|P(Bull) − P(Bear)| must exceed this

3-step-ahead signal must agree and exceed this

Signal decay must remain useful for at least this many bars

At least this many of the tracked instruments must agree

7 combos × N instruments — roughly 60–90s on startup

Leave blank to use General default

Maximum number of instruments shown in the Markov card grid and the Execution Engine live signals table. Computation always runs for all available instruments — only display is capped. Range: 1–50. Default: 10.

How many cards show the complete suite (Transition Matrix + HMM + Multi-step + Parameter Sweep). Cards beyond this count show a compact signal-only view to reduce page weight. Should be ≤ Max instruments. Default: 10.

Fraction of total capital per instrument

Aggressiveness multiplier on the risk budget. Higher = more risk per trade across all signals. 1.0 = conservative, 3.0 = intermediate, 5.0 = aggressive. Works in combination with Base risk % below.

Maximum % of current equity to risk on a single trade at the strongest signal (sig=1.0, Kelly=1.0, horizon=8h). Kelly fraction and signal strength scale this proportionally downward for weaker signals. The hard stop % is NEVER part of account risk — it only converts the dollar risk budget into a position size (size = risk$ ÷ stop%). Default 0.5%: at K=3, strong signal → ~0.95% acct risk per trade.

Populates the three cap fields below with broker-appropriate defaults. Max Risk per Trade must be set manually — it is a portfolio management decision, not a broker rule.

PRIMARY cap — hard $ ceiling on loss if hard stop is hit on any single trade (size_usd × hard_stop_pct ≤ this value). Applied FIRST in all four engine modes (daily + hourly, fixed + compounding). Example: $2,500 means no trade can lose more than $2,500 at the stop. Leave blank for no cap.

SECONDARY cap — broker leverage gate. Applied after Cap B in all four engine modes. 100 = no leverage (Alpaca crypto — fully funded); 200 = 2× Reg T overnight (Alpaca equities / IBKR Pro); 2000 = 20× IG Tier 1 (5% margin). Leave blank for no cap.

NEW 2026-07-11 — QUATERNARY cap, hourly fixed-mode only. Minimum fraction of a desired position's own size that must be coverable by real available capital (after equity-based sizing and concurrent-position commitment, times the active broker's leverage) for the trade to be taken at all. Below this, skipped entirely rather than taking a token-sized position. 20 = default.

TERTIARY cap — absolute dollar backstop. Compounding mode only. Applied last. Example: IBKR $100,000 precautionary default. Leave blank for no cap.

Portfolio-level cap on simultaneous open positions

On (default) = Kelly fraction sizing above, completely unchanged. Off = the simpler Risk-% model below: size = risk-per-trade % × signal-strength multiplier, with a hard ceiling you set directly.

Halt new entries if portfolio drawdown exceeds this

Resume new entries once drawdown recovers to this level

Session open — when VWAP anchor resets

Skip intraday entry if no fill by this time

For instruments combining OPEN with a zone method (POC/VWAP): if no zone touch by this time, fall back to an OPEN_FALLBACK entry rather than waiting until the entry cutoff with no trade. Must stay earlier than the entry cutoff above. Shared by backtest and live trading.

Entry bar volume must exceed N-bar average

Used as the default on the Backtest page (overridable per-run)

Controls the scanner's automatic scheduled runs (pre-market scan, gap scan, intraday monitor) — the scanner doesn't place any real trades, purely evaluation and ranking, so this exists to fully stop that activity when not needed. A manual "Run Scan" from the Scanner page still works even with this off, since that's a deliberate action, not the automatic run this setting controls.

Master on/off. When disabled no new entries are placed. Existing positions and resting stop orders are not affected.

Hard-locked by config.LIVE_TRADING_PAPER_ONLY. Set to False in config.py to allow live order routing. Cannot be changed here for safety.

Block all new entries when portfolio drawdown from peak exceeds this percentage. Does not close existing positions.

Maximum number of open positions across all instruments at any time.

Maximum entry attempts per ticker, per scheduler tick. Guards against duplicate fills when overlapping scheduler runs both evaluate the same signal before the first order's fill is reflected in the broker's position snapshot. Default 1 — one trade per valid signal.

Price move % at which the hard stop triggers. Equity default; override per-instrument for crypto.

How often the trailing-stop activation job checks price and swaps the hard stop for a trailing stop. Confirmed default 15s — fast-moving instruments (e.g. QQQ) can reverse within minutes; 5 minutes was too coarse. Takes effect immediately, no restart needed.

Real, calendar-day lookback for the ORB regime state (Narrow/Middle/Wide) calibration — converted to a real bar count from each instrument's own actual bar density (a 24/7 crypto instrument and a limited-hours index have genuinely different densities for the same real day count). Confirmed default 14 — tested directly against real cached bars; extending to 30 days showed no meaningful improvement during a genuine, sustained trend, since the window still contained no larger prior move to compare against.

Real, direct percentile bands of the rolling abs(EMA20-EMA200) distribution itself — Narrow below the low percentile, Wide at/above the high percentile, Middle between. Replaces the original thirds-of-a-single-peak formula, which had no such guarantee and collapsed to "Wide" during a sustained trend. Confirmed default 10/90 — tested directly against 100 real recorded breakout candidates, correctly compressed "Wide" to a genuine minority state (13-18%) over the full 60-day history while still honestly reflecting a real, extreme trend when one is actually happening.

Real, independent, optional ceiling on the upper (Wide) band — effective upper bound = min(percentile-high value, rolling max × this/100). At the honest default of 100%, this never binds (a percentile can never exceed the true max) — confirmed, verified numerically as an exact no-op at this default. Lowering it below 100% pulls the Wide threshold tighter than the percentile alone suggests, as a genuine, separate calibration lever.

When on (default, matches the existing, current behavior), the ORB chart only marks a breakout once it has actually aligned with an active regime filter. When off, every real candidate breakout scanned that day gets its own marker, aligned or not — useful for auditing real breakout timing and timezone conversions independent of regime state. Display-only: never changes which breakout actually triggers a trade.

No new hourly-engine entries at or after this time. The last scheduled hourly slot leaves too little runway before the 15:55 EOD close job to safely manage a freshly-opened position. Existing open positions are unaffected — this only blocks new entries. Applies to Alpaca-routed equities only; crypto (24/7) is exempt, and this won't apply once a CFD broker is introduced.

Hourly engine only, per instrument. Before an entry is taken, checks whether the most recent OPEN/POC-entered trade for that instrument actually worked out: if it closed at a loss, or is still open but the most recent completed hourly bar moved against the original signal's direction (open > close = bearish, open < close = bullish), and the current signal is proposing the same direction as that failed attempt, its Signal T+1 is downgraded to the minimum value that still keeps it tradeable — never blocks a genuine reversal. Two such failures in a row for the same instrument skip that signal entirely for the current cycle only. Off by default — this changes real entry decisions once enabled.

Only relevant when "Allow risk assessment" above is also on. Prevents the first tradeable hourly signal of a new trading session from being compared against the previous session's last bar (e.g. NY open 9:30 ET's first signal shouldn't be judged against yesterday's close). Detected structurally, not by a fixed clock time — if the gap since an instrument's last recorded signal is longer than a normal hourly cycle, that signal is treated as the start of a fresh session and the comparison is skipped for it only. Off by default.

Only relevant when "Allow risk assessment" above is also on. When enabled, enough consecutive same-direction failures (see threshold below) skip the signal entirely for that cycle. When off, a signal is always downgraded at most — never fully skipped, regardless of how many consecutive failures exist. On by default, matching this whole feature's original behavior before this control existed.

Only relevant when "Allow full signal skipping" above is also on. How many consecutive same-direction failures trigger a full skip rather than just a downgrade.

Only relevant when "Allow risk assessment" above is also on. A downgraded trade's final, fully-sized position (after Kelly/Risk-% sizing and every cap) is multiplied by this percentage before the order is placed — e.g. 50 means the downgraded trade is taken at exactly half the size it would otherwise have been, regardless of the original signal's strength. REPLACES the platform's original minimum-signal-strength downgrade mechanism entirely.

When more instruments qualify to enter in one cycle than available capacity allows, which get priority? Confirmed via direct code review: today's actual behavior with nothing checked is arbitrary — whatever order instruments happen to be listed in the roster, first-come against the concurrent- position cap. Checking more than one uses each as a tiebreaker for the one above it, in this fixed top-to-bottom order — not a blended/weighted score. Applies to the hourly engine only.

Aggressiveness multiplier for Kelly sizing model.

Base account risk % at maximum Kelly signal strength.

Cap B — maximum dollar loss if stop is hit. Blank = no cap. Applies to all new instruments.

Master on/off. When disabled, no trades are relayed to any slave account regardless of individual slave settings — matches the same "master switch" pattern as Live Trading above.

How often mt5_daemon.py checks price and updates each position's trailing stop (a direct position-property modification, not a separate order). Confirmed default 15s. Re-read by the daemon before every check, so changes take effect on its next cycle — no daemon restart needed.