How do you date a claim that has no timestamp?

By its contents rather than its metadata. The latest datable element inside a version sets the earliest that version could exist — archaeology calls this the terminus post quem. A note citing a figure released in March is a post-March document whatever date is on it. This gives you one direction of inference only: it tells you the earliest the claim could have formed, never the latest.

A thesis feels long-established but every citation for it is recent. What does that indicate?

Usually that it has been patched. A recent figure inserted into an older argument sits deep in the structure while being new, which reads as fresh confirmation of a durable thesis when what happened is repair. Archaeologists call the equivalent an intrusion — a trench cut through older layers and backfilled, sitting physically low and chronologically late.

Two analysts hold the same view. Did one take it from the other?

Shared conclusions prove nothing, because two people can reach the same correct answer independently. Shared errors are the evidence — the same rounded number, the same slightly wrong date, the same stale assumption. Textual critics have used shared error rather than shared correct reading as the test for common descent for two centuries.

Can you trace a market narrative back to its origin?

Rarely, and the attempt is where most provenance work goes wrong. The honest reconstruction is usually a partial order — some versions clearly precede others, several cannot be placed at all, and there is no root. A single chain looks like what an origin story should look like, which is why a wrong one is hard to notice.

What makes an assumption inside an argument a warning sign rather than a detail?

When it stopped being true and nobody removed it. A rate environment that no longer exists or a competitive set that has since consolidated gives an argument the appearance of having been tested over time, while being the component most likely to be wrong. In stratigraphy this is residuality — old material redeposited into a later layer, making the layer look older than it is.

Dating a Rumour by What It Picked Up

The diagnostic model

You are seeing

A claim circulates in several versions and no version cites a source
You can order the versions by content but not by time
A narrative feels long-established and every citation for it is recent
Two accounts share a detail that neither could have observed independently

The mechanism

A thing that travels accumulates detail from each place it passes through, so the accumulations can be read as a sequence even when nothing carries a date.

The older apparatus

Archaeology, transplanted into narrative transmission and from there into claim provenance. Folklore has the structure and treats it descriptively; archaeology built the formal method.

The false friend

Independent convergence. Two accounts can arrive at the same correct detail separately. Only shared errors and shared arbitrary specifics indicate shared descent.

The discriminating test

Find the latest datable element in each version. If a version contains something that did not exist before date D, that version is later than D regardless of how established it sounds or what date it claims.

On your own data

Take a recurring market claim, collect its variants across sources, and build a stratigraphic sequence from the datable elements each version carries.

Flinders Petrie had several thousand Egyptian graves and not one date for any of them.

He cut strips of cardboard, wrote each grave’s pottery types along the edge of a strip, and shuffled them around a table until the styles lined up into the longest continuous run he could manage.

That is seriation, and it works because a pottery style behaves like a fashion. It appears, spreads, peaks, and fades. Two graves holding the same mix of styles were filled around the same time. A grave holding styles that never otherwise appear together has been disturbed. Petrie could not date a single grave, but he could put all of them in order, and order was enough to build a chronology of predynastic Egypt out of a pile of undated pots. Brainerd and Robinson turned the shuffling into arithmetic in the 1950s, and correspondence analysis does it today, but the reasoning is Petrie’s and it did not require a calendar.

The problem he was solving is more general than Egyptian graves.

Folklorists inherited the same problem with worse records #

The Vanishing Hitchhiker is the story where a driver picks up a stranger, the stranger gives an address, and somewhere before arrival the passenger is gone. At the address, someone explains that the person died years ago.

Two folklorists collected 79 written accounts of it across the United States in the 1940s and sorted them into four version-types. Ernest Baughman gave the plot a motif number with lettered subtypes running from (a) to (j), depending on whether the passenger leaves an object behind, is later identified as a dead relative, or turns out to be a phantom nun. Jan Harold Brunvand traced American tellings back to the 1870s. The earliest known relative is in a Swedish manuscript from 1602, where the vanishing figure is a maid who has just delivered a prophecy about grain, war and pestilence.

The same skeleton has been recorded in England, Ethiopia, Korea, France, South Africa and Tsarist Russia.

Almost none of these tellings carries a date. What each one carries is detail, and the detail is local. The 1602 maid prophesies because that is what a supernatural encounter was for in 1602. The American versions have a car, which narrows things considerably. Folklorists have documented three axes along which a travelling story changes: the setting swaps, the roles get redistributed, and the moral gets reframed. The apple-shot legend keeps its roles and swaps its setting. The Hitchhiker keeps almost nothing but a passenger who vanishes and a subsequent revelation of death, and redistributes everything else — who prophesies, who investigates, who is warned.

Each of those changes is a deposit. The story is carrying a record of where it has been.

Which makes it a stratigraphic object, and archaeology has spent a century building tools that folklore uses informally and finance does not use at all.

The latest thing in the layer sets the floor #

The sharpest of those tools is the terminus post quem — the limit after which. A layer of soil containing a coin minted in 1340 cannot have been deposited before 1340. It might have been deposited in 1600. The coin does not tell you when the layer formed; it tells you the earliest it could have formed, and that single direction of inference does most of the work in field archaeology.

Applied to the Hitchhiker, a telling with a car in it is later than cars. That sounds trivial until you notice it is the only hard date in the entire corpus of variants, and it was obtained from the content rather than from any record of the telling.

Applied to a market claim, it is not trivial at all. A research note containing a figure that was released in March is a post-March document, whatever date is on it and however established the argument feels. A thesis that references a company’s segment reporting cannot predate the reporting change that created the segment. The claim’s own contents date it, and the contents are usually the only evidence available, because the thing being dated is not a document with a header. It is an argument circulating between people who did not write it down.

Some details were dug in later #

The complication that makes archaeological stratigraphy harder than the geological version is that people dig.

A drain trench cut through three earlier layers and then backfilled produces a context that sits physically low in the sequence and is chronologically recent. Read the depth naively and the trench fill dates to the layers it cuts through. This is why archaeologists record contexts — discrete events of deposition or destruction — rather than treating depth as a proxy for age.

There is a matching diagnostic. Unconsolidated material settles flat under gravity, so a tilted layer was either laid on an existing slope or has been disturbed since. The tilt is a warning that the sequence has been interfered with, available before any dating argument is made.

Both have narrative equivalents, and both are common.

An intrusion is a recent element inserted into an old story, sitting deep in the structure while being new. In market terms it is a fresh number retrofitted into a thesis that has been circulating for two years, which makes the thesis look freshly validated when what actually happened is that someone patched it. The argument reads as long-standing and confirmed. It is long-standing and repaired.

Residuality runs the other way. Old material gets redeposited into a later layer — a Roman coin in a Victorian foundation trench — and makes the layer look older than it is. The market version is an assumption nobody stripped out when the thesis was updated: a rate environment that no longer exists, a competitive set that has consolidated, a regulatory regime that lapsed. The stale element gives the argument a patina of long-tested durability while being the part most likely to be wrong.

Neither is detectable from the conclusion. Both are visible in the composition.

Nobody finds patient zero #

The last tool is the one that matters most and gets used least.

Edward Harris, working in Winchester in 1973, faced a site with thousands of contexts and no way to show how they related across the whole excavation rather than along a single trench wall. What he built is a directed graph — contexts as nodes, edges to whatever sits immediately above and below. Its important property is that it represents a partial order rather than a sequence. Two contexts with no physical relationship to each other appear as parallel branches with the relationship left undetermined, instead of being forced into a single false line.

That last part is the discipline. The matrix is designed to make it possible to say these two things are not ordered with respect to each other and stop there.

The instinct when tracing a market narrative is to find the origin. Who said it first, which note, which desk. Occasionally that works. Usually the honest reconstruction is a partial order — several tellings that clearly precede several others, a few that cannot be placed relative to anything, and no root. Forcing a single chain onto that produces a clean answer that is wrong, and the wrongness is invisible because a chain looks like exactly what an origin story should look like.

The archaeologists built a notation whose main feature is that it can decline to answer. Nothing in research provenance does that.

What this is worth #

Take a claim circulating in your market. Collect the versions, ignore what each one asserts, and record only what each one contains — which figures, which comparisons, which company facts, which vintage of assumption. Then sort.

The version with the most recent datable element is late. The version carrying an assumption that stopped being true in 2024 is either early or has been sitting in a drawer. Versions sharing an arbitrary specific — the same rounded number, the same slightly wrong date — are related, and the shared error is better evidence of that than any shared conclusion, because two analysts can reach the same correct answer independently and cannot independently make the same mistake.

What falls out is a sequence, and a position in it. Not who started it, which you will probably not recover, but whether the version that reached you is early or late.

Petrie could not tell you what year any of his graves was dug. He could tell you which ones came first, which turned out to be the useful question, because the ordering is what a chronology is made of and the dates were never going to arrive.

The ordering is also what tells you whether you are early.

If the version you are holding carries last week’s figure, forty people repaired it before it got to you, and it is in the price.