Kill Criteria
reout · Systematic BTC
These thresholds apply from record inception (Methodology, section 1) and are anchored before it, in the pre-registration manifest. They are not applied to the pilot. This document is the public excerpt of reout's internal document, whose hash is included in the same manifest. The thresholds are not modified while the account is in a drawdown. Each monthly report states the account's current position against them.
1Scope and principle
Exits of individual positions (stop-loss, take-profit, time limit) are part of the strategy and are applied by the system to every trade. This document concerns the account as a whole.
P&L thresholds are alerts and review triggers, not automatic stops. A stop triggered by P&L alone cannot tell a normal losing run from decay (section 2), and, once triggered, it removes the recovery that has historically followed a drawdown. Tail risk is handled by position sizing (Methodology, section 1), not by stopping.
Automatic halts are reserved for system faults, that is, evidence that the system is not operating as designed. Fault signals have a low false-alarm rate and a clear remedy. Decay of the strategy itself, meaning that market behaviour no longer resembles the history on which it was built, cannot be told apart from a normal losing run by P&L alone within any useful time. The thresholds of section 3 therefore trigger review rather than a stop, and that judgement is made in the review and in the scheduled renewals of the program (Methodology, section 12).
2Pre-declared ranges
The thresholds in section 3 are the ranges the program is expected to stay within. They were derived from the program's backtest before the record began, are anchored in the pre-registration manifest and are not changed while the account is in a drawdown. They bound what the program is expected to survive; they are not a forecast of results. Results within them are normal and are not evidence of failure; results beyond them are reviewed as set out in section 4.
3Thresholds
Each threshold is derived from the program's backtest and fixed before the record began (section 2). No threshold stops trading by itself.
| Metric | Review (alert) | Caution (investigation) | Deep review (full re-validation) |
|---|---|---|---|
| Account drawdown | −37% | −41% | −52% |
| Trailing 30-day return | −25% | −37% | −46% |
| Trailing 65-day return | −22% | −30% | −38% |
| Days underwater | 187 | 199 | 249 |
All thresholds refer to the live account at three times notional. Account drawdown is the current drawdown and days underwater are the days since the last high, both measured from daily valuation points on the index of daily time-weighted returns, so that an external transfer is neither a loss nor a recovery. The trailing 30-day and 65-day returns are time-weighted returns of the account over the trailing window on the same daily valuation points (Methodology, sections 4 and 6). Assessment is made once a day at the daily valuation point.
4Review and outcomes
Review. Within 24 hours reout (1) decomposes live results against expected behaviour, (2) examines the market regime and (3) checks data and infrastructure. The finding is recorded in the monthly report. Trading is suspended only if a fault under section 5 is confirmed.
Caution. The same review, and in addition new positions may be sized at one half while the account's drawdown exceeds the caution threshold. The rule is optional and its use is logged (Methodology, section 1). Existing positions are not closed early on account of P&L.
Deep review. A full re-validation of the program, ending in one of three outcomes announced in the monthly report: the program continues unchanged; it continues at reduced size; or it is retired, which closes the record (Methodology, section 12). Retirement is decided on the findings of the review, not on the level of P&L that triggered it.
5Automatic halts: system faults only
The system stops opening new positions, while continuing to manage existing positions under its rules, on evidence that it is not operating as designed: an integrity check failing at start-up, a required data feed going stale, the live process deviating from its reference behaviour, or execution deviating from what was modelled. The specific signals and their thresholds are internal. A manual halt is available. It blocks new entries only and does not close positions.
6Prohibited actions
- Editing a threshold in place, or treating a current episode as an exception.
- Increasing leverage during a loss.
- Closing existing positions on discretion, an action the system never takes.
- Stopping the strategy on P&L alone, contrary to section 1.
- Changing the deployed version in response to a loss, or resetting any metric in this document on a change of version (Methodology, section 12).
This page renders KILL_CRITERIA.md at track commit cbdaf12, built by d2a944c of the site. What is anchored is the file, not this page.
SHA-256 de912deff7b25c49f65adede2f4848a31995eca92b8d4762d0cb5845b443951a