Method: how a surge class is defined and defended
A classification scheme is only worth anything if the definitions are fixed before the data arrives, if each class states what evidence it needs, and if every label ships with the finding that would remove it. This note sets out all three, plus the sentences this desk will not write under any circumstances.
How a class is defined
A surge class on this site is defined by the evidence that would put a window into it, not by the story that sounds most plausible afterwards. Each of the six classes carries three fixed parts: a one-line description of the mechanism, a required evidence set, and the observation that moves a window out of that class and into another one.
Definitions were fixed before any window was read, and they are deliberately coarse. Six classes is few enough that a reader can hold them all in mind while a chart is moving, and few enough that the boundaries can be stated in a sentence each. A finer scheme would look more rigorous and would be harder to defend, because most extra distinctions cannot be settled from a transaction record.
Two of the six overlap in practice and are expected to. Routing and produced turnover both generate activity that returns inventory to roughly where it started, so a window can sit on that boundary legitimately. When it does, the correct output names both classes rather than choosing the more interesting one.
The class matrix
The table below is the whole scheme. The third column is the part that does the work: it is written at the same time as the class, before anyone has had time to become attached to the reading.
| Class | Evidence required | Moves out of this class when |
|---|---|---|
| S1 Attention | A high share of trading addresses never seen in this pair before, a counterparty tail that keeps growing through the window, and trade sizes spread across a wide range. | Reclassify if the address set stops expanding while turnover holds, or if most addresses trace to a small funding set. |
| S2 Plumbing | Pool initialisation, liquidity added or removed, a migration transaction, or a second venue appearing inside the same window as the turnover. | Reclassify if reserves and pool membership are unchanged across the window, which leaves the turnover unexplained by structure. |
| S3 Distribution | Large one-sided flow against a small number of accounts, a measurable shift in balance concentration, and net inventory that ends the window clearly directional. | Reclassify if the same accounts finish the window holding what they started with, which points at routing or produced flow instead. |
| S4 Routing | Legs that open and close within a slot or two, aggregator and arbitrage programs carrying a large share of the flow, and inventory that returns to flat. | Reclassify if inventory ends directional and the counterparty set is wide, which is a demand shape rather than an intermediation shape. |
| S5 Produced | Trade sizes drawn from a narrow band, spacing that stays inside a tight range regardless of the hour, and funding paths that converge on a small number of sources. | Reclassify if size dispersion is wide and the address tail keeps lengthening, which no fixed configuration produces on its own. |
| S6 Echo | A second burst that lags the first by a consistent interval, arrival timing clustered near round minute boundaries, and participants with no prior history in the pair. | Reclassify if the burst leads rather than lags the first move, since an echo cannot precede the thing it echoes. |
Nothing in that table is a detection algorithm. Each row describes what a window would have to look like for the class to be defensible, which is a much weaker claim than saying the class has been proved. A window can satisfy a row and still belong somewhere else, which is precisely why the third column exists.
What counts as evidence
Evidence here means a field that can be read from public transaction data and read again by someone else from the same window. That restriction rules out a great deal of what usually gets treated as evidence in market commentary, and the exclusions matter more than the inclusions.
- Signer and fee payer, which on Solana need not be the same account, so both are recorded.
- Slot and block time, which is what makes any statement about spacing or lead-lag possible.
- Pre and post token balances, because the amounts that actually moved are more reliable than the amounts an interface displayed.
- Program identifiers, which say which venue executed the swap and whether one user action became several legs.
- Pool reserves at the start and end of the window, which is the only way a plumbing event can be separated from a behavioural one.
- Address first-seen status inside the pair, which is a derived field and is recorded as derived.
Screenshots, aggregate figures with no stated window, dashboard totals whose deduplication rule is unpublished, and anything relayed second hand are treated as leads rather than evidence. A lead can start a reading. It cannot support one.
Fixing the window first
Every reading on this site begins by fixing a window and a comparison period, and both are stated before any field is measured. This is not bureaucracy. Turnover figures are extraordinarily sensitive to window choice, and an analyst who picks the window after seeing the data can produce almost any conclusion without writing a single false sentence.
The comparison period matters just as much as the window. A spike is a ratio, and a ratio needs a denominator. A pair that traded almost nothing for six hours will show an enormous multiple on very ordinary activity, which is a statement about the previous six hours rather than about the current ten minutes.
What would change this practice
If windows chosen after the fact produced the same class assignments as windows fixed in advance across a broad sample, the discipline would be costing effort for nothing. Until someone shows that, the desk assumes window choice is doing real work and states it up front.
How confidence is expressed
Confidence is expressed in four named bands and nothing else. No percentages, no scores out of ten, no probability figures, because a number implies a calibration exercise this desk has not carried out and could not honestly publish.
| Band | Meaning and requirement |
|---|---|
| Undecidable | The available fields do not separate the candidate classes at all. Published as undecidable, never rounded up to the nearest interesting class. |
| Provisional | One field points somewhere. Enough to write down, not enough to act on, and expected to move when a second field is measured. |
| Working | Several fields agree and their ordinary explanations are not compatible with each other. This is the highest band most live readings reach. |
| Firm | The class is forced by a recorded structural fact, such as a pool being created inside the window. Reserved for facts, not for inferences. |
The band is chosen by what was measured, never by how interesting the conclusion is. Firm is deliberately hard to reach: it applies to structural facts recorded on chain, such as a pool being initialised inside the window, and inference never reaches it however many fields agree with each other.
Most live readings land on Provisional or Working, and a good number end at Undecidable. That distribution is the honest result of reading short windows with incomplete venue coverage, and a scheme where everything reaches the top band would be evidence that the bands are not doing anything.
What this desk will not claim
The refusals are as much a part of the method as the classes, because most of the damage in this field comes from claims that were never supportable in the first place. These are absolute rather than case by case.
- No named token, wallet, team, launchpad or venue is described as manipulating anything. Intent, identity and ownership are absent from the record, and a wrong accusation about a named party cannot be repaired by a correction.
- No price expectation of any kind. No targets, no entries, no exits, no directional language dressed up as analysis.
- No frequency claim without a measurement behind it. Words like usually, typically and most of the time are treated as numeric claims and are removed unless a cited figure supports them.
- No trading signal, no watchlist and no ranking of assets by anything.
- No guidance on making produced activity harder to detect, in any framing.
- No confident single-field verdict. Every field on this site carries an ordinary explanation, and a reading resting on one field is written as Provisional at best.
Why there are no base rates here
The most common question this desk receives is what proportion of Solana volume spikes belong to each class. The answer is that this desk does not know, and the reason is worth being precise about rather than apologising for.
A base rate requires a defined population, a sampling method that does not favour the interesting cases, complete venue coverage across that population, and a labelling process whose error rate is known. None of those four conditions is met by reading windows that came to attention because they looked unusual. A figure produced from such a sample would describe the selection process rather than the market.
Because no base rate is available, arguments on this site are written so they do not depend on one. Where an argument genuinely needs a frequency to work, it is either dropped or the missing frequency is named in the paragraph, so a reader can see exactly which step is unsupported. Inventing the number would be the same failure this site exists to describe.
Corrections and revision
Notes are edited when they are wrong. The modified date on each page is real and moves when the text changes, so a reader who cites a page should cite the date they read it. Nothing is quietly deleted, and a class assignment that turns out to have been wrong is more useful published with its correction than removed.
A correction that names the page, quotes the sentence and states what should replace it will be acted on faster than a general disagreement, because it can be checked the same day. The route is on the contact page, and the standing scope of the desk is described in the note about the desk.