Research and paper only · not financial advice · not an income forecast

Kalshi Income Plan

Saturday October 17, 2026 · written 2:48 AM EDT

Two research agents, called Astra and Fable here, were each asked the same question and wrote a plan without reading the other's. Then each reviewed the other's plan, and both revised. This page puts the two side by side. It is the companion to the Kalshi money plan, which covers what runs today.

1. One screen

The two plans' bankroll tables, side by side

Read these as “what the arithmetic says if the measured edges turn out to be real”, not as forecasts. Astra's is deliberately conservative. Fable's is its revised table; its first version was withdrawn (section 4 explains why).

Where they agree: both put the subscription bill out of reach at $250 and $2,500, and both say $25,000 is a supplement at best until something is proven. Where they differ: Astra won't name a number it can't defend. Fable names the number you would earn if the measured edges hold, with odds attached.

The recommendation, in four points

1. Be the patient side (the one whose order waits on the book) only in slow markets: mentions (will a speaker say a given word), entertainment, weather, and 15-minute commodities. These are where impatient buyers lose the most per dollar staked: mentions −3.0%, entertainment −2.0%, weather −1.4%, 15-minute commodities −1.2%. Mentions, weather, gas and rain also charge the patient side no fee.
2. Build a broken-prices scanner. It looks for prices that break simple logic, and nothing else: a ladder where “above X” is cheaper than “above X+1”; a narrow claim priced above the broader claim that contains it (“wins by 7+” above “wins”); old orders left resting after the thing they track has moved; and the first seconds after a scheduled data release, before resting orders are updated.
3. Reward pools are the one thing found that grows with capital. Kalshi pays pools to traders who keep orders on both sides of some markets. But the scoring rules must be read in full first, and the first pilot runs on paper.
4. Sports stays on paper. Before kickoff, Kalshi's price sits a median 0.94¢ from the sharp bookmaker's (Pinnacle's) fair price, and only 0.4% of sides showed a gap of 2¢ or more after fees. In fast markets the buyer who takes the posted price tends to be the better-informed one (section 2).

2. Where the money goes

Every trade has a taker, who accepts a price already on the board, and a maker, whose resting order was there. Before fees, whatever one wins the other loses. Fable measured this on two weeks of Kalshi's own trade record (October 2–15, 2026, 127.5 million trades, settled markets only). That is a snapshot dominated by in-play sports and 15-minute crypto, not a law.

The pie: Kalshi's fees are about six times what takers lose to makers

Per day, October 2–15. In sports alone, fees run $5.3 million a day against a $0.48 million transfer: eleven to one. Every strategy's first competitor is the fee.

Who loses, by category, as a share of what they stake

Takers' result before fees, as a percent of what they paid. Red = takers lose (makers win). Green = takers win (makers lose).

Science and tech runs off the scale (dashed end). In dollars the biggest losses are in crypto and sports, simply because that is where the volume is. The slow small families (mentions, weather, entertainment) add up to about $28 thousand a day of taker losses across the whole exchange, against about $24 thousand of fees in the same families. In gas, yields and politics the takers win, so makers there lose; a reward-pool business in gas has to pay for that.

The surprise: cheap-side takers win when the money is counted

Older studies of daily closing prices find that people who buy contracts under 10¢ lose over 60% of their money. Weighted by dollars on this October trade record, it runs the other way: takers who bought the cheap side won, and takers who bought the expensive side lost.

Why it matters: most of this volume is in-play sports and 15-minute crypto. There, the person who grabs a 30¢ price the moment a touchdown is scored knows something the resting order doesn't. So “be the house” only works where the market is slow. In a fast, deep sports book, a patient order is the one getting picked off, by about 2¢ a contract. Astra's separate one-night check on a single fight card found the same thing: takers gained +0.74¢ a contract on the winner markets and +1.13¢ on the method-of-victory markets.

3. The two plans compared

What to measure instead of dollars per hour

Both agreed “net dollars per hour” is the wrong target to run the desks on. It divides by hours nobody trades, hides the capital behind it, and only moves after settlement. The combined scorecard:

4. Peer review: the system working

Each agent wrote alone, saved its work, and only then read the other's. Both caught real mistakes, and both changed their plans. That is the point of running two: the errors below were caught before any money moved.

How each responded

• Fable put a correction banner on its first plan and wrote a revision that recomputes every dependent number. It adopted Astra's capital rules: at most 20% of the bankroll reserved for positions plus resting orders, 1% per event, 3% per group of related bets. It renamed “central estimates” to “if the measured edges hold” and added owner hours and return on total bankroll.
• Astra accepted that its 20% cap, turnover and profit-kept figures are policy assumptions, not a forecast. It split sports-maker research from reward-market research, and added a first-week watcher for fee, rule and reward-term changes. It also pushed back on one point. A 1¢ spread does not cap a maker's value at half a cent unless the fair price sits exactly at the midpoint. The question is whether an outside-priced edge survives freshness, fees and bad fills.
• The lag disagreement resolved: both were right. The related market prints within seconds, but only 52% of those first prints go the same way as the main move. The price level finishes adjusting over the next minute (+4.4¢ at one minute, falling to +1.4¢ at five). Neither shows a price you could actually fill at a profit.

5. Ideas

A sixth round looked for new ways to extract value, generalised from the owner's two leads into mechanisms rather than trades.

Fable's top three

Must read before any reward pilot: the scoring sets a reference price where resting size reaches one fifth of the target. If a lone quoter's own orders can set it, unclaimed pools may be real. If not, the reward business is the cost of fills plus a share of the pool.

Astra's round 6: measuring now

• Ladder consistency (“above X” never cheaper than “above X+1”). Scanned on both the archived price history and the recorded order books.
• Final seconds: as 15-minute and hourly markets lock in a settlement average, is the price still under-confident?

These runs finished their first passes between 2:36 and 2:41 AM. Astra has not written them up yet, so this page reports no results from them. The raw scanner alerts have not been checked for false matches.

The owner's two outside leads, as mechanisms

6. The maker proposal: written down, not yet measurable

The owner reopened maker strategies, the ones that post a patient order and wait to be filled. The decision is recorded as maker strategy research reopened (M-151). It allows research, testing by the referee and forward paper trading. The referee is our independent scoring program, which judges a rule on data it was not built on. Live money still needs a passed test, paper evidence and the owner's separate yes.

Astra froze one rule and published a fingerprint (hash) of the file, so it can't be quietly changed after the results are seen.

Why it can't be measured yet

The stored Pinnacle prices were 204 to 1,464 seconds old, median 264 seconds. The rule needs 30. “Not measurable” is not “failed”: fills and profit are unknown, not zero. No threshold was loosened after seeing the result. The fix is a recorder that captures fresh, time-stamped prices going forward. Older held-back data would not help.

Fable's warning stands beside it: in deep sports books our fills are the ones least likely to be good ones. Astra's answer: that is exactly what the frozen rule is built to test, and reopening makes testing allowed, not profitable.

Decision: maker strategies are reopened for research and paper. This proposal waits for the referee owner to review it and for fresh data. Nothing has been submitted.

7. Next steps and gates

No live money moves without all three: a passed test, forward paper, and the owner's yes.

• Week one: fresh, rule-matched prices with source and receipt times, and realistic fills. Add a watcher that halts a desk if a fee, rule or reward term changes.
• The broken-prices scanner runs as a paper desk. It logs every violation it finds, how big it was, how much size was there and how long it lived, before anyone considers trading it.
• Reward rules are read in full before any pilot. The reward pilot runs on paper first, with rewards at zero until a credited payment reconciles.
• Sizing goes up only on evidence, never by the calendar. A step needs 100 settled fills across at least 10 dates, a mean above 0¢ after fees and no price move against us after fills. Never two steps in one week.
• Day 30: an income statement, a drawdown and capital report, owner hours, and a continue-or-stop decision. A positive month alone does not earn more size.

8. Kill criteria: when to admit there is no income here

Both say the same thing at the end: if it's dead, say so, with the numbers, the same day. Then ask what else the time should go to.