Every Strategy declares a stop, and R is planned risk
The detectable floor is declared in net R (ADR-0004 context: see the floor decision), which requires the risk unit to be unambiguous. We fix **R = config.risk_per_trade, a constant**, with position size derived as R / stop_distance — the donor's sealed convention (riskUsd = 175.0, _sized(stop_dist, params, costs)). **A StrategySpec without a stop fails to compile.**
Considered options
- Volatility fallback (ATR x size) where no stop exists. Rejected: it is a unit, not a bound. Losses stay unbounded per position, which is unacceptable when losses are not capped by an entry fee, and the entry-obtainability gap auditor would have nothing to audit.
- Money for stopless specs. Rejected: reintroduces the stranded-at-one-capital-level problem and splits verdicts into two incomparable kinds.
Consequences
Regime-conditional strategies — the first proposer — carry a disaster stop alongside their signal exit. The signal normally fires first; the stop exists to define R and bound the loss. The disaster stop's width is therefore a declared parameter subject to the same registration discipline as any other, and must not be tuned after the fact.
R is planned, not realised. A position that gaps through its stop is recorded honestly as (say) -2.7R rather than clipped to -1R. On equities this is the difference between a truthful loss distribution and a flattering one.