EdgeWatchFXby ErgodicLabs

What EdgeWatchFX does differently

A closed-trade balance leaves out the losses on open positions, so it can make a grid or a martingale look like the smoothest account on a list. EdgeWatchFX was built around that problem. Below is what it does, why, and what it looks like.

At a glance

QuestionA common approachEdgeWatchFX
What the curve showsClosed-trade balance, or equity sampled now and thenEquity from the bridge every minute while positions are open; earlier history rebuilt from the deals and one-minute prices, with each day's low kept
Maximum drawdownMeasured on the balanceAt the intraday equity low, never reset, with the part only open losses showed drawn apart
History before connectingStarts on the day of connection, or imported as balance onlyRebuilt minute by minute from price bars where the terminal has them, and marked as rebuilt
Grids and martingalesA high win rate and a smooth curveFound from sizing and added positions; a basket counts as one bet; from High upward no edge is claimed for them
The headlineOne score or one gain figureThree readouts kept apart, so a good return cannot offset a dangerous way of trading; the Edge Matrix Score beside them, lowered by the danger rules
Short recordsShown with the same weight as long onesThe chance the edge was above zero, its likely range, and roughly how many more trades a firm claim would need
After an EA goes liveWatched by eyeEach EA is checked against its own earlier trades, and the app shows how often its results monitor would raise a false alarm
ProjectionsNone, or the average return drawn as a straight lineThe most likely course from the account's own days, a zero-edge line, and the same reading from three months ago set against what happened since
Breakdowns by symbol, session, weekdayRaw tables, where noise can look like a patternShrinkage and false discovery control; filters only after later trades agreed
NewsA general calendarReleases for the currencies you trade, and how each EA did around past ones, compared with trades held about as long
ReturnOne gain figureTime-weighted return next to the money actually gained or lost, with a flag when the timing of deposits made them disagree
Public profilesStart date and included trades vary, and are not always labelledStrict, or a custom view that says so; changes logged in public; a withdrawn record keeps its reference

"A common approach" describes what is widespread among account trackers and signal platforms, not any one product, and several products do parts of what EdgeWatchFX does. Every diagram below is a synthetic example made for this page.

The record itself

Every readout on the site rests on these choices. If the curve leaves out open losses, nothing calculated from it can put them back.

Equity, not balance

The risk that was carried, not only the risk that was closed

A grid or martingale closes its losing positions late, so its balance rises in small steps while the equity underneath can fall far below it. The bridge sends equity every minute while positions are open, and history from before the bridge is rebuilt from each deal and the one-minute price of every open position. Each day's low is kept.

In this example the balance never fell more than 4.9%. The equity fell 36.7% in one basket.

Example grid account: the closed-trade balance rose smoothly, with a largest dip of 4.9 percent, while the equity low fell 36.7 percent below its peak in one basket.-30%-15%0%+15%+30%Day 1Day 200
Closed-trade balanceLowest equity each day
Where each figure comes from

From the first trade, and clear about how it was measured

When the bridge is attached it uploads the whole deal history and the one-minute bars the terminal holds for every symbol traded, so the record starts at the first deposit rather than today. Days that could be rebuilt from bars are shaded as rebuilt; older days without bars are marked as closed trades only. After that, the data is sent by the bridge and signed with the account's key.

The record panel says how much was rebuilt, when data last arrived, and how long the bridge was silent in the last 30 days.

A record in three parts: early days with closed trades only, days rebuilt from minute bars, and days measured live by the bridge.RebuiltBridge attached
Closed trades onlyRebuilt from minute barsMeasured by the bridge
Return and money

A deposit cannot turn a loss into a gain

Time-weighted return takes deposits and withdrawals out, so new money never counts as profit. It can still mislead the other way: a strong run on a small balance followed by a large deposit and a loss gives a good return and lost money. Both figures are shown side by side, with a flag when deposit timing made them disagree.

Here: +50% on $1,000, a $10,000 deposit at the peak, then -10%. Time-weighted return +35%. Money: $650 lost.

An account that gained 50 percent on 1,000 dollars, received a 10,000 dollar deposit at its peak, then lost 10 percent: a 35 percent time-weighted return and 650 dollars lost.$1k$2k$5k$10kDeposit of $10,000$1,000 start$10,350 now
Equity on a log scale, so the early run on $1,000 is visible next to the deposit.

Judging the account

Three separate questions, each with its own evidence: how dangerously the account trades, whether each EA still behaves like itself, and how much the record shows so far.

No single score

A good return cannot buy back hidden risk

A blended score adds return and risk together, so a grid that is still winning can score well right up to the basket that ends it. Here Danger, Health and Evidence are never added up. Danger is judged from how the account trades, so profit does not lower it. The app also shows the Edge Matrix Score, the backtest score of Edge Matrix worked out for a live record; the danger rules and the record checks count against it, and it sits beside the three readouts, never in their place.

One blended score
82 / 100

High return and a high win rate lifted it. The open baskets lowered it a little.

The same account here
DangerCritical
HealthHealthy
EvidenceWithheld
Illustration. Healthy only means the EA behaved like its own past. Evidence is withheld because the closed trades did not describe the risk.
Danger

A grid basket is one bet, not a string of wins

Counted trade by trade, a grid shows a high win rate, because each small position closes in profit until the basket that does not. Here, positions by the same EA, on the same symbol and side, that were added while earlier ones were still open count as one basket, and every statistic counts the basket once.

The Danger rules look at lot size growing after losses, positions added at worse prices, trades without a stop, past stop-outs, the distance to stop-out in daily ranges, the gap between closed-trade and equity drawdown, outsized tail losses, and losers held much longer than winners. Each is shown with its own figures, worst first.

Four grid baskets: three winning baskets of small positions and one losing basket that took back more than the three wins together.Basket 1+1.4%Basket 2+1.6%Basket 3+1.2%Basket 4-6.9%Each bar is one position; taller means a larger lot.
Per position: 25 wins, 12 losses. Per bet: 3 wins, 1 loss, and the loss took back more than the three wins together.
Evidence

Lenient with good records, honest about luck

A short profitable record is called what it is: good news that is still early. The edge per trade is estimated from a sceptical starting point centred on zero, which gives the chance that it was above zero, its most likely size and an 80% range.

The top level, Established, also needs a conservative bound to be above zero. That bound stays valid however often the page is checked, so looking every day does not make a lucky record pass.

In this example the estimate was positive almost from the start; the conservative bound only rose above zero after 395 trades.

The estimated edge per trade with its 80 percent range, which narrows as trades come in, and the conservative line, which crosses zero much later.-0.4%-0.2%0%+0.2%+0.4%0 trades500 trades
Estimated edge per trade80% rangeConservative boundUnder 30 trades: very early
Illustration of how the figures move as trades come in. The app shows the figures, not this chart.
Health

Each EA against its own past

Once an EA has enough trades, its early record becomes its baseline and every later trade is checked against it: a running sum of results below the baseline, the chance of its longest losing streak, its drawdown against resampled copies of its own history, and changes in sizing or holding time. The baseline stops growing, so a faded edge does not slowly become its own yardstick.

For each account the app shows how often the results monitor alone would raise a false Watch on an EA that had not changed. The other monitors can add to that.

A cumulative sum of results below the EA's own baseline. It stays near zero while the EA is unchanged and climbs through Watch into Degraded after its edge faded.WatchDegradedAfter baselineEdge fadedTrade 240
Example picked to show no false alarm before the change: the line stayed low while the EA was unchanged and climbed through Watch into Degraded after its edge faded. An unchanged EA can still reach Watch now and then.

Looking ahead, and slicing the past

The two places where statistics most easily mislead: a projection, and a table of results cut many ways.

Projection

A most likely case you can hold against what happened

The account's own trading days are resampled in blocks so calm and rough stretches stay together, intraday lows included. Each path first draws a plausible mix of the recorded days, so the luck of a short record widens the spread instead of being replayed as certain. On each day the most likely level is drawn, with the densest half of the paths around it; best and worst cases are left out.

A dotted line shows the most likely level for the same days with zero edge: the gap between the lines is the part that rests on the estimated edge. The same reading is also built from the record as it stood three months ago, and the equity since is placed against it, so a reading that was too narrow shows up over many accounts.

The most likely level and the densest half of resampled paths: one reading built from the record as it stood three months ago, with the live equity since drawn through it, and today's reading with a dotted line for the same days with zero edge.3 months agoToday+3 months
Recorded equityMost likely levelDensest half, three months agoDensest half, from todayMost likely, zero edge
If the recorded behaviour continued. In this example the equity ended at the 29th percentile of the paths from three months earlier, inside their densest half.
Edge map

Slicing that does not invent patterns

Cut results by symbol, session, weekday, holding time and other ways, and some cells always look strong by chance. Small cells are pulled toward the average by their own noise, all cells shown at once are tested together, and cells with too few trades carry no verdict.

A filter is only suggested when it was found on earlier trades and held on the trades that came after. Choosing filters from live results without that check would be curve-fitting with real money.

Taken at face valueShrunk and tested together
The same slices twice: taken at face value 9 of 42 cells look strong or weak; after shrinkage and false discovery control two cells remain, and cells with too few trades carry no verdict.
Did betterDid worseToo few trades
Synthetic example with two real effects among 42 cells. At face value 9 cells looked clearly strong or weak, 3 of them from a handful of trades; the controlled view kept the 2 real ones.
News

Your releases, and how your EAs did around them

The bridge reads the economic calendar from the terminal. The News screen lists the coming releases for the currencies in your open positions and in the symbols you traded over the last 90 days, and, for each EA, how trades that were open through past releases did.

Longer trades are more likely to be open through a release, and in many EAs they also do worse. So each EA's trades are split into fifths by holding time, trades around releases are compared only within their fifth, with a permutation test and a correction for testing many releases. Where no difference shows, it says so.

A plain comparison shows trades open through a release doing much worse, but within each holding-time group trades around releases did about the same as trades held as long without one.All tradesShortest2nd3rd4thLongestPlainWithin each fifth of holding time
Open through a releaseOther trades
Example: the plain comparison suggested releases hurt; within each fifth of holding time they did about the same. Verdict shown: not told apart from chance.

Built by the makers of Edge Matrix

EdgeWatchFX is made by ErgodicLabs, which also makes Edge Matrix, a validation app for backtests. The Edge Matrix statement parsers and resampling engine are copied into EdgeWatchFX unchanged and checked file by file. The live readouts are new code, written with the same approach: say what the record shows, and how sure that is.

Edge Matrix

Before going live: is the backtest robust, or fitted to its history?

EdgeWatchFX

After going live: is the account carrying hidden risk, and is each EA still behaving like its record?

The methods behind the readouts

Stationary block bootstrapResamples whole stretches of days, with the block length chosen from the data by the Edge Matrix engine, so streaks and calm periods are kept. Used for projections.
Bayesian bootstrap weightsEach simulated path first draws a plausible mix of the recorded days, so uncertainty about the record itself widens the spread.
Sceptical prior on the edgeStarts from "no edge" and lets the trades move it, which gives the chance the edge was above zero and its likely range.
Confidence sequencesBounds that stay valid however often they are looked at, unlike a test repeated every day until it passes.
CUSUMA running sum of shortfalls against the baseline, a standard tool for noticing a change early without reacting to every bad week.
Resampled control limitsAn EA's current drawdown and time under water are compared with what resampled copies of its own baseline produced over the same span.
ShrinkagePulls each small group toward the overall average in proportion to its noise, so a lucky handful of trades cannot top the table.
False discovery controlBenjamini-Hochberg across every cell or release tested at once, so testing many slices does not manufacture findings.
Stratified permutation testsShuffles labels only among comparable trades, here trades in the same fifth of holding time, so a difference cannot come from holding time.
Forward checksA pattern found on earlier trades is only reported as a filter after the trades that came later agreed with it.

The methods are public; the exact thresholds and weights are not, so the readouts cannot be tuned against. How each readout is measured

Checks on the record, and on us

Profiles say what they are

A strict profile starts at the first deposit, includes every trade and never resets its drawdown. A custom view may start later or leave EAs out, and is labelled as one with the list of what was changed. Danger always covers the whole account.

Changes are on the record

Start dates, left-out EAs, product names, publishing and withdrawing are logged on the profile. A withdrawn or deleted profile keeps its reference and last figures, so anyone can still look it up on the verify page.

Our own interests, stated

The founder of ErgodicLabs sells EAs and earns introducing-broker commission from one broker, and every public profile says so. Once the community forum opens, sellers we identify carry a Vendor label there, the founder included. Conflict policy

A read-only bridge

No broker password is shared. The EA cannot place, change or close a trade. Every request is signed with the account's key; the secret is never sent with a request and is not stored in the database, and old requests are refused.

Check any record free

Upload an MT4, MT5, cTrader or MQL5 signal statement and get the Danger rules the file has data for, without an account. A statement holds closed trades only, so open losses are reported as not measurable. The file is not kept, and a backtest is named as a backtest. Check a record

Limits stated

History before the bridge is rebuilt from minute bars and can miss moves inside a minute. A short record or one market regime can look stronger or weaker than the strategy is. No figure here is a forecast or investment advice.

See your own account measured this way

Start free with one account and its Danger, Health and Evidence readouts. Pro and Portfolio track more accounts, with 50 EAs per account.

Track an account free