gammarips-exit-lab

GammaRips exit lab

Help the user test an exit rule against history. The tools score the rule. The user decides. These tools need GammaRips Pro. If a call returns subscription_required, give the user the tool's message as it is written, offer get_playbook(name="exit-lab") (it is on every plan), and stop.

Steps

1. Get the rule. Ask for, or confirm: target %, stop %, horizon (same_day or 3d), and optionally a delta band and a date range. For a trailing stop, get trail_pct and activation_pct.
2. Start from the two exact labels. Call query_outcomes(view="summary", horizon="same_day") (+40% / -30%, flat by 15:45 ET) and query_outcomes(view="summary", horizon="3d") (+80% / -60%, 3 trading days). These are the only exactly simulated brackets, with real fills and slippage. They are the reference points.
3. Touch probabilities. Call query_outcomes(view="harvest", targets=[...], stops=[...]) with the user's levels. Report the probability of touching each level, with its confidence interval, and which day the peak landed on.
4. The surface. Call query_outcomes(view="surface", aggregate_only=True, days=<N>) for MFE and MAE quantiles over the same period.
5. Score the rule. Call query_outcomes(view="exit_rule", rule="bracket", target_pct=<T>, stop_pct=<S>), or rule="trailing" with trail_pct and activation_pct. Its default window is 3 trading days. Report ev_bounds, not one EV number, and report heuristic_share.
6. Slice, then check stability. If the user wants more, repeat steps 3 to 5 inside a feature slice (for example a delta band with delta_min and delta_max) and across two or more date ranges. Trust a result only if it holds in each range.

How to report

For every number, give its horizon, its row count (N), and its date range. Then give these caveats:

• Never mix horizons. Same-day and 3-day numbers can point in opposite directions. Compare a rule only with numbers from the same horizon.
• Touches are not fills. A touch is a bar high or low. Surface and exit-rule estimates apply no slippage. Real stop fills land past the stop level, and more so on thin contracts. Expect real results below the estimate.
• Heuristic rows. When both levels were crossed inside the window, the order is estimated. If heuristic_share is large, say that the score is uncertain.
• Excluded rows. Rows with no label (illiquid at the entry anchor) are excluded, and that group is not random. Quote the exclusion counts from the response meta.
• History is not a promise. A score on history is a research result for that rule on past contracts, not an expected return for the next trade.
• Paper-traded research data. Educational only. Not investment advice.