Every order carries a client ID derived from the signal, the intent, the quantity and the day, so the same decision always produces the same ID. A retry after a timeout therefore reaches the broker as the same order rather than a second one, and an order whose outcome cannot be established halts new trading instead of being guessed at.

The backtester calls the production evaluate_entry at every simulated fill rather than a mirror of its arithmetic. A copy would have drifted within a quarter, and the backtest would then be measuring rules the live system no longer had. Two golden fixtures fail the build if a refactor changes what either path would do.

Twice every weekday the system compares what the broker holds with what its own database says it holds — positions, quantities and the resting exit orders that protect them. A discrepancy it cannot explain halts new entries and says why. It will adopt the broker's version of an order, but it never rewrites the quantity it recorded when the position was opened.

An unhandled error in a scheduled trading run — the nightly screen, the five daily manage runs, the two reconciles — marks that run failed and halts new entries, rather than carrying on with a partial picture. Exits are deliberately never blocked by a halt: whatever went wrong, a position that needs to close still closes — and if the broker refuses the sell, the protective stop is put back the same run and the refusal is reported as a failure, never as an exit.

Each lane screens after the close, records its signal with the same frozen evidence a live signal gets, and fills and exits against a simulated book sized by the live risk engine — so a lane's record means the same thing as a backtest of the same rules, and it is written before the outcome is known. The code path that runs a lane cannot construct an order; a test reads its imports and fails if that ever changes. The cohort includes a control that replays the live signals, single-variable variants of the live strategy, three different return sources, and two options overlays — a protective put and a covered call — priced against the options market with spreads charged in full. Every hypothesis is written into the experiment ledger before its lane records a single night.

Try enough ideas and the best-looking one is selected noise. The Deflated Sharpe Ratio is computed across the whole experiment ledger — each backtest, each walk-forward, each lane — so the bar rises with each attempt and can never be lowered by forgetting one. Factor attribution regresses each result on market, size, value and momentum, so exposure an index fund sells for a few basis points is never mistaken for skill. The current answer, published above: nothing clears the bar.

Parameters are chosen on a rolling training window and judged only on the months that follow it, stitched into one out-of-sample record that counts as a single trial. Separately, three regime-diverse years are locked: the backtester and the walk-forward harness both refuse any run that touches them, and the one permitted evaluation is recorded before it runs and can never be un-spent. Transaction costs are an explicit input, defaulting to zero so that a cost-free figure is always labelled as the upper bound it is.

A signal is stored with the indicators that produced it, frozen at that moment by a database trigger, and the dashboard explains each one against the threshold the screen required. The operator's own agree/disagree verdicts count only when written before the trade resolves, and are then scored against what actually happened. Each signal's page also asks the options market whether its target is realistic — how many expected moves away it sits, what a put at the stop would cost — as measurement, never as an instruction to trade.