Transaction Timing Patterns and Trade Size Signatures

Why the gap carries the most information

The gap between one transaction and the next is the richest single signal in a trade record. Spontaneous flow arrives in bursts separated by long silences. Scheduled flow arrives with gaps drawn from a narrow band around a configured interval. That difference survives even when the two records show identical daily turnover, which is why spacing is read before size and long before totals.

The reason is structural. Turnover is a sum, and a sum discards the order and the spacing of the things it adds up. A gap distribution keeps both. Two markets can move the same value in a day while one did it in four frantic minutes and the other in eight hundred evenly spaced instalments. Only the second describes a machine, and a machine has no motive attached to it in the log. Cadence can say software. It cannot say why.

Irregular arrival versus scheduled arrival

Human-driven arrival is irregular in a specific way. Traders react to something they saw, so activity clumps around the trigger and then decays. A social post, a listing, a price move: each produces a burst, and between bursts the market is close to silent. The resulting gap distribution has a heavy tail, and no favourite value.

Scheduled arrival looks different because a scheduler has no attention to spend. It fires, waits, fires again. Even with jitter added, the gaps concentrate around the configured interval and the tail is truncated: there is a longest gap the schedule will ever produce, and the record does not exceed it. A firm ceiling on silence is unusual in a market driven by attention.

It is worth being direct about why cadence is legible at all. Schedulers are not hidden infrastructure; they are openly sold products with configurable intervals, amount ranges and session lengths, and an automated Solana volume bot exposes those controls as ordinary settings. A finite set of dials produces a finite set of rhythms, and a finite set of rhythms is what a distribution can recognise.

An illustrative arrival-gap profile

The table below uses invented round numbers, describing no real pair, no real market and no real operator. It exists to show the shape of the comparison rather than to report a measurement. The buckets are the ones a first pass usually starts with, and the last column matters most, because it states each row's limit.

Read the columns as shapes rather than as numbers. The scheduled column is concentrated and truncated; the spontaneous column is spread and open-ended. A real session producing the left column would carry weak timing evidence for automation whatever its turnover, and one producing the right column would carry evidence worth stating, at moderate confidence.

Size quantisation and narrow bands

Sizes are the second axis, and they quantise in two recognisable ways. The first is a short list of round values: the same handful of amounts repeating across many transactions, often in whole units rather than the odd fractions that fall out of a manual swap. The second is a tight band, where every amount sits between a floor and a ceiling, which is what a randomised range setting produces.

Both patterns are legible because sizes are stored in base units. An SPL token amount is an integer scaled by its decimals, and lamports are integer billionths of a SOL, so exact repeats are exact, not approximate. That makes an amount histogram cheap to build and hard to argue with. What it means is a separate question, handled below.

Round numbers alone prove nothing. People buy in round amounts constantly, and interfaces encourage it with preset buttons for a half or a whole unit. The informative case is not the presence of round values but the absence of everything else: a record where round values dominate and the scatter around them has gone. The same logic drives the clustering heuristics that group addresses behaving as one operator.

Slot timing and what blockTime cannot resolve

Solana has its own timing structure and it shapes what timing analysis can claim. The network targets roughly 400 milliseconds per slot, and transactions confirm inside slots rather than at arbitrary instants, so the finest grain any on-chain clock offers is a slot boundary. The protocol basics are set out in the Solana documentation.

The practical consequence is resolution. The blockTime value returned by RPC and shown by explorers is an estimated wall-clock timestamp in whole seconds, not a precise instant. Two transactions in the same slot can share a timestamp, and two seconds apart can be reported closer together than they were. Any claim resting on sub-second differences is unsupported, and should be dropped rather than hedged.

Slot numbers are the better clock when precision matters. The distance between two slots is an exact integer even though the wall-clock length of each slot varies with network conditions. Measuring cadence in slots removes the estimation problem and introduces another, since a constant slot gap is not a constant time gap. Say which clock you used.

Curve phase cadence and pool phase cadence

Launch venues add structure that a naive gap profile will misread. On a bonding curve, price is a function of supply along a fixed formula, so a run of small buys moves price in a smooth staircase and trades tend to land in dense early clusters while attention is highest. After migration to a standard pool, the same market prices against a reserve pair, and arrival changes with the participants.

That distinction matters for anyone comparing a launch window with a later one. Cadence during the curve phase is naturally faster and more clustered than pool phase cadence, and tooling built for the phase reflects it: a Pump.fun volume bot tool is configured around curve mechanics rather than pool depth. Spanning both phases without splitting them at migration produces a bimodal distribution belonging to the venue.

The correct handling is to cut the record at the migration point and profile each side separately. A single distribution across both phases will show two modes and invite the wrong conclusion. This does not prove that either phase was produced. It means the venue changed underneath the measurement, and a measurement that ignores that is measuring the venue.

Priority fees as a secondary timing signal

Fee settings sit alongside spacing and carry weaker, useful information. Solana charges a base fee of 5,000 lamports per signature, and a transaction may add a compute unit price to bid for earlier inclusion. That price is a client setting, which means it records a decision made once by whoever configured the sender rather than a market outcome.

A compute unit price that never varies across hundreds of transactions, through quiet periods and congested ones alike, suggests a fixed configuration. A price that tracks network conditions suggests a person watching, or a more capable client doing the watching. Neither observation decides anything, because ordinary wallets frequently keep a static fee setting.

Treat fee constancy as corroboration only. It is worth recording next to the gap profile because it arrives free with the same transaction fetch, and it occasionally separates two wallets that otherwise look identical. This desk would not raise a confidence band on fee constancy alone. The glossary keeps the field definitions used here.

The innocent explanations that defeat each reading

Every timing signature has a boring explanation that fits it perfectly, and a reading not tested against the list below is unfinished. These are not edge cases. They are the ordinary mechanics of how transactions reach a validator, and each manufactures the exact pattern a careless analyst would call produced volume.

• Retry and rebroadcast logic. Clients resend a transaction until it lands, and a failed leg may be retried after a fixed delay, stamping regular short gaps onto a record that had none.
• Cron-driven treasury operations. Scheduled rebalancing, payroll, buyback programmes and reward distributions all run on timers and all produce evenly spaced on-chain activity by design.
• Dollar-cost-averaging tools. A retail user splitting a purchase into fixed instalments produces textbook scheduled arrival with quantised sizes and no manipulation anywhere in the chain.
• Market maker quote refresh. A maker re-quoting on a timer produces two-sided flow at a steady cadence, which is exactly the shape a naive detector flags.
• Router leg splitting. An aggregator can break one user order into several venue legs that land within the same second, creating a burst that represents a single human decision.

Router behaviour deserves particular care because it inverts the count. Aggregators such as those documented at the Jupiter station route across multiple pools, so one intent becomes several trades. A gap profile built without collapsing legs into intents over-reports short gaps and under-reports silences, which biases the reading toward the automated column before the analysis has begun.

The desk does not publish guidance on making produced activity harder to detect, and the list above exists to protect readings from false positives rather than to be read backwards. Where a signature has an innocent explanation the data cannot exclude, the output is a lower confidence band, as in the round-trip analysis.

Reconstructing a gap profile from an explorer

A first-pass profile does not need custom infrastructure. A public explorer such as Solscan exposes enough to build one by hand for a modest window, which is the right scale for deciding whether a deeper pull is worth the effort. The sequence below is deliberately manual so every step can be checked.

1. Fix the window first. Choose a start and end time before looking at any data, and write down why those boundaries were chosen. A window selected after seeing the pattern is not evidence.
2. Pull confirmed trades only. Open the market or token page, filter to successful swaps, and discard failures. Failed attempts describe a client, not a market.
3. Collapse legs into intents. Where several transfers or venue hops share one signature or land in the same slot from the same signer, treat them as one arrival.
4. Record slot and timestamp together. Keep the slot number next to the blockTime for every arrival, so that fine spacing can be checked in slots when whole seconds are too coarse.
5. Compute consecutive gaps. Sort by slot, subtract each arrival from the one before it, and bucket the results using the bands in the table above.
6. Build the size histogram alongside. Tally trade amounts in base units and count exact repeats, since quantisation and cadence are stronger together than either alone.
7. Write the falsifier before the conclusion. State the observation that would overturn the reading, check whether the data you hold contains it, and record the answer.

The output of that sequence is a profile and a caveat, not a verdict. If it points somewhere, the next step is a reproducible pull rather than a louder claim, and the method for that is in the evidence collection notes. A timing profile earns a hypothesis. Nothing here identifies anybody.

Questions this desk is asked