/Culture/Weather/Other 0.05; Geopolitics 0 (fee-free); Combos 0.05 | [F] same page |
| Polymarket maker rebate share | Crypto 20%, Sports 15%, most others 25%, Combos 0%. Paid daily in proportion to filled maker volume, at Polymarket's discretion. | [F] same page; [maker rebates program](https://docs.polymarket.com/developers/market-makers/maker-rebates-program) |
| Polymarket fee rollout | Taker fees started on 15-minute crypto markets in Jan 2026 and extended to most categories on 2026-03-30. Dynamic fees on short crypto markets (reported 2026-07) explicitly target latency arbitrage. | [F] [The Block](https://www.theblock.co/post/384461/polymarket-adds-taker-fees-to-15-minute-crypto-markets-to-fund-liquidity-rebates), [Finance Magnates](https://www.financemagnates.com/cryptocurrency/polymarket-introduces-dynamic-fees-to-curb-latency-arbitrage-in-short-term-crypto-markets/) |
| Polymarket liquidity rewards | Score is `S=((v−s)/v)²·b` within a per-market max spread `v` and minimum size. Two-sided score = min(bid, ask). Single-sided orders score at 1/3 in the 0.10–0.90 range and must be two-sided outside it. Book sampled **once per minute at a random offset**, 1,440 samples per UTC day. Paid daily, $1 minimum. | [F] [Polymarket docs – liquidity rewards](https://docs.polymarket.com/market-makers/liquidity-rewards.md), fetched 2026-11-08 |
| Polymarket reward pools (examples) | Crypto TWAP program (now ended): $1M through August across 5m/15m/4h markets. ">$5M" April 2026 sports/esports incentive. World Cup 2026 program ran 06-11 to 07-19. | [F] docs above; [F, secondary] search summaries |
| Polymarket holding rewards | 4.00% annualized on position value in a small set of long-dated markets. Variable, treasury-funded, paid daily. | [F] [Polymarket docs – holding rewards](https://docs.polymarket.com/polymarket-learn/trading/holding-rewards) |
| Polymarket US | CFTC Amended Order of Designation: a DCM with an intermediated (FCM) model. App launched Dec 2025 via waitlist, broadened in May 2026, sports first. Public REST/WebSocket API with Ed25519 auth reported. | [F, secondary] [Covers](https://covers.com/industry/polymarket-removes-waitlist-launches-for-american-ios-users-may-12-2026), [DeFi Rate](https://defirate.com/news/polymarket-launches-public-api-unlocks-permissionless-liquidity/). [U] Whether Polymarket US has its own maker reward program, and what its fees are. |
| Kalshi fees | Taker fee ≈ `0.07 × C × p(1−p)`, rounded up. Maker fees only on listed series (e.g. Fed rate, CPI, jobs, GDP, NFL/NBA in an Oct 2026 list) at 0.0175, about a quarter of the taker rate. No settlement fee. | [F, secondary] search summary of [Kalshi fee schedule PDF](https://kalshi.com/docs/kalshi-fee-schedule.pdf) and guides |
| Kalshi incentives | Volume Incentive Program to end "no earlier than Oct. 13" 2026 amid wash-trading allegations. Liquidity Incentive Program page lists an end of 2026-09-01, but an amended version was filed effective 2026-10-28. A "Designated Liquidity Provider Market Making Program" was self-certified 2026-10-28. | [F] [The Block 2026-09-30](https://theblock.co/news/business/2026-09-30-kalshi-ends-trader-incentive-program-417247); [Kalshi help – LIP](https://help.kalshi.com/en/articles/13823851-liquidity-incentive-program); [F, secondary] CFTC filings list. **[U] Current LIP pool sizes and terms.** |
| Kalshi maker/taker returns | Over 300k contracts: takers lost about 32% on average and makers about 10%. All contracts averaged about −20% to −22% depending on version and fee treatment. Favorite-longshot bias, stronger at taker-accepted prices, shrinking over time. | [F] Bürgi, Deng, Whelan, ["Makers and Takers: The Economics of the Kalshi Prediction Market"](https://www.karlwhelan.com/Papers/Kalshi.pdf); [VoxEU summary](https://cepr.org/index%2Ephp/voxeu/columns/economics-kalshi-prediction-market) |
| Practitioner reward estimates | Claims that naive Polymarket reward calculators overstate net earnings 3–5× because they ignore adverse selection. One example: $5–30/day on $5–20k before losses. | [F that the claim exists; U whether it is true] [DEV](https://dev.to/fatherson/polymarket-liquidity-rewards-build-a-real-calculator-min-size-max-spread-break-even-risk--117o), [Medium](https://medium.com/@ezzekielnjuguna/polymarket-liquidity-rewards-calculator-estimate-maker-rewards-min-size-max-spread-and-risk-7ddc4319590f) |
| Polymarket maker markouts | One independent note finds positive volume-weighted markouts in every maker-volume decile. Single dataset, no book reconstruction. | [F, weak] [microstructure note](https://avdheshcharjan.github.io/notes/polymarket-mm.html) |
| 15-minute crypto markets | Only 36.8% of wallets were profitable in a studied week (pre-fee). | [F, secondary] [Telonex](https://telonex.io/research/top-crypto-traders-polymarket-15m) |

### 1.2 Perps, carry, DeFi

| Item | Fact | Source |
|---|---|---|
| Hyperliquid fees | Tiered by 14-day volume. Base is about 0.045% taker / 0.015% maker. Rebates (−0.001% to −0.003%) only for >0.5% / >1.5% / >3% share of maker volume. | [F, secondary] [Hyperliquid docs](https://hyperliquid.gitbook.io/hyperliquid-docs/trading/fees) via summaries |
| Hyperliquid US access | Interface ToS lists the US as a restricted jurisdiction. Enforcement is by frontend geofence, but the terms still bind users. | [F, secondary] [Datawallet](https://www.datawallet.com/crypto/is-hyperliquid-available-in-the-usa), [OneKey](https://onekey.so/blog/ecosystem/geo-block-hyperliquid-dydx-us/) |
| Funding levels | Trailing 12-month average annualized funding to May 2026: BTC ≈ 5.2%, ETH ≈ 4.1%. Ethena moving into equity perps because crypto funding compressed. Historical ETH OI-weighted funding: 16% (2021), 0.6% (2022), ~9% (2023), ~13% (2024). | [F, secondary] Odaily newsflash via search; [Ethena docs](https://docs.ethena.fi/backing-assets/crypto-basis-trade.md) |
| US-regulated perps | Coinbase Derivatives nano BTC (0.01) and ETH (0.10) perpetual-style futures since 2025-07-21. Up to 10x, funding mechanism, fees "from 0.02%". Available via IBKR and others since Feb 2026. | [F] [The Block](https://www.theblock.co/amp/post/359852/coinbase-derivatives-two-us-perpetual-style-futures-contracts-to-begin-trading-on-july-21), [The Block – IBKR](https://www.theblock.co/post/389256/interactive-brokers-adds-nano-bitcoin-ether-futures-via-coinbase-derivatives). [U] Current funding formula and fee tier. |
| Kalshi perps | The Block reports Kalshi ETH perpetual futures volume (>$5B/month in disputed repeated trades), and Kalshi denies wash trading. | [F] [The Block 2026-09-30](https://theblock.co/news/business/2026-09-30-kalshi-ends-trader-incentive-program-417247). [U] Contract specs and access. |
| Drift | Exploited 2026-04-01 for roughly $200–285M (durable-nonce pre-signed transactions, attributed to a DPRK-linked group). | [F] [Decrypt](https://decrypt.co/363087/solana-drift-protocol-exploited-285-million) |
| Lighter | Zero fees for standard accounts, premium tiers paid. Volume reportedly fell ~80% from its Nov 2025 peak by Mar 2026. Reportedly pursuing a CFTC license, so no US access is established. | [F, secondary] [Blockworks](https://blockworks.co/news/lighter-opens-public-mainnet), [Coinage](https://www.coinage.media/2026/how-lighter-is-looking-to-take-the-perp-dex-crown-in-2026) |
| Papertrade.xyz | Launched 2026-10-10 on HyperEVM. Users trade against a protocol pool at the **Hyperliquid best bid/offer midpoint (no independent oracle)**, up to 1000x, BTC/ETH only. Wins pay an "impact haircut" plus a 2% fee. Unpaid profits go to a FIFO debt queue repaid from new losses. Pool capped at ~$5M with surplus to PAPER stakers. PAPER is minted on losses (100/$ decaying), transfer-restricted. A 2-of-3 multisig can pause, freeze or change fees. No public audit report. A $1.28M launch-day midpoint-manipulation incident. Stakers received $46.15M of $49.78M gross revenue through mid-October. | [F, secondary] [Datawallet](https://www.datawallet.com/crypto/papertrade-explained), [Crypto Briefing](https://cryptobriefing.com/papertrade-staker-rewards-launch-hyperevm/), [Blockworks Research](https://blockworks-research.beehiiv.com/p/lose-money-earn-tokens). [U] US geo-policy. |

---

## 2. Candidate grid (all return figures are [E])

Returns are **annualized, net of fees, adverse selection and infrastructure** (assumed about $30–80/month for a VM and data), on deployed capital. Ranges are deliberately wide: they are roughly 10th–90th percentile desk judgements, not modelled distributions.

| # | Venue + strategy | Edge source (who pays) | Net @5k | Net @50k | Capacity before decay | Competition / tech bar | US access | Verdict |
|---|---|---|---|---|---|---|---|---|
| A | **Kalshi selective MM** | Uninformed and longshot-biased takers pay the spread. Incentive programs where live. | −15% to +30% | −10% to +20% | ~$100–300k per operator across many series | Medium: pro MMs and DLPs in top series, thin competition in the long tail. Seconds-level latency is fine outside news events. | Yes (CFTC DCM, KYC) | **#1** |
| B | **Polymarket US ↔ Kalshi RV (maker-first)** | Price dislocations between two retail pools. Taker-paid spread on the stale side. | −10% to +25% | −5% to +15% | ~$50–200k (pairs are few, and capital is locked to resolution) | Medium-high: many bots scan the same pairs. Resolution-rule mapping is the moat. | Yes (both regulated). [U] Polymarket US API maturity. | **#2** |
| C | **Regulated basis / funding carry** (Coinbase nano perp-style futures or CME vs spot or ETF) | Leveraged longs pay funding and basis. | +0% to +6% (less after fees on small notional) | +2% to +7% | Very large | Low. Basis largely arbitraged. | Yes | **#3 (benchmark)** |
| D | Polymarket international CLOB MM + liquidity rewards + rebates | Rewards (treasury and taker-fee funded) plus spread | −10% to +40% (if accessible) | −5% to +25% | Pools are fixed and shared, so per-market capacity is small | High in rewarded markets: rewards attract MMs until marginal reward ≈ marginal adverse selection. | **No.** International platform restricts US persons, and US users go to Polymarket US. | Rejected (legal) |
| E | Polymarket/Kalshi news and resolution-speed trading | Stale quotes from slow MMs (you are the sniper) | Highly skewed, −30% to +100% | Capacity-limited | Tiny per event | Very high: latency races, colocated feeds. Polymarket dynamic fees aimed specifically at this. | Yes on Kalshi | Rejected: arms race, and venues are fee-taxing it |
| F | Hyperliquid perp MM / funding / HLP / vaults | Spread and rebates (top tiers only), funding, HLP liquidation profits | MM at base tier: negative to small. HLP historically positive but drawdown-prone. | — | Large | Very high for MM: maker fee is 0.015% until large share | **No** (ToS excludes US persons) | Rejected (legal plus competition) |
| G | Lighter / Aster / Drift MM or points farming | Zero fees, points/airdrop speculation | Speculative | — | — | High | Not established. Drift exploited Apr 2026. | Rejected |
| H | Papertrade.xyz (trade, LP, or stake PAPER) | Losers fund the pool, and stakers get surplus | Trader EV **negative by construction** (haircut, 2% fee, payout queue). Staking needs losses to mint PAPER first. | — | Pool cap ~$5M | Mid-price oracle invites manipulation. The only edge is manipulation, which is excluded. | [U] Likely gambling/bucket-shop exposure for a US person. | Rejected |
| I | Other PMs (Limitless, Myriad, etc.) | Same as A, thinner | [U] | Small | Low competition but thin flow | [U] US access, token incentive dependence | Rejected for now (insufficient evidence; revisit) |
| J | On-chain DEX LP / MEV / CEX-DEX arb | Swap fees minus LVR, or ordering rents | LP: often ≤0 net of LVR for passive. MEV: concentrated in specialist builders/searchers. | — | — | Very high | Legal grey areas for some MEV | Rejected |
| K | Polymarket holding rewards (4%) | Treasury subsidy | ≈4% minus price risk of the position | Small set of markets | — | Low | Not for US persons (international) | Rejected (legal; below T-bills on a risk basis) |

**Why the 5k and 50k ranges differ:**
- Fixed costs: infrastructure plus operator time are a larger drag at 5k.
- Queue and size limits: at 50k you cannot stay inside top-of-book in the long tail without moving it.
- Prediction-market capital is locked until resolution, so turnover limits realised return.

[I]

---

## 3. Testing the "Paradigm challenge ≈ live PM market making" claim

**What the challenge is [F]** ([Paradigm](https://www.paradigm.xyz/puzzles/prediction-market-challenge/about), fetched 2026-11-08):
- **Contract and price process:** a binary contract on a 1–99¢ FIFO book. The latent score follows a Gaussian walk plus compound-Poisson jumps (about 2 jumps per run).
- **Arbitrageur:** an informed actor that knows p_t and sweeps every stale quote.
- **Retail:** Poisson arrivals, buys and sells with equal probability, LogNormal size.
- **Competitor:** a static ladder that never re-centres.
- **Timing:** you quote *before* the price moves, then the arb hits, then retail arrives.
- **Scoring:** passive edge against p_t.

A placed competitor reported that inventory skew and size control were decisive [F, secondary: [write-up](https://hn.jakechampion.name/reader/https://github.com/octavi42/prediction-market-maker)].

**Where it matches live markets [I, supported]:**
- Kalshi data show the structural split the challenge assumes: takers (largely uninformed retail) lose much more than makers [F, Bürgi et al.].
- Polymarket's own design reacts to latency sniping of stale quotes. Dynamic fees on short crypto markets were built to "redirect value from latency arbitrage toward genuine liquidity provision" [F].
- On Kalshi, single-name markets show the highest informed price impact, yet makers still earn more per contract there [F, secondary: NBER paper via search summary] (consistent with spread being compensation for adverse selection).

**Where it breaks [I], and what each break implies:**

1. **Retail is not symmetric or uninformed in the challenge's sense.**
   - Live flow has a **favorite-longshot bias**: takers overpay for cheap contracts [F].
   - Implication: the maker's edge depends on which side is quoted and at what price. Leaning offers into longshots is the structural edge.
   - The challenge's symmetric retail hides this. A simulator must model price-dependent buy/sell imbalance.
2. **The "arbitrageur" is a distribution, not a god.**
   - Live snipers are news-fast bots on scheduled releases (CPI, FOMC, game events), plus other venues' prices (sportsbooks, CME FedWatch, Polymarket ↔ Kalshi).
   - Adverse selection is concentrated in time, so a maker who pulls quotes around scheduled events and when correlated venues move avoids most of it.
   - The challenge's continuous omniscient sweep overstates adverse selection in quiet periods and understates it at events (jumps there are larger and more frequent).
3. **The average live maker lost about 10%** in the Kalshi sample [F].
   - In challenge terms, the median live maker quotes too tight or too large through jumps.
   - Edge exists only for the selective maker: quote-pulling, skew and series selection.
   - [I] So the challenge's lesson ("adapt from observed fills") is necessary but not sufficient. Live, the strongest signals are exogenous: event calendars and cross-venue prices.
4. **Fees and rewards alter the P&L identity.**
   - Live P&L = spread edge − adverse selection − maker fees (Kalshi listed series) + rebates/rewards (Polymarket: 15–25% of taker fees; Kalshi LIP/DLP where live) [F].
   - Polymarket's reward sampler (once per minute at a random offset, quadratic in distance from mid) pays for *time-at-touch*, not fills [F]. That creates a known exploit-and-defence game: quote tight but small, and pull fast.
   - The order-age rule on taker-delay markets [F] is the venue's defence against that.
5. **Resolution and settlement risk.** The challenge settles mechanically. Live markets carry rule ambiguity (UMA disputes on Polymarket, Kalshi rulebook interpretation) and multi-week capital lock [F, secondary: [Orrery](https://orrery.me/learn/polymarket-vs-kalshi-resolution-rules)].
6. **Competition is adaptive.** The challenge competitor is static. Live competitors re-centre and also chase reward pools, so reward APYs compress until marginal reward ≈ marginal adverse selection [I].

**Net verdict:** the structure is right and the parameters are wrong. Use a challenge-style simulator as the core, but add four things:
- price-dependent retail imbalance
- event-clustered jump intensity tied to a real calendar
- an external "reference price" process that leads the book
- venue-specific fees, rewards and resolution lags

[I]

---

## 4. Ranked shortlist

### Rank 1: Kalshi selective passive market making

**Why it ranks first:**
- It is the only venue that is US-legal, deep, programmable (REST/WebSocket/FIX API), and has peer-reviewable evidence that the maker side outperforms [F].
- Maker fees apply only to listed series [F], so you can pick fee-free series.
- Incentive programs are live or being restructured (LIP amended 2026-10-28, new DLP program) [F]. Treat them as upside, never as the thesis, because the Volume program was just killed [F].

**Edge mechanism:**
- Earn the spread from structurally uninformed and longshot-biased takers in **mid-liquidity series** (economics, weather, culture/mentions, long-dated politics). There, pro DLPs are thin and informed flow is predictable in time.
- Avoid adverse selection by:
  - pulling or widening quotes around scheduled releases and live game states
  - skewing on inventory
  - leaning against longshot overpricing
  - refusing to quote when an external reference (Polymarket price, sportsbook line, CME FedWatch, weather model update) moves

**Data needed for a faithful simulator:**
- Kalshi historical trades (public API) and **forward-recorded L2 order-book snapshots/deltas** via WebSocket. Assumption [U]: no free historical L2 archive, so start recording on day 1.
- Market metadata and rulebooks, settlement outcomes and timestamps.
- An event calendar (BLS, FOMC, NOAA model runs, game schedules).
- External reference prices: Polymarket public data, sportsbook odds, FedWatch.
- Fee schedule history and LIP/DLP terms, including reward pool per market and scoring.
- Queue-position model: FIFO, so estimate queue ahead from L2 deltas.

**Minimum viable live setup:**
- One VM in a US-East region, a Python/TypeScript bot, Postgres or Parquet recorder, and a risk process separate from the quoting process.
- Hard caps: per-market notional, per-event aggregate, daily loss.
- Kill switch: cancel-all on disconnect, stale data or clock skew.
- Funding: Kalshi KYC account funded by ACH.

**Staged rollout and kill criteria [E thresholds]:**

| Stage | Duration | Capital | Pass criteria | Kill criteria |
|---|---|---|---|---|
| 0. Record and simulate | 3–4 weeks | 0 | Simulator reproduces realised spreads and fill rates within ±25% on held-out weeks | Cannot model fills (queue) within ±50% |
| 1. Shadow quoting (compute quotes, log hypothetical fills vs. real tape) | 2–4 weeks | 0 | Hypothetical markout at +5 min and at settlement > 0 after fees across ≥500 hypothetical fills | Markout ≤ 0, or edge concentrated in <3 markets |
| 2. Small live | 6–8 weeks | $1–3k, ≤$100 per market | Realised net P&L > 0 with 30-day markout positive. Max drawdown <10%. Fill-rate gap vs. shadow <30%. | Drawdown >15%, two consecutive negative 2-week periods, any operational incident causing an unbounded position |
| 3. Scale | ongoing | Step up 2× per month to $25k, then $50k | Return per $ does not fall >40% per doubling | Edge per contract halves, or incentive cut removes >50% of P&L |

**Key unknowns to measure first:**
1. Per-series adverse-selection cost (markout curve) versus quoted spread.
2. Queue position and fill probability at the touch.
3. The share of P&L from rewards versus spread.
4. Current LIP/DLP terms and eligibility for a small account [U].
5. How fast Kalshi DLPs re-centre after releases.
6. Capital lock-up and the turnover it allows.

### Rank 2: Polymarket US ↔ Kalshi cross-venue relative value (maker-first)

**Why it ranks second:**
- Two regulated retail pools quote the same events (sports first on Polymarket US [F]), and prices diverge for reasons the sources document: fees, user bases, resolution sources [F, secondary].
- It is US-legal.

**Why it is not first:**
- Polymarket US's API, fees and reward terms are not verified from primary docs [U].
- Pure taker arbitrage is heavily scanned, and a 3¢ gross edge can be 0¢ net after fees [F, secondary].
- **Resolution mismatch can lose both legs** [F, secondary].
- Capital is split and locked across two venues [I].

**Edge mechanism:**
- Quote passively on the venue whose book is stale or thin, using the other venue's mid as the reference price. Hedge on fill with a taker order only when the net edge clears both fees.
- The moat is a **curated, human-reviewed mapping of economically equivalent contracts**, plus fee-aware hedging logic. It is not secret.

**Data needed:**
- Synchronized L2 recordings from both venues, with clock-synced timestamps.
- Contract-equivalence table (subject, threshold, source, cutoff time).
- Both fee schedules.
- Historical settlement divergences.
- Hedge-latency distribution (time from fill to hedge).

**Minimum viable live setup:** the Rank 1 stack, plus a Polymarket US account through its FCM path, a pair-mapping file with a mandatory human sign-off per pair, and a hedge-failure handler (unwind or hold-to-resolution rules).

**Staged rollout:**
- **Shadow (3–4 weeks):** log detected dislocations, simulated fills and hedge slippage.
  - Pass: ≥2¢ net per pair-trade median with ≥80% hedge success within 5 s.
- **Small live:** $2–5k split across venues, at most 10 pairs.
- **Kill criteria:**
  - any resolution-mismatch loss on a reviewed pair (stop and re-audit mapping)
  - net edge per trade under 1¢ for 4 weeks
  - hedge success under 70%
- **Scale:** only when there is enough pair supply (count dislocations per week over $X).

**Key unknowns:**
1. Polymarket US maker fees, rewards and API rate limits [U].
2. Overlap breadth beyond sports [U].
3. The frequency and half-life of dislocations above fees.
4. Withdrawal and transfer friction between venues.

### Rank 3: Regulated funding/basis carry (benchmark and parking, not an edge)

**Why it is on the list:**
- It is a legal, scalable, backtestable way to earn above cash when funding is rich.
- It gives the desk a baseline. If strategies 1–2 cannot beat this plus T-bills, stop.

**Why it is weak:**
- BTC/ETH funding averaged about 4–5% annualized over the year to May 2026 [F, secondary].
- That is roughly cash-like after fees and margin drag, with tail risk from funding flips and margin calls in a spike.
- At $5k, nano-contract granularity and fees dominate [I].
- Ethena itself moved away from crypto basis because it compressed [F, secondary].

**Edge mechanism:**
- Collect funding/basis paid by leveraged longs.
- Size up only when 7-day realized funding exceeds the cash yield by a threshold (regime switch). Otherwise hold T-bills.

**Data:** Coinbase Derivatives funding/fee history [U on public archive], CME basis history, spot/ETF prices, margin rules.

**Minimum viable live setup:** an FCM account (e.g. via a broker offering Coinbase nano perps), spot or ETF hedge at the same broker, daily rebalance, and a margin buffer of at least 3× historical worst daily move.

**Staged rollout:**
- **Backtest:** must clear cash by at least 2% annualized net in the backtest window.
- **Small live:** 1 contract pair, to verify the funding accrual mechanics match the docs.
- **Scale:** only in rich-funding regimes.
- **Kill criteria:** net spread under cash for 30 days, which means returning to T-bills.

**Key unknowns:**
1. The exact Coinbase perp-style funding formula and fees [U].
2. Margin offsets between spot ETF and futures at the chosen broker [U].

---

## 5. What was rejected and why

- **Polymarket international CLOB (rewards, rebates, holding rewards).**
  - It is the most generous reward ecosystem [F], but the international platform excludes US persons, and US users are routed to the DCM [F/I].
  - VPN or offshore-entity workarounds create legal and account-seizure risk that compliance will not sign off on.
  - Revisit if Polymarket US gets equivalent programs.
- **Hyperliquid (MM, funding, HLP, vaults, builder codes).**
  - The ToS excludes US persons [F, secondary].
  - Economically, small makers pay 0.015% until they reach a large share of maker volume [F], so a small MM pays to provide liquidity against professional HFTs.
  - Builder codes require building a front-end business, which is out of profile.
- **Lighter / Aster / Drift.**
  - US access is not established.
  - Drift suffered a roughly $285M exploit on 2026-04-01 [F].
  - Points-farming EV depends on airdrop speculation, which cannot be backtested.
- **Papertrade.xyz.**
  - Trading is negative-EV by design: haircut plus 2% win fee, with profits subject to a FIFO payout queue [F].
  - The house pool is small relative to open interest [F, secondary].
  - The mid-price oracle has already been manipulated [F, secondary]. The only structural "edge" is manipulating Hyperliquid's mid, which is market manipulation and rejected outright.
  - PAPER is transfer-restricted and minted to losers. Staker dividends (large so far [F]) are a claim on future trader losses with dilution and admin-key risk.
  - US legal status is unclear (bucket-shop or gambling classification risk) [U/I].
- **News and resolution-speed sniping.**
  - It is a latency arms race against colocated bots.
  - Polymarket explicitly re-priced fees against it [F].
  - Defensibility is near zero for a single non-colocated VM.
- **Other PMs (Limitless, Myriad).**
  - Not enough verified data on depth, US access or incentive sustainability gathered in this pass [U].
  - Revisit with a one-week data pull before dismissing permanently.
- **DEX LP / MEV.**
  - Passive LP loses to LVR versus informed arbitrage in most volatile pairs [I, consistent with the LP/LVR literature].
  - MEV is dominated by vertically integrated searcher-builders.
  - Wrong profile.
- **Kalshi Volume Incentive farming.** The program has been terminated [F].

---

## 6. Cross-cutting risk register (risk/compliance)

| Risk | Applies to | Mitigation |
|---|---|---|
| Incentive cut or rule change (Kalshi Volume killed in 2026; Polymarket fees changed several times in 2026) [F] | 1, 2 | Require spread-only P&L > 0. Rewards are upside only. |
| Resolution dispute or ambiguity | 1, 2 | Avoid markets with subjective criteria. Cap per-market exposure. Keep a rulebook review log. |
| Regulatory (CFTC actions on event contracts, state-level sports challenges) [I] | 1, 2 | Diversify series categories, and avoid relying on sports alone. |
| Operational (bot bug leading to runaway position) | all | Separate risk process, cancel-on-disconnect, position caps enforced server-side where available. |
| Correlated event jumps (one release moves many series) | 1 | Aggregate exposure by underlying event, not by market. |
| Tax (US event-contract and futures treatment) [U] | all | Consult a tax professional. Section 1256 treatment of event contracts is not settled here. |

---

## 7. Defensibility summary

None of the top 3 relies on a secret rule [I]. The edge in 1 and 2 comes from execution quality, which competitors can copy only by doing the same work:
- quote-pulling discipline around events
- inventory management
- a reference-price feed
- a human-curated contract map

That is a moderate moat at small size and weak at large size, where professional DLPs compete directly. Rank 3 has no moat; it is a regime bet.

---

## 8. Open questions (unanswered in this pass)

1. Current Kalshi LIP/DLP reward pool sizes and small-account eligibility (amended 2026-10-28).
2. Polymarket US fee schedule, maker programs, API rate limits and product breadth beyond sports.
3. Whether any historical L2 order-book archive for Kalshi or Polymarket US is purchasable (e.g. third-party vendors) or must be self-recorded.
4. Current (Oct–Nov 2026) BTC/ETH funding levels on US-regulated venues.
5. Papertrade's geo-policy and audit status.
6. Limitless/Myriad US access and depth.
7. Exact current Hyperliquid fee table (the sources conflict).

## 9. Sources

Polymarket docs: [fees](https://docs.polymarket.com/trading/fees) · [liquidity rewards](https://docs.polymarket.com/market-makers/liquidity-rewards.md) · [maker rebates](https://docs.polymarket.com/developers/market-makers/maker-rebates-program) · [holding rewards](https://docs.polymarket.com/polymarket-learn/trading/holding-rewards) — [The Block: taker fees](https://www.theblock.co/post/384461/polymarket-adds-taker-fees-to-15-minute-crypto-markets-to-fund-liquidity-rebates) — [Finance Magnates: dynamic fees](https://www.financemagnates.com/cryptocurrency/polymarket-introduces-dynamic-fees-to-curb-latency-arbitrage-in-short-term-crypto-markets/) — [Covers: Polymarket US](https://covers.com/industry/polymarket-removes-waitlist-launches-for-american-ios-users-may-12-2026) — [DeFi Rate: Polymarket API](https://defirate.com/news/polymarket-launches-public-api-unlocks-permissionless-liquidity/) — [Kalshi fee schedule](https://kalshi.com/docs/kalshi-fee-schedule.pdf) — [Kalshi LIP](https://help.kalshi.com/en/articles/13823851-liquidity-incentive-program) — [The Block: Kalshi ends volume program](https://theblock.co/news/business/2026-09-30-kalshi-ends-trader-incentive-program-417247) — [Bürgi, Deng, Whelan](https://www.karlwhelan.com/Papers/Kalshi.pdf) · [VoxEU](https://cepr.org/index%2Ephp/voxeu/columns/economics-kalshi-prediction-market) — [Paradigm challenge](https://www.paradigm.xyz/puzzles/prediction-market-challenge/about) — [Sniper's Tax](https://sybilpm.substack.com/p/the-snipers-tax) — [PM market-making note](https://avdheshcharjan.github.io/notes/polymarket-mm.html) — [Telonex](https://telonex.io/research/top-crypto-traders-polymarket-15m) — [Orrery: resolution rules](https://orrery.me/learn/polymarket-vs-kalshi-resolution-rules) — [Hyperliquid fees](https://hyperliquid.gitbook.io/hyperliquid-docs/trading/fees) — [Datawallet: Hyperliquid US](https://www.datawallet.com/crypto/is-hyperliquid-available-in-the-usa) — [Ethena basis docs](https://docs.ethena.fi/backing-assets/crypto-basis-trade.md) — [The Block: Coinbase perp-style futures](https://www.theblock.co/amp/post/359852/coinbase-derivatives-two-us-perpetual-style-futures-contracts-to-begin-trading-on-july-21) — [Decrypt: Drift exploit](https://decrypt.co/363087/solana-drift-protocol-exploited-285-million) — [Blockworks: Lighter](https://blockworks.co/news/lighter-opens-public-mainnet) — [Datawallet: Papertrade](https://www.datawallet.com/crypto/papertrade-explained) · [Crypto Briefing: Papertrade](https://cryptobriefing.com/papertrade-staker-rewards-launch-hyperevm/) · [Blockworks Research: Papertrade](https://blockworks-research.beehiiv.com/p/lose-money-earn-tokens)