September split in two: entry activity grew for three weeks, then fell, while the market mix kept rotating toward Gold

02 · Entries and management

Management instructions came back, and changed shape on the way

Pulse divides tracked instructions into entries, which carry a recognised entry command such as buy or sell, and management and other instructions: closes, partial closes, stop and target modifications and pending-order cancellations. The entry share is entries as a percentage of all tracked instructions. Because it is a share, it moves when either group changes.

In August the two groups pulled apart. Entries rose 8.1% and management fell 22.0%, lifting the entry share from 57.0% to 64.8%. September partly reversed it. Entries were unchanged at 0.0% on calendar months, management and other instructions rose 16.0%, and the entry share fell to 61.3%. Over the quarter the share went 57.0%, 64.8%, 61.3%: a swing, not a trend. The share was 62.5% in the first window of September and 58.3% in the second, so management made up a larger part of the smaller late-month stream.

The rebound was not a return to July's mix of commands. Changes in activity by command family, September against August, with August against July for reference:

The direction is consistent. Instructions that end or withdraw something rose: pending-order cancellations by 93.1%, close-all commands by 24.2% and side-specific closes by 24.1%. Instructions that adjust something already in place fell: stop and target modifications by 31.0% and volume-based partial closes by 30.1%, after a fall of 60.1% in August. Pending entry commands rose 23.0%, although at 1.7% of entries they remain a small family. More pending orders sent and more cancelled is internally consistent: strategies that place conditional orders also withdraw the ones that do not trigger.

The strongest alternative reading is design rather than behaviour. If some strategies moved their exits into the entry instruction or into receiving-EA settings during August, and moved them back in September, management instructions would fall and recover with no change in how anything was managed at a broker. These are counts of command text. A rise in close commands does not show that more positions were closed: a strategy can send a close instruction when nothing is open, and exits can happen through broker-side stops that never pass through the bridge. What the record does show is that August's low level of management instructions was not a new normal. A reader who took August's entry share as the platform's settled mix would have been wrong a month later.

03 · Market mix

The rotation toward FX and metals continued, and the class mix concentrated

Three numbers describe an asset class, and each answers a different question. Its share is its slice of all entry instructions in the month. Its activity change is how much its own entry activity grew or shrank. Its share change is measured in percentage points: a class moving from 30% to 33% of entries gains 3 points, whatever happened to its own activity.

With total entries unchanged, a class could only gain share by taking it from another, and in September the exchange ran one way. Metals rose from 32.1% to 35.5% of entries, with its activity up 10.6%. FX edged from 35.2% to 35.9%, its activity up 2.2%. Equity indices fell to 17.3% (activity −9.1%) and crypto to 10.2% (activity −17.0%). FX and metals together took 71.4%, up from 67.3% in August and 59.7% in July.

August's report also showed each class's contribution to the net change in entry activity. This report does not. When net entry activity barely moves, that calculation divides by a number close to zero and produces values in the thousands of percent: arithmetically correct and meaningless to a reader. Share and activity change answer the question instead.

Two kinds of concentration

"Concentration" can mean two different things, and September separates them cleanly. Across asset classes it can be summarised with the Herfindahl–Hirschman index, which economists use to describe market structure: square each class's share and add the squares. Its reciprocal is easier to read: the "effective number" of equally sized classes that would produce the same index. Five equal classes would score 5.0; a single class 1.0. On consistent rules the effective number of asset classes fell from 3.91 in July to 3.60 in August and 3.38 in September. Entry activity is being spread across fewer classes each month.

Across leading markets the story is different. The four leading markets took 59.3% of entries in July, 59.7% in August and 60.1% in September. Gold's gain was mostly offset inside that group: Bitcoin fell from 12.1% to 10.0% of entries and Nasdaq 100 from 7.8% to 7.0%.

Read together, the two measures say something precise. Activity has been leaving crypto and equity indices for FX and metals, which concentrates the class mix. Within the leading markets it has been moving from Bitcoin and Nasdaq 100 to Gold, which reshuffles the top four without making them a larger part of the whole.

Inside the classes

Within metals there is effectively one market: Gold made up 99.1% of metals entries. Its entry activity reached 97.9% of the entire FX class's, one market almost equal to every currency pair combined. FX hides its components. The class grew 2.2%, but each of its three leading pairs declined: EUR/USD by 2.1%, GBP/USD by 1.2% and USD/JPY by 1.0%. The three pairs' combined share of FX activity fell from 53.4% to 51.5%. FX grew through the wider range of pairs beyond its leaders. Crypto's decline is essentially a Bitcoin decline: Bitcoin made up 98.6% of crypto entries. That decline was already under way in the strong first weeks of September, so it is a separate story from the late-month fall.

A rising share of entry instructions is not rising popularity or market volume. A small number of high-frequency strategies can move a market's share with no change in how many people trade it, and Pulse does not measure the latter. September's figures show that a larger part of the instruction stream referred to Gold and a smaller part to Bitcoin and equity indices. They do not show who sent those instructions, whether a broker executed them or what happened next.

04 · Gold

Gold now sets the platform's rhythm

Gold took 35.2% of September's entry instructions, up from 31.8% in August and 29.7% in July, and its entry activity rose 10.8%. At more than a third of all entries, its rhythm increasingly shapes the platform's. Every full week of September stayed above August's monthly share; the two partial weeks at either end of the month cover four and three weekdays and are not comparable with a full week.

Gold is also a weekday market in the instruction stream. It made up 36.1% of weekday entries in September and 12.2% of weekend entries. Its daily shape is the most pronounced on the platform: the ratio of its busiest to its quietest source-clock hour was 28.6, against 1.37 for FX and 1.18 for crypto, whose instruction flow is almost flat around the clock. As Gold's share grows, the platform's busiest hours carry more of the day. The hour labels follow the source timestamps, whose time zone is still unconfirmed, so these figures describe the shape of the day, not its alignment with any market session, and they say nothing about when it is good to trade.

05 · Macro backdrop

A tightening month in rates and the dollar

This section is external context. Every value comes from a named public source and is dated, and none is combined with a Pulse measurement in a single statistic. Nothing here explains the instruction data.

September brought a cluster of central-bank decisions. The European Central Bank raised its three key rates by 25 basis points on 10 September. The Federal Open Market Committee raised the federal funds target range by 25 basis points to 3.75–4.00% on 16 September and published new projections. The Bank of England held Bank Rate at 3.75% on 17 September, and the Bank of Japan set its policy-rate guideline at around 1.25% on 18 September. The US macroeconomic releases published during the month all measured August: the employment report on 4 September, CPI on 11 September and PCE inflation on 30 September, the last day of the window. On 2 October, after the month had ended, August's payroll figure was revised down. That revision was not available to anyone during September.

Market prices moved with the decisions. Between 31 August and 30 September the US two-year Treasury yield rose from 4.34% to 4.88% and the ten-year from 4.75% to 5.29%. By 25 September, the latest date published at the time of writing, the Federal Reserve's broad trade-weighted dollar index had risen from 118.57 to 120.33, the euro had fallen from 1.1617 to 1.1400 dollars and the dollar from 159.74 to 157.18 yen.

It is tempting to connect the two records. Management instructions rose in a month of central-bank decisions. Gold's share kept climbing while yields rose. Entry activity weakened in the final week, when yields rose most. Each of these is a temporal overlap, not a finding. Monthly and window totals combine every instruction from every strategy, the source clock's time zone is unconfirmed, and the method that would test an announcement effect, an event study around precisely timed releases, is not available until that time zone is confirmed. The late-September decline was also spread evenly across asset classes, so whatever lay behind it was not specific to any one market.

What the backdrop does supply is a set of questions an approved method could later test without stretching the data. Do instruction rates in the hours around a scheduled policy statement differ from the same hours on comparable days? As more months accumulate on consistent rules, does the composition of management instructions differ between months with and without major policy decisions? Strategies driven by price indicators generate alerts when prices move, so does aggregate instruction activity track market movement, measured without combining internal and external figures in one statistic? None of these is a working hypothesis of this publication, and none has been tested here.

06 · Exit fields

The exit-field average moved because classes changed, not because the mix did

An entry instruction can carry a recognised stop-loss (SL) field, a take-profit (TP) field, both or neither. Pulse reports each as a share of entry instructions. This measures instruction text. It does not show whether a level was valid, whether a broker accepted it or whether anything was protected. Exits can live in receiving-EA settings, later instructions or strategy logic, so an entry without a recognised field does not mean the trade had no exit.

At platform level September moved modestly. Entries carrying any SL field changed by −0.9 pp to 66.4% and any TP field by +1.4 pp to 57.4%. The four-way partition moved more: entries with both fields +3.0 pp to 55.5%, SL only −4.1 pp to 10.8%, and entries with neither field +2.7 pp to 31.8%. The full partition is in the reference section below.

An average like this can change for two different reasons. The class mix can tilt toward classes that carry fewer fields, or the classes themselves can change. A shift-share decomposition, a standard tool in regional and labour economics, separates the two. It splits the change into a mix effect (class weights change while each class keeps its earlier rate), within-class effects (class rates change while weights stay fixed) and a small interaction term. From August to September, the neither-field share changed by +2.7 pp; the mix effect was −0.1 pp and the within-class effect +2.7 pp. From July to August the totals were −3.7 pp, +0.2 pp and −4.5 pp. In both months the considerable rotation toward FX and metals barely moved the platform's exit-field profile, because those classes' field rates sit near the average. Almost everything came from changes inside classes.

The within-class effect has a clear leader. In September the share of metals entries carrying neither field rose 6.7 points, which on its own contributed +2.2 pp of the platform's +2.7 pp change. Equity indices went the other way: their neither-share changed by −2.4 pp and their TP-field share by +9.8 pp. In August the picture was reversed. Metals' within-class change then reduced the neither-share by 1.8 points, while FX's raised it by 1.0. Metals' field profile has now turned around twice in two months while its share of activity rose in both. One possible reading is that the mix of instruction templates behind metals changes from month to month, some carrying exits and some not. The aggregates cannot confirm or rule that out.

Fewer recognised fields on metals entries is not evidence of less careful risk control. Strategies that manage exits through the receiving EA, or close with a later instruction, would produce exactly this pattern with no change at the broker.

07 · July to September

What continued, what reversed and what remains open

September closes the third quarter, which invites a three-month story. Pulse's separately published reports cannot supply one directly. When the August edition was finalised it recomputed July under refined eligibility rules, so the standalone July report and the August edition give slightly different July figures. A line drawn through the July report's entry share and the later months would join numbers built on different definitions. The table below avoids that: all three months use the same rules, with July as it appears in the August edition's comparison. They are three comparable observations, not a quarterly statistic. Nothing is averaged, pooled or compounded.

What continued. The move toward FX and metals, and specifically Gold, ran through all three months, and the class mix concentrated each month. Crypto and equity indices lost share each month. The weekend became a smaller part of the week: an average weekend day carried 14.9%, 13.4% and 11.3% of an average weekday's entries. Crypto's share of weekend activity narrowed with it, from 72.3% to 63.1%. The two are linked: crypto trades every day, so a smaller crypto share shrinks the weekend relative to weekdays. The weekend measure compares average days, so it is far less sensitive to the calendar than a monthly total.

What reversed. The balance between entries and management swung up in August and partly back in September. The neither-field share fell in August and rose in September, driven both times mainly by metals, in opposite directions. Entry activity on a standardised week rose strongly into August, then fell in September's last nine days.

What stayed put. Direct entries made up 96.3%, 96.8% and 96.5% of entries. Top-four market concentration stayed near 60%. The platform's daily shape kept the same busiest and quietest hours, with the ratio between them easing from 3.32 to 3.20 to 3.11.

What remains open. Three months is a short record. A pattern that holds for a quarter describes that quarter; it is not evidence of a lasting shift. Pulse published no working hypotheses for these months, so there is nothing to score. Whether the entry and management swing is cyclical, whether metals' changing field profile reflects changing instruction templates, and what lay behind the late-September fall all need more months on one set of rules.

08 · Findings

What September establishes

1. September split in two. On a standardised week against August, entry activity was +4.0% in 1–21 September and −18.4% in 22–30 September. The first window was a rotation between asset classes; the second was a decline shared by every class, with management instructions still above August.
2. The calendar flattered the headline. All tracked instructions rose 5.7% on calendar months. A calendar model puts the effect of September's extra weekday and fewer weekend days alone at +2.7%.
3. Management instructions drove the rise in the total. They rose 16.0% while entries were unchanged, with closes and cancellations up and modifications down.
4. The rotation toward Gold and FX continued and the class mix concentrated. FX and metals took 71.4% of entries; the effective number of asset classes fell to 3.38 while top-four market concentration stayed near 60%.
5. The exit-field average moved because classes changed, not because the mix did. The neither-field share changed by +2.7 pp; the mix effect was −0.1 pp.

The lesson that connects them is about reading monthly instruction data at all. A single total combines a calendar, a composition and a timeline, and each can move in a different direction. None of September's findings is visible in its headline, and each needed its own comparison. Methods, definitions and the standardised-week calculation are described in Methods.