Service Status

Every run_* process is live or completed in its expected sub-daily or business-day window.

In plain English: Like checking every machine on the assembly line ran on time today. If a scheduled job didn't finish when it should have, something is stuck and needs a look.

Target: 100%  ·  Actual: 10/11 (90.9%)

Fraction of sampled instruments with a price update in the last 26 hours, excluding instruments with no up-to-date data (retired/stale, or feeds confirmed dead at source).

In plain English: Are we getting fresh prices for the things we trade? This is the share of markets whose prices we refreshed recently — trading on stale prices is like driving with a fogged-up windscreen.

Target: 99%  ·  Actual: 252/255 (98.8%)

Fraction of FX codes with an update in the last 26 hours.

In plain English: The same freshness check, but for currency exchange rates. We need up-to-date rates to correctly value positions we hold in foreign currencies.

Target: 99%  ·  Actual: 12/12 (100.0%)

Contract orders unfilled, older than 2.0h, while their market is open.

In plain English: Orders we sent to the broker that are still sitting unfilled while the market is open — like a letter we mailed that never got delivered. This should be zero.

Target: 0  ·  Actual: 0

Every active strategy has an optimal-position update since the start of the previous business day (evening producers, morning snapshot).

In plain English: Did every strategy do its homework and recalculate what it wants to hold? If one didn't, it's running on yesterday's (or older) plan instead of today's.

Target: 100%  ·  Actual: 1/1 (100.0%)

Each active strategy's trailing realised annualised vol (63 business days) is within [0.6x, 1.5x] of its percentage_vol_target, and its ex-ante expected risk stays below 1.5x target (overshoot guard).

In plain English: Are we taking about the amount of risk we aimed for — not too sleepy, not too wild? This checks our actual ups-and-downs stay close to target, and that we're not set up to take far more risk than intended.

Target: 100%  ·  Actual: 1/1 (100.0%)

Instruments feeding the optimiser a non-zero position target whose value is more than 5 days old. Keyed on the age of the value, not the age of the last write.

In plain English: Every night we work out how much of each market we ought to own. This checks that those numbers are actually being recalculated, rather than an old one being copied forward and traded on. A number can be rewritten every day and still be months out of date.

Target: 0  ·  Actual: 0/174 (0.0%)

Mean ratio of realised bid/ask half-spread to currently configured spread across instruments traded in the last 30 days. 1.0 = on budget; >1.0 = paying more than expected. Informational while threshold is calibrated.

In plain English: Every time we buy or sell we pay a small 'toll' — the gap between the buy and sell price. This compares the toll we actually paid recently to what we budgeted: 1.0 means on budget, above 1.0 means we're paying more than expected.

Target: informational (calibrating)  ·  Actual: 44/587 (127.3%)

Mean ratio of realised bid/ask half-spread to currently configured spread across instruments traded in the last 90 days. 1.0 = on budget; >1.0 = paying more than expected. Informational while threshold is calibrated.

In plain English: Same trading-toll check as the 30-day version, but over a longer three-month window so a few unusual days don't dominate. 1.0 means on budget, above 1.0 means we're paying more than expected.

Target: informational (calibrating)  ·  Actual: 82/587 (436.3%)

Count of instruments whose realised bid/ask half-spread is more than 2.0x the currently configured spread over the last 30 days. Informational while threshold is calibrated.

In plain English: How many markets charged us a much bigger trading toll than we'd budgeted for. A high count means our cost assumptions are off for several instruments and need refreshing.

Target: informational (calibrating)  ·  Actual: 8/44 (800.0%)

Cumulative (realised − theoretical) PnL over the last 30 days, divided by absolute cumulative theoretical PnL. Theoretical PnL is computed from the most recently recorded optimised position per (strategy, instrument) carried through subsequent business days, multiplied by back-adjusted price changes and base-currency point size. Realised PnL is the daily change in accumulated capital. Captures execution drift (slippage, integer rounding, fill timing). Informational while threshold is calibrated.

In plain English: We have a 'perfect world' simulation of what our trades should have earned, and what they actually earned in real life. This measures the gap — caused by things like slippage and rounding. A small gap means real trading is tracking the plan.

Target: informational (calibrating)  ·  Actual: 538.71/14023.65 (3.8%)

Realised max drawdown depth at each of 1-month, 3-month, and 1-year trailing windows, expressed as a ratio to a reference depth. Value is the worst (most adverse) ratio across horizons. <1.0 = inside expectations; >1.0 = at least one horizon is deeper than reference. Reference depth at each horizon is the 95-th percentile of historical max-drawdown-in-window values computed from the system's own pnl history; horizons with fewer than 20 sample windows fall back to conservative hardcoded depths so the SLI never goes blank. Informational while threshold is calibrated.

In plain English: A 'drawdown' is how far we've fallen from a recent high — the size of a losing streak. This checks whether our worst recent losing streaks are normal for a system like ours, or unusually deep. Below 1.0 means within expectations.

Target: informational (calibrating)  ·  Actual: 0.0623/0.0847 (73.5%)

For each instrument: sum of |fill quantity| across all strategies in the last 30 days, divided by recent average daily volume of the currently-priced contract times 30. Reports the maximum across instruments as a raw fraction (0.005 = 0.5% of ADV at the worst instrument). value_numerator counts instruments above the 1% capacity ceiling; value_denominator is the count of instruments measured. Informational while threshold is calibrated.

In plain English: How big are our trades compared to how much that market trades each day? Small means we're a tiny customer who can get in and out easily; big means we're such a large share of the market that trading is slow and expensive. Lower is safer.

Target: informational (calibrating)  ·  Actual: 1/46 (2.0%)

For each (strategy, trading rule, instrument), the mean of |capped forecast| over the trailing 250 business days, compared to the configured average_absolute_forecast scalar (default 10.0). value reports the worst deviation fraction across all pairs; value_numerator counts pairs outside the ±25% band. Carver-framework scalars target 10; sustained drift outside the band means positions are systematically over- or under-sized vs design. Informational while threshold is calibrated.

In plain English: Each strategy turns its market views into a 'conviction score' that should average around 10. This checks the scores are sized right — if they drift too high or low, our positions end up systematically too big or too small versus the design.

Target: informational (calibrating)  ·  Actual: 3198/5831 (100.0%)

For each (strategy, trading rule, instrument), the annualised Sharpe of the rule's standalone PnL contribution over the trailing 250 business days. A pair is deviated when trailing Sharpe falls below the 5th-percentile of the rule's historical 250-BD rolling Sharpe distribution; pairs with too little history (< 50 distinct sample windows) fall back to a fixed-threshold delta of 0.50 Sharpe units below the full-period mean. value reports the most adverse delta (most negative trailing − full); value_numerator counts deviated pairs. Informational while threshold is calibrated.

In plain English: Each trading idea ('rule') is graded on its recent risk-adjusted profit. This flags any rule performing much worse lately than it normally does — an early warning that a strategy may be breaking down.

Target: informational (calibrating)  ·  Actual: 363/5831 (-2563.4%)

The scheduled jobs installed on the production host, and each armed account's process windows, compared against the schedule generated from the client registry. 1 means they disagree: a task that never runs, or a window shut against its own cron time, stops an account trading without crashing. A commented-out approved line counts as drift like any other missing one; there is no disableable task carve-out.

In plain English: Every account has its own scheduled daily jobs. This checks that the schedule actually installed on the machine still matches the one the system says it should be running. If they disagree, an account can quietly stop trading with nothing crashing to say so.

Target: 0  ·  Actual: 0