Perp venues leak position changes; spot venues hide them

• 51 of 61 published v2.0 signals originated on a perp venue; 10 on spot chains.
• Perp position changes are explicit, sized and directional — legible almost immediately.
• Spot decisions fragment across routed swaps, so the same conviction must clear a higher bar of evidence before publication.
• The uneven chain mix reflects a constant evidence standard, not uneven attention.
• Rotation dominates the perp output; spot output so far skews to distribution and inflow.

The observation

We crawl four chains continuously and label wallets on all of them without preference. The published output does not reflect that symmetry. Of the first 61 signals from the v2.0 engine, 51 came from a perpetuals venue and 10 from spot chains, with the perp share dominated by rotation and the thin spot share skewing toward distribution and inflow.

The instinct is to read this as a coverage gap. It is not, or at least not primarily. The same gates were applied everywhere; they were simply easier to satisfy in one place than the other.

Why perps are legible

On a perpetuals venue, the position is the primitive. Opening, adding to, reducing and closing a position are each explicit events with a size and a direction attached, and they are recorded as such. There is no interpretive step between the transaction and its economic meaning.

That has a direct consequence for detection: when several independent participants lean the same way inside a short interval, the pattern is apparent almost immediately, and the evidence needed to establish it is exactly what the venue already published. Leverage sharpens it further — conviction expressed with borrowed size is louder than the same conviction expressed with cash.

Why spot hides the same decision

A single spot decision rarely looks like a single transaction. It arrives as a routed path through one or more pools, sometimes through an intermediate asset, often deliberately split to manage slippage. The economic event — one actor decided to move capital from A to B — has to be reconstructed from fragments before it means anything.

Reconstruction is tractable but it is not free. Every inference introduces the possibility that we have joined together transactions that do not belong together, or split apart transactions that do. A filter that is honest about that uncertainty must demand more evidence before publishing a spot pattern than an equivalent perp pattern. That is precisely what ours does, and the uneven output is the visible result.

We would rather the mix stay uneven than flatten it by lowering the spot bar. A balanced-looking feed assembled from weaker inferences is worse than an unbalanced one assembled from strong ones, even though it presents better.

What this means for a trader

Treat the two venue types as different instruments rather than duplicate coverage of the same market.

Perp flow tends to lead. Leverage lets a view express itself before spot capital has to be reallocated, so directional crowding often appears there first. This is the flow to watch for anticipation, with the corresponding caveat that leveraged conviction can be unwound as quickly as it was expressed.

Spot flow tends to confirm. When capital actually rotates on spot, the participant has accepted the cost and the holding period; the decision is slower and stickier. This is the flow to watch for durability.

The sequence is the real signal. A name that appears first from the perp venue and then, hours later, from spot has been endorsed twice by different kinds of capital. That pairing is more informative than either alert read alone, and it is the pattern we would look for as the spot side of the feed deepens.

Limitations

The sample is small and young: 61 published signals across roughly one day of operation. The chain mix will move as spot coverage matures, and readers should not treat today's ratio as a permanent property of the system.

This note describes venue structure, not detection technique. It deliberately omits how spot paths are reassembled, how participants are judged independent, and every threshold involved in either — those remain internal. It also makes no claim about the performance of the signals discussed; no measured outcome sample existed at time of writing.