Assumptology Case Study: The Palme Murder, Part 2

Part II, Technical Evidence

Bullets, Weapon Inference, Trajectory, DNA, Test Shootings, and the Missing Artefact

If we want to reduce the suspect space, we should start with the most anchored layer of the web: technical evidence.

Not because technical evidence is infallible, but because it is the closest thing an investigation has to reality that does not depend on human memory. Physical traces still require interpretation, but they are less socially volatile than testimony and less vulnerable to political narrative than motive-theories.

So in this series we proceed in the natural Assumptology order:

Technical evidence (D-heavy): bullets, fragments, casings, trajectories, what is physically anchored and what is inferred.
Witnesses (noisy D + strong R): how identification becomes a suspect.
Grand theories (A-driven): PKK and other explanatory regimes that expand the hypothesis space primarily because of background assumptions rather than new physical data.

This post is the first category: technical evidence, and in particular the single most important technical object in the Palme case:

the gun we never found.

1) What counts as data here?

In Assumptology terms:

D (Data) is what is observed/recorded: recovered bullets/fragments, clothing damage, wound paths, scene observations (including the presence/absence of cartridge cases), timestamps, distances, photographs, medical documentation.
The weapon is not D. The weapon is a hypothesis, a model built to explain D.

This distinction sounds trivial, but it is exactly where cases quietly drift into storytelling. The public starts speaking as if “the gun” is a fact, when the gun is precisely what the investigation does not possess.

So the correct posture is:

The bullet is data. The weapon is an inference.

2) D is not just the object, it’s the object plus its provenance

In a strict Assumptology sense, D is rarely “the thing” alone. It is the thing plus the chain by which the thing entered the web.

A bullet found at a scene is not just a projectile; it is a projectile with a provenance: when it was found, by whom, where exactly, under what conditions, and with what documentation.

This matters in the Palme case because a controversial upstream feature in many retellings is that key bullets were found by civilians rather than recovered during the initial forensic sweep. That doesn’t make them false, but it does lower the integrity of the data node relative to a tightly controlled chain of custody.

Assumptology’s lesson is blunt:

If the provenance of D is weak, every subsequent inference rule (R) and background assumption (A) is multiplying by a fraction.

So before we ask what the bullets imply, we should ask:

How stable is the bullet node itself?

3) The investigative flow is not one-way

It’s tempting to describe an investigation as a simple reverse chain:

bullets → gun → murderer

But real inquiries don’t run in straight lines. They loop, jump levels, and run in both directions.

The causal chain in the world is:

murderer → weapon → bullets → wounds → scene.

An investigation often runs that chain backwards:

scene/bullets → weapon hypothesis → suspect space.

And then it flips direction again:

suspects → possible access/ownership networks → attempt to locate candidate weapons → test and eliminate → refine suspects.

So the real structure is a feedback system:

D constrains hypotheses → hypotheses select targets → targets generate new leads → leads update D.

Assumptology makes this visible: every hop uses interpretive rules (R) and settles background assumptions (A) about what is “normal,” what is “rare,” and what is “worth pursuing.”

4) Stabilizing the bullets: how an object becomes “evidence”

In this case, the bullets carry enormous inferential weight. If you accept them as linked to the event, they constrain the weapon space. If you don’t, the entire technical scaffold becomes unstable.

That is why the first technical question is not “what gun?” It is:

How do we know we have the right bullets?

When key objects enter the evidential chain under imperfect conditions, the investigation must attempt to settle them via convergent links, binding them to the event through consistency with medical traces, clothing damage, scene coherence, and other physical observations.

One stabilizing link often discussed in summaries is that the recovered bullets were treated as consistent with lead fragments found in the victims’ clothing, a way of binding the projectiles to the event rather than treating them as free-floating objects.

The point here is methodological: technical evidence becomes action-guiding when it is stabilized by convergence, not when it is merely asserted.

5) From bullet to calibre: be precise about what is being inferred

Once the projectiles are accepted as linked to the event, investigators identify what they can about them. In many summaries, the bullets are described as Winchester–Western .357 Magnum, 158 grain, metal-piercing.

Even here, it’s easy to slide from bullet facts into weapon facts. A projectile’s characteristics constrain the weapon space, but they do not uniquely determine the firearm. The mapping has the following structure:

Projectile data (D) ──R (rules)──▶ Weapon Hypothesis │ A (Assumptions)

The conditional bar matters. If A changes (local killer vs travelling actor; conventional sourcing vs specialised logistics), the “same” D can support a different hypothesis profile.

6) .357 Magnum vs .38 Special, and the hidden rule of market normalcy

This is worth stating cleanly, because it’s a perfect example of A interacting with D.

A .357 Magnum revolver can typically fire .38 Special cartridges.
The reverse is not normally true in standard firearms: a typical .38 Special revolver is not designed to accept .357 Magnum cartridges, largely because the .357 case is longer (a deliberate design choice to prevent dangerous misuse).

So, if the ammunition is genuinely identified as .357 Magnum, the “standard configuration” mapping points strongly towards a gun chambered for .357 Magnum.

But there is a hidden interpretive rule here, rarely stated, but often decisive:

R (Conventional sourcing): the ammunition and weapon were obtained through ordinary commercial or criminal channels, and therefore reflect what is typical in those markets.

If you silently accept that rule, “metal-piercing .357 → standard .357 platform” feels natural, and non-standard branches are pruned as noise.

But that pruning is not forced by the bullet. It is driven by A: an assumed model of the perpetrator.

If A = “a standard actor,” conventional sourcing is a reasonable default.
If A shifts towards “a specialised actor with logistics,” then “market normalcy” is no longer safe to privilege.

The point is not that exotic logistics are likely. The point is that the missing weapon means you cannot mechanically close off branches; you can only assign weights. Under one A-regime, edge cases are properly ignored. Under another, pruning them becomes a mistake.

The “batch accuracy” trap: when high-tech becomes noise

In the 1990s, investigators reportedly invested heavily in lead-isotope analysis to connect the bullets to a particular “batch” or box of ammunition.

This is a perfect Assumptology example of hidden optimism in technical reasoning. The method only bites if a strong assumption holds:

A: the chemical/isotope profile is unique enough to act like a barcode.

If the same signature is common across millions of rounds, or varies unpredictably across production runs, then the “high-tech” data stream doesn’t narrow the space. It becomes noise dressed as precision.

So even technical science needs Assumptology: the question is not “is the test scientific?” but “what assumptions must be true for the test to be discriminative?”

7) Two closures: from weapon class to weapon identity

There is another closure step hiding inside the “find the gun” narrative.

Even if we accept the bullets as settled data, technical reasoning does not jump directly from “bullet” to “the murder weapon.” It moves through two distinct closure levels:

1) Class closure (coarse): A bullet’s calibre and construction constrain the class of compatible firearms and ammunition. This is where investigators talk about revolver-versus-pistol, calibre families, and what configurations are plausible under the governing assumptions.

2) Individual closure (fine): To claim that a specific seized gun is the murder weapon requires a different kind of evidence: a way to individualize. This is typically framed as a match between test-fired bullets and recovered bullets, but the Assumptology point is that this is never pure deduction. It is a probabilistic settlement under a method and a threshold.

So the statement “these bullets came from this gun” is not a fact in the same sense as “two bullets were recovered.” It is a licensed conclusion, a closure decision:

R: the matching procedure and its error model (what counts as a match, how ambiguity is treated)
σ: the confidence threshold at which a court, a lab, or an investigation is willing to call it “the same gun”
A: assumptions about uniqueness and stability (that signatures are sufficiently individual; that deformation and wear don’t erase discriminative features; that recovered bullets are suitable for comparison)

This matters because the gun was missing. Without candidate weapons in hand, the investigation could do a great deal of class closure, but it could not reliably achieve individual closure. And without individual closure, even strong suspicion remains structurally vulnerable.

In other words: the missing gun didn’t just remove an object. It removed a whole mode of closure.

8) “No casings found” is open-system data, a fragile constraint

Another observation often repeated in retellings is:

no cartridge cases were found at the scene.

Under ordinary expectations, that is consistent with a revolver, because revolvers retain cases rather than ejecting them automatically. But again: this is not the scene speaking. This is R, a rule connecting D to a conclusion.

In an open system, a public street, at night, with movement, weather, imperfect scene control, absence is weaker than presence. The non-observation of an object can mean many things: the object was never there, the object was missed, the object was moved, or the object was removed.

So “no casings” is best treated as an inference of exclusion, not an inference of inclusion. It can constrain the hypothesis space, but it is a fragile constraint.

There is also a competing rule-set that can produce the same D without requiring a revolver: casings could have been controlled or removed (by behaviour, by luck, or by design). Assumptology’s point is not to assert which is true; it is to keep the logic clean:

In an open system, the absence of an object should rarely be treated as a hard lock, especially when the missing object would have been small, mobile, and easy to overlook.

9) Test shootings as projection, and the invisible pruning event

Once investigators treat a weapon hypothesis as action-guiding, the next step is elimination: locate candidate weapons and rule them out.

One way to do that is through test shootings and technical comparison. In practice, this can include outreach to holders of legally owned weapons thought to be potentially relevant, requesting that firearms be submitted for test firing.

At first glance this sounds like a straightforward technical program. In Assumptology terms, it is larger:

A test-shooting program is not just a procedure. It is a web-shaping event.

Because it decides which parts of the hypothesis space will ever be systematically searched.

The geographic filter

A consequential feature, often noted in discussions of the case, is that the systematic effort was heavily Stockholm-centred. The investigation did look at specific guns of interest outside the region in particular episodes, but the broad net was not cast uniformly nationwide.

If the relevant legal weapon pool was on the order of many thousands nationwide (figures like “~10,000 registered .357 revolvers” are often cited), while only hundreds were ever test-fired across decades, then most of the legal pool was never technically cleared.

That gap is not just “lack of work.” It is structural pruning: the inquiry didn’t disprove that a gun from Malmö or Göteborg was used; it simply never built a system capable of clearing that space.

The local-killer assumption

This is where the hidden assumption comes into focus:

A: the perpetrator (or the source of the weapon) is local to Stockholm.
R: geographic proximity increases probability of involvement.

Reasonable as a heuristic, devastating as an invisible pruning rule. Branches are not eliminated by evidence, but by a prioritization regime needed to cope with volume.

If the killer was a travelling actor, or the weapon was sourced from another region, then a Stockholm-centred test-shooting strategy becomes a category error: the right object in the wrong data silo.

The confidence threshold for “must be tested” is rarely applied uniformly. Status, bureaucracy, and perceived “type” quietly affect what is treated as necessary versus optional.

Whether or not any single anecdote is perfectly remembered, the structural point is stable:

σ is not only legal; it is social. The threshold for demanding a technical elimination is filtered through status, ease, and institutional friction.

The legal boundary condition

Finally, police powers are not unlimited. A “fire-everything nationwide” policy runs into legal thresholds. In many systems, testing privately owned firearms requires consent or a warrant; and “you own the same model class” may not be sufficient for compulsory seizure without additional suspicion.

So the scope of the test-shooting program is shaped by:

σ (legal thresholds): what counts as sufficient suspicion to compel testing
R (procedural rules): how requests, warrants, and compliance are handled
A (background heuristics): which owners are treated as relevant enough to pursue

All of this belongs to the technical layer, not because it is ballistics, but because it governs which technical comparisons are ever allowed to happen.

10) The signal in non-compliance

Test shooting programs generate a second-order effect:

They create new data: who complies, who doesn’t, and why.

In theory, non-compliance can be innocent: fear, distrust, confusion, absence, bureaucracy. In practice, it becomes an investigative signal, because the inquiry is forced to manufacture constraint wherever it can.

Assumptologically, this is where technical work blends into interpretive work:

D: requests for test shooting; records of response/non-response
R: rules about what non-response should mean (suspicion? irrelevance? nothing?)
A: assumptions about how ordinary citizens behave when contacted by police
σ: what level of suspicion warrants escalation

The key point is not that “people who didn’t show up are guilty.” It’s that missing artefacts force investigations to treat administrative behaviour as evidential proxy.

11) Distance, trajectory, and shooter geometry: constraints, not fingerprints

Technical reconstruction can also generate constraints about distance and positioning.

In general forensic work, close-range firing can often be inferred from clothing/tissue patterns and residue behaviour. Trajectory and wound-path analysis can sometimes constrain shooter position, although real street scenes rarely yield crisp geometric certainty.

The important methodological point is that “geometry” is not pure D either. It is produced by interpretation rules (R) under posture and movement assumptions (A).

With that caveat, technical reconstructions have often been described as consistent with a shooter of roughly average-to-tall height, sometimes summarised around ~180 cm. But the number matters less than what it is: a constraint generator.

A height estimate is not an identity. It is a boundary condition.

12) DNA and the moving boundary of technical evidence

Technical evidence is shaped not only by what was left at the scene, but by what technology can extract from what was left.

In the Palme case, some witness accounts describe the shooter being close enough to make contact, often retold as a hand placed on the shoulder just before the shots. If that kind of contact occurred, it raises a natural modern question:

Could DNA have been recovered?

The crucial Assumptology point is not whether a specific sample exists. It is that the space of possible technical evidence changes over time. In 1986, forensic biology and DNA profiling were early relative to today. Many traces that would now be considered potentially analyzable, touch DNA, low-template samples, were either not recoverable, not recognised as valuable, or not processable with high confidence.

But there is a deeper layer: even if usable traces existed, whether they can be analysed decades later depends on how evidence was handled at the time.

This is the temporal trap that haunts pre-1990 cold cases:

the 1986 team optimised for their closure regime (fingerprints, immediate procedures),
and in doing so, they often de-optimised for future regimes (DNA sensitivity, contamination control).

Contamination, in that sense, is not just “a mistake.” It’s the collision of two eras of σ: two different standards for what counts as extractable and what must be preserved.

The background assumption is recognizable:

A (often implicit): today’s forensic limits are close to the permanent limits.

Once you recognise that technology advances, evidence handling becomes a choice about which future branches you keep open, or quietly destroy.

13) The missing artefact and why it dominates everything

Put the pieces together:

D gives you projectiles, clothing damage, scene observations with provenance and integrity profiles.
R turns those into working constraints (calibre-consistent, revolver-consistent, possibly trajectory-consistent).
A governs the pruning: local killer vs travelling actor; conventional sourcing vs specialised logistics; “absence as strong evidence” vs “open-system fragility.”
σ shapes what gets preserved, compelled, and treated as “good enough”, both in 1986 and decades later.

And then the critical fact:

The weapon was never recovered.

That means the central technical hypothesis can rarely be closed mechanically. It remains action-guiding, but not action-ending.

Here’s the technical paradox in plain English: the case is rich in constraints, but poor in closures, because the central bridge object, the gun, never enters custody.

This also reframes the “missing artefact” problem in a sharper way:

The gun is not necessarily a needle in a haystack. It may be a needle in one of ten haystacks, and the investigation only ever systematically searched one.

If the killer was a travelling actor, or the weapon was sourced from another region, then even a diligent Stockholm-centred technical effort can become the wrong search in the wrong silo.

The technical lesson

The weapon question is not just “what gun was it?”

It’s: what kind of closure is even available? Much of the technical work yields class closure, constraints on what could have produced the bullets. But the conclusion people intuitively want, individual closure, the claim that a particular gun is the murder weapon, is a probabilistic settlement governed by rules (R) and thresholds (σ).

And the refinements sharpen it further:

D includes provenance. Weak collection integrity multiplies downstream uncertainty.
High-tech can become noise if its discriminative power rests on optimistic assumptions (the “batch barcode” trap).
Absence is fragile in open systems.
Technical programs are not neutral. Their geography, legal thresholds, and social σ perform invisible pruning.
Evidence handling is time-indexed. The 1986 closure regime de-optimised for the future one.

This is what makes the Palme case such a strong Assumptology example: even before witnesses and theories, the technical layer already shows how inquiry is shaped by hidden assumptions, non-uniform thresholds, and time.