Every setup is a claim about where the money comes from. The map draws
the whole territory; the ranking scores each family as a game; the crowding chart shows
where the players actually are. Ratings are v2 — adversarially revised: two independent
adversaries attacked every cell, a third ran the convergence check, and the crowding axis is the
median of three blind raters.
ROOT — who is on the other side of your P&L?
Someone paying for direction they lack (1) · a related contract mispriced against yours (2) ·
a holder paying a recurring toll (3) · someone mispricing future movement (4) ·
an impatient taker crossing the spread (5) · a forced convergence (6).
The map
Thirty-two setup families around one root. Node color is the game score —
darker (light mode) / brighter (dark mode) is a better game. Hover any node
for its P·R·A·C breakdown.
< 2.6 poor game2.6–2.92.9–3.153.15–3.45≥ 3.45 best games
The ranking — which game to playV2 · POST-ADVERSARIAL
Score = 0.5·P + 0.3·R + 0.2·A (each 1–5).
P — probability a dedicated non-institutional trader who plays this family ends up with real
positive expectancy after costs, unconditional (if the door is shut, P is low — the round-1
adversaries killed the earlier conditional scoring that let gated games rank on edge no solo can
touch). R — payoff shape. A — accessibility. C — crowding (5 = saturated
relative to capacity, 1 = deserted); shown, deliberately not in the score — it is the
second axis, charted below.
#
Game
Score
Σ
P
R
A
C
Adversarial round changes: funding capture demoted from co-#1 (P 4.5→3.5 on
post-2024 institutionalization, R 2.5→2 on the exchange-ruin tail); market making, latency, and
replication arb collapsed to the bottom (unconditional-P fix); trend following P 4→3.5 (behavioral
survival haircut); basis and roll-yield trimmed (ETF-era compression, carry-as-tail-risk); long-vol
R 4.5→3.5 (retail overpays for wings); breakouts R 3→3.5 (same cut-losses shape as trend); macro
A 2→3.5 (micros make it accessible — the barrier is informational and already priced in P). Two
families the reviewers found missing were added: session & seasonal timing, and
liquidation-engine participation.
Odds × crowding — where the good games are still empty
The second axis you asked for: vertical = P (your odds of real edge),
horizontal = C (how many players are already there). The prize is the upper-left — decent
odds, few players. Hover any dot for detail.
The decision tree — reaching a top-5 game
The same six counterparties, read as a decision: start at the root, choose who pays your edge, follow the branch. Only three branches carry a top-5 family (solid path, gold-ringed leaf); the other three are pruned — their best family never clears the cut (dashed).
Root / counterparty Branch — a decision Game — shaded by score Top-5 rank path to a top-5 game pruned branch
The read: every path to a top-5 game runs through Direction, Relative value, or Carry & income — the counterparties who pay a structural toll (a persistent trend, a mispriced relative, a recurring carry). Volatility, Microstructure and Arbitrage are pruned: post-2024 a solo trader there is usually the food, so even their best family stalls below the five.
The verdict — v2, after the adversarial loop
Best game outright: systematic trend following (3.55) — the only family whose score
survived both adversaries near-intact. Positive skew, century-scale evidence, moderate
crowding against vast capacity. Its price is behavioral: years of chop, every signal taken.
The prize quadrant (P ≥ 3, C ≤ 3):positioning extremes (the headliner — public
data, confirmed marker in your own registry, and still only moderately used), trend following,
crypto cross-section RV, gap fills, calendar spreads, processing
spreads, and vol relative value. Decent odds, and the crowd hasn't arrived.
The demotion that matters: funding capture fell from co-#1 to ~#5. The odds are still
good — but C = 5: Ethena-scale vaults and ETF-era basis desks industrialized the toll booth
after 2024, and its tail is loss-of-entire-stack. Good game, long queue, ruin-shaped risk.
The trap, unchanged: the most crowded corners with the worst odds — structure
frameworks, breakouts, scheduled-event scalping, exhaustion fades (still rank-last among
directional games). Popularity is a cost. The uncrowded-but-unproven curiosities — long
volatility, session & seasonal timing — are empty for a reason worth testing, not assuming.
The full key
1Directionnet long or short — paid when price goes your way
1.1 Continuation — the move is real; join it
Breakouts
Range / Donchian breakout — close beyond an N-bar extreme
ADL & socialized-loss awareness — position for, or defend against, the clawback mechanics
6Arbitragea mispricing with forced convergence — near-riskless when pure
6.1 Replication arbitrage — the future vs what it's made of
Index arb — futures vs the replicating basket or ETF (ES vs SPY)
FX futures vs spot-forward parity — covered-interest deviations
Single-name futures vs the underlying
6.2 Convergence & cross-venue
Expiry-basis convergence — basis must die at settlement
Perp / dated / spot triangle — three prices of one asset kept honest
Cross-exchange gaps — same contract, different venue, regional premia
DEX-perp vs CEX-perp — on-chain vs off-chain funding and price
Funding-vs-borrow arb — perp funding against the spot borrow rate
The multipliers — axes, not branches
Holding period
sub-second → scalp → intraday → swing → position → cycle. Every leaf exists at several of these.
Decision engine
discretionary read → rules-assisted → fully systematic → ML ensemble.
Market family
rates · equity index · FX · energy · metals · agriculture · crypto dated · crypto perpetual (funding replaces expiry).
Risk posture
outright · spread-hedged · delta-neutral · market-neutral book. The same idea, four ways to wear it.
The standing rule
A leaf on this tree is a hypothesis, not an edge. Your own record: every setup that was
measured properly either died to a null or shrank to a risk marker — and the ones that survived
did so only after vol matching, era splits, and cost accounting. Before trading any leaf: name the
counterparty, name the null, pre-register the look, and price the sweep.