Dispute resolution can vaporize a “winning” position when oracle text and reality diverge. This chapter zooms inward: a composite $100,000 prediction-market bankroll—not a whale, not paper—spread across venues, themes, and time. Numbers are illustrative; copy the structure, not tickers.
Snapshot of how capital sat
Roughly twenty-eight percent sat in slow macro and rates theses on regulated order books. Thirty-two percent lived in a US elections cluster—Senate, president, swing props—correlated by design. Eighteen percent targeted regulatory and tech binaries echoing ETF-style dossiers. Twelve percent covered international and conflict books with wider spreads. Six percent funded play-money shadows for calibration without hero trades. Four percent stayed cash as dry powder after volatility spikes.
Open risk in a bad scenario might approach thirty-eight percent of bankroll—not one hundred percent notional, because binaries cap per contract but correlation does not. A “diversified” book of five YES legs on the same election night is one concentrated bet wearing five tickets.
A representative open book
Fed cut by DATE YES near forty-one cents with a thesis stop on hawkish dot plots. Senate control YES near fifty-two cents tied to a seat tree. A presidential hedge leg NO near forty-seven cents watching certifier ambiguity. A BTC flow proxy YES echoing regulatory case study lessons. Ukraine ceasefire NO flagged after a recency spike. A thin cabinet-nominee YES sized small because liquidity failed a volume test. A shadow CPI band logged at 2.4 percent with zero dollars deployed.
Three positions failed a re-buy audit—“would not buy this size again”—scheduled for Friday review, not midnight mood. Endowment is silent until you force the question.
Ninety days of P&L as story
Month one: plus four thousand two hundred, macro limits obeyed, process grade A. Month two: minus six thousand eight hundred, election chase after a poll drop, cooling violated, grade F. Month three: plus two thousand one hundred, contrarian regulatory sizing at half-Kelly, grade B. Net minus five hundred.
Macro calibration bins around fifty-five to sixty-five percent beat market; politics losses were execution and sizing, not “the crowd knew more.” Bias systems hardened after month two; month three improved compliance, not luck. The story is familiar: edge in slow books, leak in fast narratives.
Where dollars worked
Kalshi turnover dominated with roughly 1.2 percent all-in costs and positive macro edge—keep. Polymarket bursts cost near 2.8 percent on volatile weeks; mixed—limits only. PredictIt caps and fees made it a signal venue, not home. Manifold at zero fees expanded shadows.
Paying curve tax on automated-market-maker legs without simulator discipline turned small edges negative—three trades tagged slippage in the journal. Hybrid venue chapters explain why simulators belong in the workflow, not optional tooling.
Correlation stress on one bad night
Senate, president, and three swing props YES could lose together: roughly twenty-one thousand four hundred dollars, twenty-one percent of bankroll. Policy capped correlated open risk at twenty-five percent; after a cluster hit, no new correlated entries until Sunday review—preventing narrative double-down.
Stress tests are not pessimism. They are how you discover that five well-reasoned legs share one shock.
One trade through the full stack
“Party A wins State X by at least two points.” Scout blind forecast forty-four percent at 08:12. Market open fifty-one cents—negative edge after fees, pass. Forty-eight hours later market forty-six cents, scout forty-seven percent, thin positive edge, half-Kelly suggests eleven hundred dollars, limit at forty-five fills over three days, resolves YES, plus fourteen hundred fifty realized.
Without gates, a market order at fifty-one would have destroyed expectancy. Limits saved the trade. The same thesis at the wrong price is a bad trade.
Quarterly review rhythm
Freeze story twenty-four hours: export positions, no trades. Tag leaks to bias labels; name top two. Brier by theme on locked forecasts only. Venue audit: fees, slippage, rule mismatches—keep or kill. Draw correlation tree for open risk. Install one friction change in the journal.
If more than forty percent of losses tag recency and affect, halve unit size for thirty days—do not “trade through it.”
Cashflow discipline
This teaching book assumes no monthly deposits. Real accounts should cap inflow when volatility is high. Withdrawals never from open cluster weeks. Fees and slippage tagged by venue. Reinvest realized P&L only after scorer review.
Treating bankroll as revolving credit after a loss turned month two into tilt; a freeze now blocks deposits seven days after a minus five percent month.
Pre-written election-night scenarios
Base case mark-to-market minus eight percent: hold limits, no new risk. Bad minus eighteen: halve units, raise cash to fifteen percent. Worst minus twenty-eight: close props, keep hedged core only. Gap risk on one leg triggers dossier exit.
Pre-writing removes affect entries at midnight. The portfolio case study is where sizing theory meets a ledger.
Defaults for a book this size
Cap a single binary near two percent after Kelly, correlation, and liquidity. Cap theme cluster near twenty-five percent open risk. Raise cash floor after large hourly spikes. Marquee on central limit books; long-tail only with simulators. Shadow daily; promote live only when ninety-day calibration beats market. Sunday scorer changes rules; no revenge Monday size-up.
Role split: Scout, Trader, Scorer on one book
The illustrative trader who lost month two was the same person scouting, clicking, and grading—a recipe for recency. After hardening systems, Scout updates forecasts on a calendar; Trader only acts inside gates; Scorer edits rules on Sunday without opening new risk. The portfolio post-mortem is where organizational habits matter even for solo accounts.
If you cannot separate roles in time, separate them in documents: two files, two timestamps, no edits after the fact.
When shadows beat live size
Manifold shadows on CPI bands and wild props trained calibration without paying spread. Promotion to live size followed ninety days where shadow Brier beat market in macro bins—not because shadows were “easier,” but because they forced daily writes without dopamine fills. The ledger’s six percent “play” bucket was not entertainment; it was tuition paid in discipline.
Withdrawals and identity
A hundred thousand dollars is psychologically a “real account” for many readers. Withdrawals to checking funded life expenses; the policy blocked withdrawals during open election clusters so P&L stress did not become rent stress. Corporate readers mapping this to hedging budgets should copy the separation: operational hedges are not ego trades.
PredictIt as signal, not home
Eight thousand dollars turnover on capped retail books bought informational prints, not scalable edge. The portfolio used PredictIt mids to cross-check regulated prices, not to deploy size. Post-mortem asks whether you mistook a capped mid for a tradable institutional forecast—a mistake the 2020 and 2024 election cases warn explicitly.
Month-two leak dissected
The minus sixty-eight hundred month traced to one election-week cluster: market orders after a poll drop, no cooling, three correlated legs. Tags: recency, cluster_cap_violation. The fix was not “trade less politics”; it was enforce cooling and caps. Month three’s recovery came from compliance, not from changing political views.
Copy the structure: buckets, caps, shadows, one rule change per quarter. The ledger tells the truth your memory will edit.
What to carry forward
A hundred-thousand-dollar book is mostly correlation and execution risk, not bragging rights. Illustrative ledgers show macro edge and politics leak until gates harden. Quarterly post-mortem flows tags to calibration to venues to one rule change. Defaults: two percent single, twenty-five percent cluster, cash after spikes, shadows before size.
Next: Market Manipulation Attempts — wash trading, spoofing, and narrative pumps on thin pools.