Pulse 001 — August 2026: the month the mix changed

Preliminary — month to date. This issue covers 1–27 August 2026 against the matched 1–27 July window. It will be restated in full after month-end. Every figure is an aggregate; nothing here is investment advice, and none of it describes positions, exposure or profitability.

The one-sentence version

Total instruction activity barely moved — but almost everything underneath it did.

For every 100 instructions the bridge carried in July, about 55 were entries and 45 were management commands. In August that split was roughly 65 and 35. Entry instructions rose 19.1% while management and other instructions fell 23.0%, and the two changes very nearly cancelled: overall activity was flat to within a rounding error.

The temptation is to read that as growth. It is not. The number of active connected accounts sending entries actually fell 1.7%. What rose was intensity: the mean number of entries per active account was up 21.1%.

That mean deserves a warning label. It is an arithmetic average, not a median, and a comparatively small number of high-volume accounts can move it without anything changing for a typical one. The honest sentence is “mean entry activity per active account rose about 21%” — not “traders were 21% more active.”

Widespread and busy are different things

The clearest structural finding this month is that the two obvious questions about a market give different answers.

Gold is the exception that proves it: 59.4% of active connected accounts touched it, and it carried 31.5% of all entry instructions. Widest and busiest, and it grew on both counts.

Everywhere else the two rankings pull apart. FX became the largest asset class by activity at 35.5% — but reaches only 27.1% of accounts. Metals carried slightly less activity at 31.8% while reaching 61.4% of accounts, more than twice FX’s footprint. FX is deep and narrow; metals is broad and comparatively shallow.

That distinction matters if you are using this data to decide anything. “The most popular market” is an ambiguous question, and the two reasonable definitions of it disagree for most of the board.

What actually moved

Compared with July, the mix rotated:

• FX gained 6.2 points of activity share, with its underlying activity up 44.1%.
• Metals gained 3.0 points, activity up 31.5%.
• Equity indices lost 5.4 points, activity down 6.6%.
• Crypto lost 2.8 points, activity down 4.1%.

FX and metals together contributed more than the entire net increase in entry activity — the other classes offset part of their gain.

Inside those classes, two patterns are worth separating. EUR/USD activity rose 52.1% while its reach was effectively unchanged: the same accounts, sending far more instructions. USD/JPY rose 45.6% and expanded its reach from 7.5% to 10.2% of accounts: more accounts, each doing more. Those are different phenomena that a single “up ~50%” headline would flatten.

Bitcoin ran the opposite way: reach held steady while activity fell 4.0%. Same footprint, less intensity.

A caution on the category labels. Gold is 99.1% of metals activity and Bitcoin is 98.0% of crypto activity. When this report says “metals” or “crypto,” it is very nearly saying “gold” and “bitcoin.” Read the category names as conveniences, not as diversified baskets.

Concentration barely moved

Despite all that rotation, the four leading markets accounted for 59.4% of entries in August against 59.5% in July. Nasdaq 100 held exactly 42.0% of index activity in both months; the three leading FX pairs held 53.8% of FX activity against 53.3%.

August was a rotation within the leading set, not a concentration event.

Direction leaned long — with two ways to count it

56.2% of classified entry instructions carried a long direction, up from 51.0%. Every major asset class moved the same way.

This describes historical entry instructions. It is not a statement about open positions, net exposure, position sizes, sentiment, or whether any of it worked.

The interesting part is that two defensible ways of counting give materially different answers. Weight every instruction equally and the busiest accounts dominate. Weight every account equally and you get closer to what a typical connected account did. Gold was 53.1% long by instruction but 62.9% long per account — nearly balanced flow overall, distinctly long-leaning among accounts. GBP/USD lands on opposite sides of 50% depending on the method: 49.2% by instruction, 56.1% per account. We publish both, because publishing one would be a choice disguised as a fact.

USD/JPY produced the strongest long skew of any published market, and it holds under both methods: 73.4% by instruction, 70.6% per account, both up sharply from July.

The day has a shape, and it is stable

One hour carries 5.68% of instructions; another carries 1.78%. That 3.19× swing between the busiest and quietest hour is almost identical to July’s 3.15×, and the same two hours held those positions in both months. A pattern that reproduces across two independent months is a structural feature, not noise.

We are deliberately not explaining why yet. The storage timezone of the underlying timestamps is still being confirmed with engineering, and until it is, we can describe the shape of the day but not reliably name the market session it corresponds to. The chart carries that caveat on its face.

By weekday: Monday through Thursday sat within 0.3 index points of each other — effectively tied, so we are not naming a busiest weekday. Friday runs about 8.6% below the average weekday, and the average weekend day produces about 12.9% of average weekday activity, down from 15.4% in July.

What instructions carried

67.5% of entry instructions carried a recognised stop-loss field and 55.4% carried a take-profit field. The mutually exclusive picture: 52.0% had both, 15.5% had a stop-loss only, 3.4% had a take-profit only, and 29.1% had neither.

The month’s real movement here was a 6.6-point jump in stop-loss-only instructions, mostly at the expense of instructions carrying both.

What this cannot tell you is whether anyone was actually protected. It measures fields present in parsed instruction text. Protection may come from connector settings, from a later management instruction, or from strategy logic — none of which appear here. “29.1% of entry instructions carried neither field” is supportable. “29% of traders don’t use stop losses” is not, and we would ask you not to quote it that way.

What we are still working on

Three things are unresolved and we would rather say so than quietly ship them:

1. Timezone confirmation for the hour-of-day axis, as above.
2. A contributor-concentration check on market-level activity-weighted direction, to confirm no single account dominates a published figure.
3. Full-month restatement. These windows stop on the 27th. Every number here will be recomputed for 1–31 August, and we will replace the whole set rather than keep whichever figures aged well.

Method, definitions and the full list of things we never publish: methodology.