Record AR-EXP-066 · Execution Study
The strategy is not what
reaches your account.
Every framework is evaluated twice. Once on paper, where entries occur at the trigger price, stops are honoured exactly and exits happen at the moment the logic says they should. And once in the market, where none of that is true. This page measures the distance between those two evaluations, names each term in it, and states which terms are worth repairing first.
Interactive demonstration — not a historical backtest
Section 01
Eight places expectancy
goes missing.
A backtest reports one number per trade. A real trade produces at least six: the price you decided at, the price you were filled at, the worst price the position saw, the best price it saw, the price you exited at, and the total you paid to do any of it.
The gap between the first and the last is not a single cost. It is a chain of eight, and they are not the same size. Most operators spend their attention on the three smallest terms because those are the ones a broker invoices for. The largest term never appears on a statement.
Unit convention
Every leak on this page is expressed in R — multiples of the risk the trade was sized against. Costs quoted in pips or currency cannot be compared across instruments, timeframes or account sizes. Costs quoted in R can be subtracted directly from expectancy.
Magnitude column shows the modelled range for a two-ATR stop on an hourly trigger under the Foundry's default cost assumptions. Ranges are illustrative, not measured.
Eight measurements,
one unit.
The execution layer is the only part of a trading system where the raw material — excursion data — usually does not exist in the operator's records. These are the eight measures the Foundry treats as the minimum viable set, and the four that carry the diagnostic weight.
Definitions · AR.EXEC_METRICS
The measure that is almost always missing
Net outcome R — outcome after fee drag — is the number that belongs in every expectancy calculation. Most journals record gross R and most platform reports record currency. An expectancy computed from gross R and then compared against a friction-inclusive benchmark is comparing two different quantities and will always flatter the strategy.
Section 03 · Specimen AR-EXP-066-S1
One trade,
under magnification.
A single modelled long position on an hourly trigger with a four-hour authority chart. Sixteen bars from fill to exit. Every number below is stated twice — once in the diagram and once in the breakdown — so that nothing depends on reading a picture.
The four coloured spans are the whole argument. The red span is what the entry cost in discomfort; the green span is what reached the account; the amber span is what the trade earned and then handed back; the grey span is the part of the stop that was never needed. Only the green span appears in a profit-and-loss report.
Execution scope
Decorative scope view: the cyan trace is the intended path, the amber trace the realised one, and the red hatching between them is the deviation. Everything it depicts is stated numerically in Figure 1 and in the breakdown below.
Specimen breakdown
| Bar | Event | Price | Open R | Note |
|---|
What this specimen is not
It is not a recommendation, a setup template or a historical trade. It is a constructed example chosen because its arithmetic is legible: one unit, no partial, one trailing exit. Real management branches split the position, which makes every measure below a weighted average rather than a single number.
The detail most journals miss
Position size was computed on the intended 54.0-pip stop. The fill arrived 2.7 pips higher and the stop stayed anchored to structure, so the position actually risked 56.7 pips — 1.05 R. A five per cent size error on every chased entry is invisible in the outcome column and compounds directly into drawdown.
Plot pain against
opportunity.
Outcome alone cannot separate a well-located trade from a lucky one. Adverse excursion against favourable excursion can. Filter the modelled population to any single quadrant and the summary recomputes over that subset only.
Each point is one modelled trade. The vertical divider sits at 0.45 R of adverse excursion — the point beyond which an entry has consumed nearly half its own invalidation. The horizontal divider sits at 1.60 R of favourable excursion, the level at which a trade had enough range to be worth managing well.
Selection summary
Assignment precedence: efficient exit → missed opportunity → clean alpha → late but correct → high pain low reward → clean but weak. Every trade belongs to exactly one quadrant.
Simulated demonstration data
Section 05
Six populations,
six different repairs.
A quadrant is not a grade. It is a diagnosis pointing at a specific subsystem: entry location, regime permission, target structure or exit discipline. Two strategies with identical expectancy can sit in completely different quadrants and require opposite corrections.
The most expensive quadrant is the one that never shows up as a loss. Missed opportunity is invisible in the profit-and-loss statement and is therefore the last thing most operators repair — usually after they have already degraded a working entry.
Interactive · Level 2
You cannot maximise capture
and participation at once.
Three exit controls. Move any of them and watch capture efficiency and tail participation travel in opposite directions, with giveback tracking one and net expectancy tracking the other. This is the central trade-off of the execution layer, and it has no free solution.
Exit policy
Interactive demonstration
Left: the share of available favourable excursion that reached the account. Centre: expected surrendered opportunity per trade, in R. Right: the share of an extended move the policy is structurally able to participate in. The left and right dials are in direct opposition.
Live reading
Why the dials disagree
Capture efficiency is measured over the whole population. Tail participation is measured over the small subset of trades that extend. A policy that banks the position at the first target converts most trades into clean, high-capture results and removes the system's access to the few outcomes that pay for everything else. A policy that trails widely does the reverse: the typical trade looks badly managed, and the distribution's right tail survives.
This is why capture efficiency must never be optimised on its own. Pushed to its maximum it produces a system with excellent execution statistics and materially lower expectancy — the most common self-inflicted wound in the execution layer.
Available R = p₁ × [p_c·peak + (1−p_c)·trail]
Expected favourable excursion per trade. p₁ is the probability of reaching the first target, p_c the probability of extending beyond it, peak the excursion high of an extended runner and trail the activation level of a runner that fails to extend.
Giveback R = Available R × (1 − capture)
Surrendered opportunity per trade, expressed so that it can be placed beside spread and commission on the same axis. Under the Foundry's default assumptions it is roughly an order of magnitude larger than all quoted costs combined.
From theoretical
to realised.
The waterfall starts from the expectancy a configuration would produce if every trade exited at its favourable excursion high and cost nothing to transact. It then removes the terms in the order they actually occur. The cost controls and the exit controls above share one configuration, so the two panels always describe the same system.
Decomposition of modelled expectancy in R per trade. The first bar is the theoretical maximum, the last is what survives. Every intermediate bar is a subtraction you can act on, though not with equal leverage.
Cost assumptions
Interactive demonstration
Proportion, not size
Research record AR-EXP-084 · Parameter Study · Validated 2026-04-29
Cost is a timeframe decision
before it is a broker decision.
Record AR-EXP-084 holds the cost assumptions completely fixed and varies only the trigger timeframe. Because the working stop is derived from that timeframe's ATR, an identical bill in pips becomes a radically different bill in R. Nothing about the broker changed.
Fixed assumption of 1.0 pip spread, 0.5 pips slippage per side and 0.6 pips commission equivalent — 2.6 pips round turn — charged against a two-ATR stop on each timeframe. The amber threshold marks 0.10 R per trade, the level beyond which friction consumes a majority of a typical modelled edge.
| Trigger | ATR | Stop | Friction | Verdict |
|---|
Finding
Identical cost assumptions consume roughly nine times more of the edge on a five-minute trigger than on a four-hour trigger. A strategy that is unprofitable on M5 and profitable on H4 has not become a better strategy — it has stopped paying a toll it could not afford. Cost negotiation cannot recover a difference of that order.
Section 08
Duration is a diagnosis,
not a statistic.
A duration distribution answers questions that outcome data cannot. A tall spike at the shortest durations is not evidence of decisiveness; it is usually a stop sitting inside the instrument's normal noise, or a break-even rule activating before the trade has had room to develop.
A long right tail with no corresponding right tail in the outcome distribution is worse: it means positions are being held through time without being paid for it. Holding cost is not only swap and financing — it is the exposure to unscheduled events that accumulates every hour a position is open.
The gap between median winner duration and median loser duration is the single most useful number here. If losers are held longer than winners, the exit logic is being applied asymmetrically — almost always by a human, almost always in the direction of hope.
Modelled holding periods in hours across the same trade population used in Figure 2.
Repair in order.
Never in parallel.
When a quadrant dominates a population, the correction sequence matters more than the correction itself. Changing entry and exit logic in the same revision destroys the attribution and you will not know which change did what. Open the quadrant that describes your dominant population.
One change per revision
Each repair invalidates the sample that preceded it. A framework revised on three axes at once has no comparable history at all, and the next evaluation begins from an effective sample size of zero.
Order of operations
Exit repairs before entry repairs, always. Exit changes are reversible, testable against the existing excursion record and do not alter which trades were taken. Entry changes alter the population itself and destroy comparability with everything measured before.
Methodology
How the figures on this page are produced
Every number on this page comes from the Foundry's preview engine — a deterministic, closed-form model of the directional relationships between parameters and system behaviour. It reads no market data, no broker export and no trade journal. Given the same inputs it returns the same outputs; no random number is drawn without an explicit seed.
The trade population in Figures 2 and 6 is generated by seeding the engine's outcome ladder and then attaching excursion values to each outcome under stated rules: winning trades receive an adverse excursion drawn from a distribution concentrated near the entry and a favourable excursion at least as large as their result; losing trades receive an adverse excursion concentrated near the stop and an independent favourable excursion, which is what produces the missed-opportunity population.
The waterfall decomposes friction using the same expression the engine uses internally, so the bars sum exactly to the engine's net expectancy rather than approximately. Timeframe friction in Figure 5 uses each timeframe's representative ATR from the engine's reference table and a fixed two-ATR stop coefficient.
Limitations
What this page cannot do
Excursion data is the hardest data to collect
MAE and MFE require the full intrabar path of every position, at a resolution finer than the trigger timeframe. Broker exports almost never contain it. Statements record fills, not the journey between them, and reconstructing excursion from candle data systematically understates both measures because the sequence of high and low within a bar is unknown.
Modelled, not measured
The excursion values here are generated to be structurally plausible, not observed. They demonstrate how the measures behave and how the quadrants separate populations. They are not evidence about any instrument, venue or strategy, and no figure on this page should be read as a performance claim.
Costs are stationary here and are not in reality
Spread widens around scheduled events, at session boundaries and in the conditions that generate the largest signals. A fixed spread assumption flatters every strategy whose entries cluster in exactly those windows. Slippage is also asymmetric: it is worse on the trades you most wanted filled.
One instrument, one convention
The specimen trade uses a pip convention appropriate to a major currency pair. Instruments with different tick structures, commission models or financing costs require the same measurements with different constants — the R conversion is what makes them comparable at all.
Continue
Execution quality is measured, then governed.
Repairing the exit layer changes what a system gives back. It does not change what a bad sequence of outcomes can do to the capital behind it. The next two chambers deal with that: the distribution of orderings, and the authority layer that decides how much may be deployed into it.