What closes our trades?

Separate entry quality from exit quality. Look for net expectancy and payout—not win rate alone.

No validated regime selected. None of the 168 configurations passed positive validation net plus five trades. No exit preset is promoted; the submitted controller and risk limits are unchanged.

Execution parity — release blockers reproduced

Actual submitted controller, pinned Hummingbot PositionExecutor, and order/event models. Exchange acceptance, private account state, fees and fills are modeled. Condor model/MCP and private Derive transport are not certified by this test.

Loading hash-verified execution evidence…

NO-GO: seven unsafe scenarios, five failure patterns. These are injected offline fault reproductions—not actual live losses. Green regression checks mean the failure is reproducible, not fixed.

Lifecycle patterns requiring a fix and a new recovery test; residuals are modeled notional, not realized loss
Failure pattern	Scenarios	Residual range	Next recovery test

Inspect an offline lifecycle / guard fixture

PASS means this fixture assertion held, not necessarily a flat account. Guard cases can intentionally retain known or unknown exposure. Risk-latch tests inject equity; that is not trading P&L.

Independent ledger versus real executor tracking at each event
Event	UTC	Ledger position	Executor status / close	Tracked entry fill	Tracked close fill	Tracked to close	Retries / held fills	Working orders	Cash / equity

Modeled order requests and fill fragments

Recorded native books: one shared $800 account

Actual controller decisions during this short capture; not a two-day execution backtest
Market	Gate / halt reason	Ticks

48-hour candle cost sensitivity — separate from executor parity

Four independent $800 market cases per source/scenario, not a shared portfolio. Old assumptions: 3bp fees + 3bp slippage per side. Submitted-bound stress: 6bp fees + 15bp slippage per side. This stresses the bound on both sides; actual entries use observed depth. Normal/restricted target-to-cost multiples remain 3×/4×.

Fixed signal/risk candle replay with modeled costs; zero trades is not a profitable strategy
Source	Cost scenario	Cases	Closes	Net sum	Turnover sum

Download sanitized execution evidence and all traces

Loading hash-audited exit research…

1. What is real, derived or simulated?

Prices and market observations are real. Strategy fills, crossing priority, slippage, fees and funding are modeled. This is not live P&L or an options-chain backtest.

Data surface and feature definitions
Surface / feature	Evidence type	Construction	Limit

2. Do different exits retain more profit?

Evidence source

Candle timeframe

Account market

Independent $800 starting accounts. Dollar sums are experiments—not one portfolio. Sources overlap in time and must not be added. ETH/BTC can remain empty because venue minimums exceed the unchanged $160 cap.

Fixed-entry, fixed-risk exit comparisons; all figures after modeled costs
Exit preset	Cases	Net sum	Wins / losses	Avg win	Avg loss	Payout ratio	Profit factor	Expectancy / trade	Avg hold range, min	Trades / account / day	Turnover sum

Six exit definitions and timing limitations

R is initial price stop distance, not the $4 account risk ceiling. Intrabar holds are intervals, not exact fill clocks. If TP and SL touch in one bar, SL wins. Trailing levels use only prior closed bars; no same-bar high lookahead. HB-net approximates the pinned market barrier logic, not actual asynchronous polling or fills.

Inspect one account and its closes

Replay account / exit preset

Exit frequency and net outcome for this account only
Exit reason	Closes	Share of closes	Wins / losses	Net

Each modeled close; known MFE excludes exit-bar extremes, so it is not a full tick-path maximum
Side	Entry UTC	Exit timing	Hold, min	Notional	Qty	SL / TP, %	Reason	Gross	Fees + funding drag	Net	Known MFE / MAE, %	Entry trend / efficiency / volume

3. Which candle regime is strongest?

Same evaluation dates and risk profile across 5m / 15m / 1h / 4h, with their real warmup histories. All periods were previously inspected: this is exploratory evidence, not an untouched holdout.

Best active validation configuration per timeframe, ranked by return minus drawdown penalty. Sparse or losing results are not an edge.
Timeframe	Exit / regime	Trades	Net sum	Status

Validation regime filter

Efficient trend: |z| ≥ 1.5 and efficiency ≥ 0.5. Participation: volume ratio ≥ 1.5. Low/high ATR: ≤ / > 0.3%. Calm/elevated CESF: < / ≥ 0.5. Gates affect entries only. Cash/no-trade rows remain explicit, not winning strategies.

Validation only, using the candle timeframe selected above. Eligibility requires ≥5 trades and positive net.
Configuration	Trades	Net	Objective	Payout	Eligible?

4. Notional is the exposure—not the leverage field

The Derive connector does not set venue leverage. The submitted field is 2, but its pinned setter returns success without setting leverage. $160 / $800 is 0.20× account exposure. We have not verified private margin requirements.

Notional $ (0–160)

Entry price $

Underlying move % (−100–100)

Illustrative leverage only

Gross P&L = quantity × price change × side. Changing leverage with fixed notional does not multiply P&L again. Notional / leverage below is a textbook illustration, not the Derive margin requirement, liquidation buffer or a trading control. No risk budget is changed.

Older runs: audit without pooling cohorts

One original cohort / source / timeframe / policy

Original recorded closes; old bar-clock holding estimates can span a full candle
Reason	Closes	Wins / losses	Net

What is worth testing next?

Separate entry eligibility from position invalidation, reconcile replay barriers with Hummingbot net-P&L semantics, and collect a longer native contract-price/book archive. Runner exits are hypotheses: proxy improvements did not transfer to native Derive. Do not increase exposure to compensate for negative expectancy.

Download sanitized exit and regime evidence

Read-only research · no order placement · no options execution claim · submission 5f68255 unchanged.