Strategy review (2026-10-06)

Profile A on Kalshi turned out to rest on two weekends a season (Kalshi history), so the owner asked for every F1 strategy so far to be backtested on Polymarket and Kalshi side by side, with rolling Kelly sizing on a $10,000 bankroll, and sorted into steady and lumpy. This page is that review: 80 jobs (19 settings sets × 2 venues × 2 seasons, plus 4 baselines) at 16,000 simulations, 342 strategy rows, run on the VM from 2026-10-05 23:46Z (queue sweeps/strategy-review.toml, report racinglines strategy search-report f1 sweeps/strategy-review.toml).

Verdict: A (T1) stays the core taker, on Polymarket only. It is steady there: up in both seasons, and still up without each season's two best weekends. On Kalshi no taker is steady. Every taker is up in 2026 because of Australia and down in 2025 once the two best weekends are removed, so Kalshi taking stays small until live weekends add evidence. The best Kalshi option is maker K, still small and concentrated. Kelly sizing multiplies the dollars but not the evidence: its totals come mostly from two compounding weekends and assume fills the books can't carry.

How to read it

• 2026 is the target season, 2025 the held-out one. P&L is in dollars at the touch (Kalshi bid/ask since PR #55), after fees.
• Without best 2: the season's P&L minus its two best weekends.
• Shape describes the record, not the risk (PR #62). Steady: up in both seasons, even without each season's best two weekends. Concentrated: up in both, but not without the best two in at least one. Mixed: up in one season, down in the other. Losing: down in both.
• Takers T2 to T6 and T10 are each job's update row. T8 and T9 are the early rows of the T1 and T7 jobs. M1 is the baseline's maker row. TB is half T1 and half T8.

Polymarket, flat sizing

Profile	Shape	2026	without best 2	weekends up	2025	without best 2	weekends up	Best 2026 weekend
T1 (A)	steady	+1,375	+112	10/16	+1,432	+350	14/23	British
T7	steady	+1,257	+67	10/16	+1,456	+374	14/23	British
C	steady	+652	+237	9/16	+629	+86	14/23	Japanese
K	steady	+535	+24	8/16	+1,033	+195	14/23	Italian
T9	concentrated	+2,794	+508	10/16	+570	−2	13/23	British
T8	concentrated	+2,707	+421	10/16	+546	−26	13/23	British
T2	concentrated	+1,243	−47	10/16	+1,400	+335	14/23	British
T3	concentrated	+948	−313	7/16	+900	−84	13/23	British
T5	concentrated	+859	−333	7/16	+816	−269	13/23	British
T10	concentrated	+728	−681	7/16	+757	−373	12/23	British
T4	concentrated	+726	−325	9/16	+881	−230	11/23	British
T6	concentrated	+688	−447	7/16	+608	−223	13/23	British
M3	concentrated	+555	−129	7/16	+1,139	+403	15/23	Italian
M2	mixed	+906	+265	9/16	−131	−750	9/23	Monaco
M1	mixed	+500	−203	8/16	−203	−796	9/23	Italian
default taker	mixed	−615	−1,744	5/16	+181	−896	9/23	British

TB (half T1, half T8): +2,041 in 2026, +989 in 2025.

Kalshi, flat sizing

Profile	Shape	2026	without best 2	weekends up	2025	without best 2	weekends up	Best 2026 weekend
T7	concentrated	+2,605	+428	8/16	+1,620	−637	6/24	Australian
T2	concentrated	+2,449	+83	8/16	+1,208	−1,065	6/24	Australian
T3	concentrated	+2,366	−80	8/16	+1,206	−1,060	6/24	Australian
T1 (A)	concentrated	+2,332	−117	8/16	+1,257	−1,025	5/24	Australian
T6	concentrated	+2,322	−22	9/16	+505	−928	6/24	Australian
T5	concentrated	+2,278	−93	8/16	+1,072	−1,204	6/24	Australian
T10	concentrated	+2,115	−352	7/16	+1,059	−1,181	5/24	Australian
T4	concentrated	+1,626	−312	7/16	+1,473	−1,207	6/24	Australian
default taker	concentrated	+1,391	−961	5/16	+1,141	−1,087	9/24	Australian
K	concentrated	+453	−552	5/16	+741	+66	13/24	Italian
M3	concentrated	+87	−534	9/16	+723	−84	13/24	Italian
T9	mixed	+1,885	+166	8/16	−746	−1,154	5/24	British
T8	mixed	+1,678	+120	8/16	−672	−1,104	5/24	British
C	mixed	−32	−329	7/16	+383	−206	12/24	Japanese
M2	losing	−568	−1,178	6/16	−953	−1,451	8/24	Hungarian
M1	losing	−1,261	−1,856	6/16	−185	−632	10/24	Italian

TB: +2,005 in 2026, +292 in 2025.

Rolling Kelly on $10,000

Half-Kelly (HK) and quarter-Kelly (QK) at the touch, $500 a market (scaled with the balance) and $2,500 deployed a weekend.

Profile	Venue	Shape	2026	without best 2	weekends up	2025	without best 2	weekends up
QK T1	Polymarket	steady	+15,551	+2,063	9/16	+10,776	+3,613	14/23
QK T8	Polymarket	steady	+23,508	+3,806	9/16	+4,942	+959	12/23
HK T8	Polymarket	steady	+31,217	+841	9/16	+6,288	+1,014	12/23
HK T9	Polymarket	steady	+30,811	+641	9/16	+6,686	+1,413	12/23
HK T1	Polymarket	concentrated	+12,647	−313	7/16	+10,078	+2,583	11/23
HK T7	Polymarket	concentrated	+10,691	−2,154	7/16	+10,436	+2,942	11/23
HK T6	Polymarket	concentrated	+3,933	−6,070	6/16	+7,363	−2,080	10/23
QK T1	Kalshi	concentrated	+33,499	+7,092	10/16	+2,328	−10,077	8/24
HK T6	Kalshi	concentrated	+16,711	+3,455	8/16	+137	−10,469	7/24
HK T1	Kalshi	mixed	+24,715	+4,818	10/16	−1,013	−9,610	7/24
HK T9	Kalshi	mixed	+19,914	+4,648	9/16	−5,961	−8,539	7/24
HK T8	Kalshi	mixed	+16,078	+2,082	7/16	−5,650	−8,598	7/24
QK T8	Kalshi	mixed	+11,985	+1,938	8/16	−5,283	−8,445	7/24
HK T7	Kalshi	mixed	+10,924	−1,915	8/16	−1,149	−6,975	8/24

Read these as rankings, not dollars. Three reasons:

1. Compounding makes the best weekends huge. HK T8 on Polymarket makes +31,217 in 2026 but only +841 without its two best weekends, so 97% of the season is two weekends. "Steady" still applies, because what's left is positive, but the label understates how lumpy Kelly is in dollars.
2. The replay has no depth limit. A $500 to $2,000 stake fills at the touch in the backtest. F1 books on Polymarket and Kalshi rarely show that much at one price.
3. Half-Kelly overbets an overconfident model. QK T1 beats HK T1 on Polymarket in both "without best 2" columns, and on Kalshi in 2026. That is the usual sign that the fair prices are sharper than they really are.

Quarter-Kelly T1 on Polymarket is the most defensible Kelly row. It is steady, has the best 2025 of any row (+10,776), and 14 of 23 2025 weekends are up. On Kalshi every Kelly row loses about $7,000 to $10,500 in 2025 once its two best weekends are removed.

Every market kind

Adding Kalshi's top 10 markets changes only Kalshi 2026: T1 goes from +2,332 to +3,969, with Monaco now its best weekend. 2025 is identical, so no 2025 top-10 market reached the replay. Polymarket lists no top 10 market, so its rows are identical. It's promising but rests on one season. It needs a 2025 top-10 tape before any profile uses it.

What this means for this weekend

This is a fantasy season for comparing strategies, so every row above stays in previews and pick lists, risky ones included. The notes below say what the record supports, not what to leave out.

• Polymarket: trade A (T1) as is. If sized off a bankroll, use quarter-Kelly with a cap the book can carry, not half-Kelly. T7 is a near twin of A (the podium coherence tolerance hardly matters). T8 and T9 earn more in 2026 but are flat in 2025 once the best two weekends are removed.
• Kalshi: trade A small and build the record. Kalshi is a big exchange; the problem is our evidence there, not its size. A's Kalshi totals are positive (+2,332 / +1,257) but come from a few long-shot weekends: only 5 of 24 weekends were up in 2025. Live weekends at a small flat stake are how we tell an edge from luck. Maker K is the steadiest Kalshi option. Kalshi has no Singapore markets linked yet, so linking comes first.
• Makers on Polymarket: C and K are steady but small, about +$500 to +$1,000 a season.

What this doesn't show

• No OG.com: there are only one or two weekends of data.
• No sprint or both-cars-in-points markets: these aren't modelled yet. No best-price-across-venues mode.
• One replicate per combo. The noise floor is the configured ±150, not measured from seed replicates.
• Fills at the touch with no depth model. The Kelly totals are the most exposed to this.
• Singapore (round 17): Kalshi has no linked markets yet, and our stored Polymarket tape shows zero 24-hour volume (Strategy preview). Check the venue's own volume before trading.

Sources

Built from data/runs/search/strategy-review/ranking.csv on the VM (f1 search-report at commit abac3e9, 2026-10-06 02:20Z). There is one saved sweep run per job in model_runs. A database backup was taken before the run (racinglines-before-strategy-review-*) and copied to the owner's Mac.