Cohorts

A cohort is the kind of bet a thesis represents. Cents tracks two:

• Directional — a single-leg position that takes a clear regime view (long X because tariffs, rates, AI capex…).
• Neutral — a paired-hedge construction (long X + short its sector ETF, sized 1:1 dollar) that cancels most of the regime exposure and isolates the company-specific signal.

Why have two

If you only run directional theses, you cannot tell whether your wins came from skill (picking the right name) or regime beta (the tape was friendly during the period you measured). The neutral cohort is the control group that makes the comparison meaningful.

Concretely: if the directional cohort returns 0.8 Sharpe over six months and the neutral cohort returns 0.7 Sharpe over the same window, the gap (0.1) is the value of taking regime bets, holding skill constant. The headline 0.8 is the wrong number to optimize against — you’d be measuring “lucky to be invested in 2026” plus “lucky in this specific name.” The gap separates those.

Without the neutral cohort: you’d never know if the gap is positive, zero, or negative. A negative gap is the most important finding you can have — your directional bets are actively destroying value — and you can only discover it with the control group running in parallel.

How it’s modeled

A neutral thesis carries:

• cohort = NEUTRAL (vs. DIRECTIONAL)
• hedge_symbol = "XLF" (the short leg)
• Two Position rows linked to the same thesis_id: one LONG on symbol, one SHORT on hedge_symbol (sides flip for bearish-signal paired theses — see Factory)
• The thesis’s spread P&L is just the sum of both legs’ P&L, since both are linked to the same thesis_id

Optionally, paired_thesis_id links to a “twin” directional thesis so you can do manual A/B comparisons.

CLI

# Create a neutral thesis directly
cents thesis create --symbol NVDA --title "..." \
  --cohort neutral --hedge-with SOXX

# Generate a paired twin of an existing directional thesis
cents thesis twin <thesis-id> --hedge-with SOXX

# Open both legs of a neutral thesis in one command
cents position open NVDA --thesis <id> \
  --size 100 --price 135 \
  --hedge-price 215     # short leg auto-sized to dollar-match

# Per-cohort P&L report
cents cohort

In the factory

The factory’s cohort_mode config decides what gets opened:

• paired (intended default) — every factory thesis is opened as neutral with a sector-ETF hedge
• directional_only — single-leg only (faster ramp, no control group)

Sector ETF lookup uses FMP’s sector classification mapped to the SPDR nine (XLK, XLF, XLE, XLV, XLI, XLY, XLP, XLU, XLB, XLRE, XLC), with SPY as the fallback when sector or FMP key isn’t available.

What it deliberately doesn’t do

• No beta-matched sizing — 1:1 dollar across legs (followup; TSLA’s beta vs XLY is the canonical mismatch case)
• No correlation-based peer lookup — --hedge-with is always user-supplied (or sector-ETF in the factory)
• No spread-based scan triggers — close manually for now