EdgeWatchFXby ErgodicLabs

How EdgeWatchFX measures an account

Every figure on this site describes recorded results. Nothing here predicts what an account will do next; projections are labelled as "if the recorded behaviour continued".

The data

The bridge EA sends every deal (with its stop and target at entry), equity and margin every 60 seconds while positions are open, symbol details, and one-minute price bars for the symbols traded. From these the service rebuilds the account's equity at every minute, including the losses on positions that were still open. Cash flows (deposits, withdrawals, bonus credit) are kept apart from trading, so returns are time-weighted and a deposit never counts as profit.

Verification levelMeaning
Publisher, signedSent by the bridge EA from the terminal, each request signed with the account's key and bound to its login and server.
Broker-linkedRead from the broker platform's own interface (planned for cTrader).
File, unverifiedA statement checked on the free statement check. Marked as unverified, holds closed trades only, and is not kept.

Danger

Danger is judged from how the account trades, not from how much it made, so a profitable record does not lower it. The rules look at lot size after losing trades (martingale sizing), positions added at worse prices while earlier ones were still open (grids and averaging down), positions opened without a stop-loss, stop-outs in the history, the distance to the broker's stop-out level measured in daily price ranges of the open positions, outsized tail losses, and losers held much longer than winners.

The maximum drawdown is measured at the intraday equity low, not on closed trades. The drawdown bar shows both: the grey part is what the closed trades showed, the red part is what only the open losses showed. Snapshots are taken every minute while positions are open, so a dip between two snapshots can be missed; equity drawdown is therefore reported as "at least" its value.

Each rule is listed with the figures behind it so it can be checked, at one of four levels: Low, Elevated, High or Critical. The levels are starting points and are recalibrated as real accounts, including accounts that later failed, build up. Grid baskets count as one unit in every statistic, because forty small wins inside one basket are one bet, not forty. Results of accounts whose Danger is Critical never count as a strength in the edge map.

The readout names what the worst rule found, in plain words such as "Holds losers longer" or "Averages down", with its level beside it, and lists anything else it found. An ordinary floating loss on positions held overnight is not flagged: open losses count against the closed trades only when they are both proportionally and absolutely deeper.

Health and alerts

Each EA is compared with its own record. The first half of its trades, up to 200, is the baseline; every later trade is checked against it. The baseline stops growing at 200 trades so that a faded edge does not slowly become the yardstick it is measured against. The monitors are a cumulative sum of results below the baseline average (CUSUM), the chance of the longest losing run at the baseline loss rate, the drawdown and the time under water against resampled copies of the baseline over the same number of days, and changes in sizing, holding time, legs per trade and trade frequency. Monitors start after 60 closed trades.

An EA is Healthy, on Watch, Degraded or Broken; when one EA reaches Degraded, the account shows it too. Each monitor leaves a level only well below where it entered it, so states do not flicker. A Health alert is sent only when the state changes. The thresholds are set from the expected number of trades between false alarms, and the page for each account shows roughly how often a false Watch would come up at its trading rate, measured on simulated records that did not change.

Noticing that a small edge has faded takes time: for a typical EA, on the order of a hundred trades from results alone. The drawdown, streak and behaviour monitors exist because they react sooner, and Danger does not wait at all.

Alerts are rules the holder sets, for one account, all of them or a portfolio: loss on the day, open loss, margin level, equity below or above a level, many positions open, a position open a long time, exposure to one currency, news on what is held, drawdown from the peak, the Danger or Health level, room left on a prop limit, a quiet EA, a silent bridge, and money moved in or out. Each fires once when its condition starts, and not again until the condition has cleared and its cooldown has passed. They are listed in the app and can also go to Telegram.

Evidence

Evidence is lenient with positive records and honest about how much they show. The average result per trade, as a percentage of equity at entry, is estimated with a sceptical starting point centred on zero, which gives the probability that the edge was above zero, its most likely size and an 80% range. A confidence sequence, which stays valid however often the page is checked, gives the conservative line. On daily returns, the probabilistic Sharpe ratio and the minimum track record length give how many more days a firm claim would need.

LevelMeaning
Very earlyUnder 30 trades. Profitable or not, too early to tell skill from luck.
EarlyA positive record, but still too early to tell an edge from luck.
DevelopingA positive record with moderate evidence of an edge.
EstablishedStrong evidence of a positive edge over the period, and the conservative line above zero.
No edge shownSo far the record has not shown an edge. A longer record may change this.
WithheldShown instead of any level above when the closed trades do not describe the risk: averaging down, sizing up after losses or a stop-out at High or above, or any danger rule at Critical. Such records tend to keep winning until one loss undoes them, so their profit is not read as an edge. The closed-trade figures stay available, labelled as leaving that risk out. A clean EA on the same account keeps its own reading.

Track record

Every account shows its age: from the first trade to now, in days under a week, in weeks under a year, and in years and months after that. Beside it are the active weeks, the calendar weeks in which a position was open, and how much of the age the bridge measured live rather than rebuilt from the deal history. An account that was emptied and funded again starts a new life: its age runs from the restart, and the earlier life stays in the history.

Age is a fact, not a score. Surviving is uncommon, but staying active has not been a sign of profit: regulators reported that 68 to 89% of retail CFD and forex accounts lost money in the periods they studied, and the one four-year regulator study found 89% lost money over the four years. Among Brazilian day traders who kept trading for more than 300 days, 97% lost money. So length counts only through the statistics: the minimum track record length (Bailey and Lopez de Prado) gives how many weeks of trading a record with its own Sharpe ratio, skewness and kurtosis needed before its returns could be told from zero at 95% confidence, and the page compares that with the weeks it has. Measured weekly, a record with a Sharpe ratio of 1 needed roughly three years; one with a Sharpe ratio of 2, under a year. After a restart, only the current life counts.

Length is not counted at all when the danger rules withhold the evidence: averaging down, sizing up after losses, a stop-out, or equity far below the closed trades. Averaging down and sizing up after losses can produce years of smooth results before one move erases them, as short-volatility products did in a single day in February 2018 after years of gains, and as the Swiss franc move of January 2015 did to accounts that had been stable for years. A long life there is exposure, not skill. No points are added for age: no study has set such a weight, so none is invented.

Sources: ESMA, product intervention measures on CFDs (2018); FCA CP16/40 (2016); AMF, study of retail forex and CFD clients 2009 to 2012 (2014); ASIC REP 828 (2026); Barber, Lee, Liu and Odean, Journal of Financial Markets (2014); Chague, De-Losso and Giovannetti, Day Trading for a Living? (2019, revised 2020); Hayley and Marsh, Journal of International Money and Finance (2016); Bailey and Lopez de Prado, The Sharpe Ratio Efficient Frontier, Journal of Risk (2012).

The edge map

Results can be sliced by account, symbol, timeframe, EA, session, weekday, direction, holding time, exit reason and month. Slicing many ways always turns up something that looks significant, so every cell shown at once is tested together with false discovery control, means are shrunk toward the overall average by an amount set by their own noise, and cells with fewer than 30 units carry no verdict. Timeframe comes from the name you give each EA, because brokers do not record it.

Strong and weak points are measured against the EA's own average: a session that made money in a profitable record is not a strength unless it did clearly better than the rest. An effect that mostly disappears once each symbol's own average is taken out is marked as confounded, because a "Monday effect" is often one symbol.

A filter is suggested only for an EA that is profitable as a whole, only for a group of at least 50 trades that lost money and did worse than the rest, and only when the effect, found using the first 70% of its trades alone, held in the last 30%. From 30 trades such a group is shown as watched. Each suggestion keeps its date and keeps being checked on new trades. Choosing filters from live results without that check would be curve-fitting with real money.

Projection

Projections resample whole trading days of the account's own record in blocks, intraday lows included, starting from today's equity. The mean block length is chosen from the record itself (Politis and White), kept between 2 days and a tenth of the record, so runs of good and bad days stay together; under 60 days of history the days are drawn one at a time. 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 certainty.

Only the most likely case is shown, never a best or worst case. On each day the simulated levels are read as a distribution, and the line is the peak of its density: a Gaussian kernel estimate, with a bandwidth suited to locating a peak. The peak is read on the scale returns compound on (log equity), because a single account compounds at that rate; read per unit of money, the peak would sit lower, further below the middle outcome the wider the outcomes spread. The shaded range holds the densest half of the paths: for outcomes with one peak, the narrowest range in percentage terms that holds half of them. The line is the most likely level day by day, not one path, so the deepest fall along the way and the longest time under water are read from the paths themselves, again as their most likely values.

Before a projection is drawn, the record is checked for days that leave out the losses its method allows, because a resample can only show risk its own days already hold. The checks look at grid, martingale and stop-out patterns, the yearly pace of growth, how many losing trades have closed, the cost of a full stop-loss against the usual win and the win rate that cost needs to break even, losers held longer than winners, trades without a stop-loss, open losses against closed ones, how smooth the daily returns were, the length of the record, and whether deposit timing flattered the return. A failed check withholds the projection and shows why; a check that needs care is shown next to it. The Edge Matrix Score reads the same checks on growth pace, losing trades, stop-loss cost, stop-loss use and smoothness: a failed check lowers it, and growth faster than a moderate position size would allow without luck caps it. The pace is judged on the lower of the return figure and the money made on the money put in, because with fixed lot sizes the days on a small early balance weigh far more in the return figure than the money they moved.

A dotted line shows the most likely level with the record's average day set to zero: the gap between the two lines is the part of the projection that rests on the estimated edge. The same reading is also built from the record as it stood about three months ago, and the live equity since is placed against it, with the percentile where it ended. If the simulations matched what followed, about half of such readings would end outside their shaded half. Scoring many of them across accounts, to show whether the simulations are too narrow, starts once enough accounts have that much history; this page will publish the result.

Sources: Politis and Romano, The Stationary Bootstrap, Journal of the American Statistical Association (1994); Politis and White, Automatic Block-Length Selection for the Dependent Bootstrap, Econometric Reviews (2004), with the correction by Patton, Politis and White (2009); Parzen, On Estimation of a Probability Density Function and Mode, Annals of Mathematical Statistics (1962); Romano, On Weak Convergence and Optimality of Kernel Density Estimates of the Mode, Annals of Statistics (1988); Hyndman, Computing and Graphing Highest Density Regions, The American Statistician (1996).

Score

The Score tab shows the Edge Matrix Score: the measurements and weighting of Edge Matrix, the backtest validation app by the same maker, worked out for a live record. Trade results are read as a share of equity at entry, so lot growth and deposits do not tilt them; drawdown is read on equity with its intraday lows; a test counts only once it has enough data, and a short record is marked provisional. The backtest-only corrections, such as the haircut for idealised tester costs, are left out, because a live record has paid its real costs.

Grids, martingales and open risk lower the score through the danger rules, and the record checks described under Projection lower or cap it. The depth of the falls is scored by the drawdown, ulcer index and drawdown endurance tests; a fall deeper than 40% also lowers the score directly. Length counts when the record held up: from 13 active weeks on, a record with steady periods and short times underwater earns a credit of up to 15%, in full at about two years of active trading. A record the danger rules read as High risk earns none, because surviving longer with a grid or a martingale is exposure, not skill. A record that ended below its start, or was wiped out and restarted, is held low. The score is shown beside Danger, Health and Evidence, never in their place: a record can score well and still be withheld as evidence.

Prop accounts

An account is put under Prop when its server or company name belongs to a known prop firm, or when the holder marks it. No firm's rules are built in: firms change them and run several programmes at once, so a built-in limit could be wrong without anyone noticing. The holder enters the daily and overall limits, the profit target and the minimum trading days of their phase.

Every limit is checked on equity including open trades. The daily limit is measured from the day's reference (the previous server day's closing balance, equity, or the higher of the two, as the holder sets it); the overall limit from the initial balance (static) or from the highest closing equity so far (trailing, optionally stopping at the initial balance). Money moved in or out moves the day's reference and a trailing high with it, so a payout is never read as a loss. How often the next 20 trading days would touch a limit is read from the account's own days, resampled with their intraday lows, and says so when the record is short. History rebuilt from minute bars is lower resolution than equity the bridge measured, and the result says which it is.

Profile rules

The account holder chooses what a public profile shows, and every profile says which kind it is.

  • A strict profile starts at the first deposit, includes every trade and EA, and never resets its drawdown.
  • A custom view may start on a later date or leave EAs out. It is labelled as a custom view and lists what was changed. Inside the period it shows, the drawdown never resets.
  • On both, Danger is measured on the whole account, including anything a custom view leaves out, so a risky EA cannot be hidden by leaving it out.
  • A portfolio profile shows accounts the holder chose, and says so. Inside it every account counts from its first deposit with every trade, its Danger is shown, and the portfolio reads as its most dangerous account. Once it has been public, every account added or removed is logged on it with its date. Accounts appear as Account 1, Account 2; their numbers and servers are never shown.
  • On both, amounts, open positions (live, delayed by 24 hours, or hidden), EA names and symbols, and the broker can be hidden. These choices change what is displayed, not how anything is measured, so they do not change the label.
  • Every figure carries the engine version and the verification level.

Posts

Holders of public profiles can write short posts about the robots behind them. A post shows on its profile and on the posts page only while the profile is public, under the profile's title and reference. Its text is checked before it goes out: links, handles, e-mail addresses and phone numbers, promises of profit and daily or weekly rate claims, gambling, account management and investment offers, cracked software, adult content, hate, harassment, profanity and spam are refused, including when spelled with common look-alike letters or spaced-out letters. A messaging app named in passing is allowed; together with an invitation to write there, it is not. A seller may say a robot is for sale; buyers reach them through the product link on the profile. Images cannot be uploaded: a chart in a post is drawn by EdgeWatchFX from the profile's own data, with the holder's drawings, and frozen when posted. Visitors can report a post, and several reports hide it until it is reviewed. EdgeWatchFX does not check whether what a post says is true; the figures on the profile are measured.

Reports

A weekly report covers Monday to Sunday and a monthly report a calendar month, both counted in the broker server's days, as the terminal counts them. A full-record report covers the whole record. Returns are time-weighted, so money in or out does not count, and profit is the change in equity with money in or out taken out. The deepest fall is measured from the period's own running high, intraday lows included, and is shown again at the daily closes; when the two are far apart, open trades went deeper during the day than the closes show.

Trade figures cover the trades closed in the period. An EA's drawdown is the deepest fall of its own closed results, in percent of the account's equity when each trade opened. Each EA's health at the end of every week or month is measured with only the trades closed by then, so a later trade cannot change an earlier verdict. Account Health, Danger, the Edge Matrix Score and each EA's recent average trade are as of the day the report was made.

A scheduled report is made once per period, after the period has ended, and kept in the app, where it opens as it was made and downloads as a PDF: a weekly one on the weekday and at the hour chosen, a monthly one on the 1st at that hour, in the holder's time zone. A weekly schedule can skip weeks in which nothing changed; monthly reports always go. A report describes what was recorded and does not forecast.

Limits

  • Intraday equity before the bridge was connected is rebuilt from minute bars, so it can miss moves inside a minute.
  • A record that is short, or that covers one market regime, can look stronger or weaker than the strategy really is. The evidence verdict exists for this reason.
  • No figure on this site is a forecast or investment advice.

Changes

Changes to how anything is measured, listed from 5 November 2026 on.

  • 6 November 2026: portfolios of two or more real-money accounts can be listed on the leaderboard, in a list of their own, ranked by the score of the combined record. The portfolio score now uses the same growth and track-record rules as an account: growth judged on the money as well, and a long record that held up credited.
  • 6 November 2026: in the Edge Matrix Score, the growth pace is judged on the lower of the return figure and the money made on the money put in, and the extra penalty for a deep fall starts at 40% (it was 35%) and grows more slowly. The three drawdown tests are unchanged. A long record that held up earns a track-record credit of up to 15%, and the track record is shown on public profiles and the leaderboard.
  • 5 November 2026: the Danger readout names the worst pattern found, such as "Averages down", with its level, and lists the others. The record is checked before a projection is drawn; a failed check withholds the projection, and the same checks lower or cap the Edge Matrix Score. Posts were added.