Guardrails

Stage 3 collects the limits every result must respect. All of them are optional, and the stage says so up front: "Optional safety limits. Leave a field blank when you do not want that rule." An empty field means the constraint does not exist — Hunter never assumes a limit you didn't set.

Historical worst-day cap (% of BP)

Reads every day of the reference window and rejects any portfolio whose single worst combined day — all picks together, contract-weighted — exceeds this share of the budget. From the help text: "with $100,000 buying power and 2%, the worst historical day may lose at most $2,000." This is a hard constraint on the actual daily series, not on an average.

Simultaneous-stop cap (% of BP)

Caps the theoretical worst day — if every strategy hits its stop at once — as a share of the budget. Loss when a spread stops equals premium × stop-multiple, so this bounds Σ(premium × stop-loss × contracts) across the picks. It is structural and regime-independent: computed from the strategy definitions alone, no daily series involved. The product describes it as "intentionally more conservative than the historical worst day" — the historical cap bounds what did happen; this bounds what could, with one caveat that matters: it assumes stops fill. In a gap or fast dislocation, stops can fail and losses approach the full spread width.

Combined DD cap (% of BP)

Caps the true combined-curve max drawdown — the worst peak-to-trough of the whole portfolio's daily equity over the reference window, read from the picks' real daily series, not an average of per-strategy percentages — at your chosen share of the budget.

What it cannot do is bind the future. The cap applies to the reference window the search optimizes — it is an in-sample shaping constraint, not a forward limit, and realized drawdowns after selection can exceed it. Our 2026-07 walk-forward study measured exactly this: one low-DD configuration replayed over 3 years exceeded its 1.5% cap in all 11 quarters tested, with a stitched max drawdown about 6.3× the cap. Tightening the cap is still useful — it makes the optimizer shed exposure and bring in wider, less-correlated spreads — but the number itself describes the past window, nothing more.

Max pairwise correlation

Forbids selecting any two strategies whose reference-window daily-P/L correlation exceeds your threshold. From the help text: "at 0.60, a pair with 0.72 correlation is rejected." Its practical value is killing near-duplicates — two 10:05 50-wides fifty cents apart are one bet, not two. Lower = stricter. Like the combined-DD cap, this reads daily series and adds a moment to the run.

Max trades in the last hour

Limits the share of the portfolio entering at or after 15:00 ET: "at 30%, a 10-strategy portfolio may contain at most 3 last-hour strategies. Contract counts do not change this percentage." Late entries have little session left to recover from a move against them; this bounds how much of the book lives there.

Max concentration in any rolling 60 minutes

The anti-clustering rule, with a mode toggle — the two modes are mutually exclusive:

• Risk % — no moving 60-minute window of entry times may contain more than this share of the portfolio's total stop-risk dollars. Risk-based: a window holding one large-premium strategy can bind before a window holding three small ones.
• # Strategies — no moving 60-minute window may contain more than this count of strategies, regardless of size: "at 3, no one-hour group may contain more than three strategies. Contract counts do not change this limit."

The window rolls — it is not fixed clock hours. Entries at 10:31, 10:59 and 11:29 all sit inside one 60-minute window starting 10:31, so clustering can't hide on an hour boundary.

Advanced limits

Collapsed by default, for portfolio-shape rules most runs don't need. Four concentration caps, each a whole number of strategies (1–40) sharing one exact value:

Below them, Minimum strategies per selected family shows one field per family present in your universe: "entering 2 requires at least two MEIC strategies; leave it blank to allow any number, including zero." This is how you guarantee a mix — for example forcing METF trend-followers into a book the optimizer would otherwise fill with MEIC alone.

Reading this stage as a whole

Each guardrail states its own limits: the worst-day cap describes the past window, the simultaneous-stop cap assumes stops fill, the combined-DD cap shapes the in-sample curve without promising anything forward. Constraints shape what the search returns; they are not promises about what the market will do to the result.

Related pages

• Match & Size
• Results & history
• Monte Carlo stress testing