• Kalshi and Polymarket both list a contract for the same real-world question — e.g. "Will the Fed maintain rates in January 2026?"
• If the two platforms price it differently enough, you can buy the opposite outcome on each platform for a combined cost under $1.00.
• Since exactly one side must resolve to $1.00, the payout is guaranteed regardless of the actual outcome — the profit is locked in at entry, not dependent on being right about the Fed.

Real example, Nov 19, 2025, 17:45 UTC:

• Kalshi: bought No at $0.47
• Polymarket: bought Yes at $0.48
• Combined cost: $0.95 → guaranteed $0.05/contract edge, 76 contracts, before fees

• Yes. Using a $0.05 gross-spread threshold sustained 5+ minutes, this single contract had 80 distinct qualifying episodes across its ~2.5-month life.
• Activity is not steady — it clusters hard around specific weeks (late Nov, one week alone had 27 events).

• Yes — one wallet's behavior looks like an automated arbitrage bot already running this exact strategy.
• Wallet 0xde0463... ("mwenya" / "Honest-Similarity"): 47 trades in a single week, 100% buys, split across both outcome sides on both platforms, executed within 3–5 seconds of each other.

• Often, no. 84.4% of qualifying episodes (54 of 64 checked) had zero realized trades on at least one leg — almost always Kalshi — during the entire window the gap was open.
• That doesn't prove a trade couldn't have filled: a live two-sided quote existed for at least part of the window in every one of those 54 episodes, and for the whole window in about a third of them.
• What's still unknown: the actual size available behind those quotes — neither platform's historical API retains order size, only price.

This is about exiting early — how long capital would need to stay locked up if closing both legs as soon as the mispricing itself corrects, rather than holding to event resolution.

• Median time to converge (gap returns to ≤$0.01): 20.6 hours
• Mean: 66.3 hours — pulled up hard by a long tail (up to ~11 days)
• 51% of episodes converge within 24 hours; 82% within a week
• Every qualifying episode in the full contract history eventually converged — none were ever "stuck" open until resolution

• Kalshi's fee is confirmed and stable: 7% × price × (1−price) per contract, taker only — makers are free.
• Polymarket's fee is not a flat platform rate — it's per-market, toggled on/off, and the exact formula is genuinely disputed between Polymarket's own docs and its on-chain contract code.
• The market actually studied here had fees switched off entirely — not something to assume will hold elsewhere.

Same 10 hedge trades, three fee scenarios:

Fees alone are the difference between this being profitable and not.

This is about entering in time — how much of a window exists to actually execute before the gap itself closes back below the qualifying threshold (a different question from #5's exit timing).

Most windows stay open over an hour on average — not a race against seconds — but the spread between "typical" and "longest" is wide enough that slippage risk shouldn't be assumed away.

All of them showed real, tradeable divergence — November baseline, by bucket:

The fourth bucket ("25bps+ Increase") is flagged N/A — it was an imperfect Kalshi/Polymarket match and excluded rather than reported unreliably. The 50+bps Decrease bucket shows the widest average gap but essentially zero average volume — wide spreads and tradeable liquidity don't necessarily show up together.

• Scheduled Fed releases (2pm ET, always): Oct minutes caused an 8-cent whipsaw — up, then reversed within 20 minutes. Net roughly flat.
• Scheduled but non-Fed data (a delayed jobs report): essentially no reaction at all — largest move all day was 3–7 cents.
• Unscheduled remarks (a NY Fed President's speech, no public calendar entry): the single biggest move of the three — 24.5 cents in under 20 minutes.

Being on the Fed's own calendar doesn't guarantee a big move, and being scheduled at all doesn't either. The largest reaction came from an unscheduled remark about a different meeting entirely.

• Yes, and more strongly than volume was: daily arbitrage-qualifying minutes correlate with total daily price movement at r = 0.63 (Kalshi) and r = 0.78 (Polymarket).
• That's clearly stronger than the r = 0.33 / r = 0.24 already found between arbitrage activity and trading volume.
• A simpler per-tick volatility measure was weaker (r = 0.35 / r = 0.47) — it's specifically cumulative daily movement, not how jumpy any one tick is, that tracks with arbitrage opportunity.

• Fees can flip the trade negative. The same 10 trades go from +$9.01 to −$0.51 depending on the fee assumption used (see #6). Mitigation: confirm the live fee rate for the specific market before sizing a trade, not after.
• Execution/leg risk — one side fills, the other doesn't. If only one leg executes, the "arbitrage" becomes a plain directional bet with real downside. Mitigation: size conservatively against the liquidity actually checked (#4), not the price shown.
• Slippage from a closing window. Qualifying gaps can close in minutes (#7). Mitigation: act promptly once a qualifying gap is confirmed rather than waiting to see if it widens further.
• Capital lockup. A hedge, once entered, may sit for hours to over a week before it can be closed early (#5). Mitigation: size positions with that holding period in mind, not assuming instant recycling.
• Platform/counterparty risk. This entire strategy depends on both Kalshi and Polymarket actually settling correctly and on time — a risk outside any of the data analyzed here.

• Yes — and it looks like it already has been, more than once.
• Wallet 0xde0463...: near-simultaneous fills across both platforms (3–5 seconds apart), consistent hedged sizing, 47 trades in a week — a clear automated signature.
• Wallet "Awkward-Stepaunt": placed the exact same size, 1,041.80 contracts, as five separate buys across five different days — a distinctive, clearly scripted (not manual) pattern, independent of the first wallet.

Multiple independent automated actors already active is itself evidence this is a known, executable strategy — not just a theoretical one.

Worth noting who these actors aren't: large-scale automated trading firms have little reason to bother. A typical entry gap here is just $0.05, and pushing meaningful volume through it collapses the very gap being traded — arbitrage at this scale is self-limiting, so it isn't worth building infrastructure around for a firm that can deploy capital more profitably elsewhere. The wallets above look like individual traders running this manually or with a simple script, not institutional players.

ROI varies meaningfully by event type, driven mostly by how large and how often each event's price gaps open — not luck. Real backtested Overall ROI (vs. starting bankroll), January–May 2026, using the strongest tested strategy variant for each event type:

• Inflation (CPI): 153.7% · Notebook 10
• Bank of Canada: 121.2% · Notebook 08
• Bank of EU (ECB): 92.5% · Notebook 09
• Bank of England: 61.7% · Notebook 07
• US Fed: 37.6% · Notebook 06
• Unemployment: 21.9% · Notebook 12 (Full-Ladder + 20% Take-Profit)

These are total returns over each backtest's multi-month window, not a daily or annualized rate — turning this into a daily figure would need a capital-recycling/compounding model this project hasn't built. Full trade-count, win-rate, fees, and PNL detail for each event lives in the Opportunity section of the Thesis page.