The Gold Standard of Archaeology | Taphonomic Surface Modification, Equifinality and Diagnostic Certainty

1. Begin with the trace, not the story

Back to contents

“Linear groove, 8.2 mm long, V-shaped in part of its cross-section” is an observation. “Stone-tool cut mark” is an attribution. Keeping those statements separate prevents an attractive behavioural story from entering the catalogue as if it were measured fact.

Record element, side, anatomical zone, orientation, dimensions, surface preservation, lighting, magnification and imaging method before assigning an agent.

2. Equifinality is a practical identification problem

Back to contents

Equifinality means different processes can produce similar outcomes. Trampling, crocodile teeth, stone edges and preparation tools can all create linear or puncture-like surface modifications. Similar appearance is therefore not sufficient evidence of identical cause.

A diagnosis gains strength when several independent features converge and plausible alternatives are actively tested. It weakens when it depends on one memorable shape.

3. Build an explicit candidate set

Back to contents

For each class of trace, list the candidate agents relevant to the deposit: butchery, percussion, carnivore tooth action, rodent gnawing, root etching, sediment abrasion, weathering, excavation damage and conservation or preparation marks. Add candidates supported by the site history.

The list disciplines attention. A laboratory air scribe cannot explain a trace documented in situ before preparation; conversely, preparation damage must remain plausible when imaging occurred only after cleaning.

4. Use morphology as a bundle of features

Back to contents

Classifiers should combine cross-section, shoulder effects, internal microstriations, wall shape, edge crushing, polish, branching and repeated scalloping where relevant. Feature definitions must be operational enough that a second observer can score them.

No feature is automatically universal. Tool raw material, bone portion, freshness, sediment and later alteration can change mark expression. Reference collections should span the conditions actually under discussion.

5. Context and chronology constrain the diagnosis

Back to contents

Cross-cutting relationships, staining and weathering can order events. A pale groove cut by a darker root trace differs from a bright fresh scratch that crosses an older patina. Spatial clustering near muscle attachments may support butchery; random distribution on exposed surfaces may support abrasion.

These are likelihood-changing observations, not shortcuts. Record the context before interpretation so the inference can be rechecked.

6. Worked example: from marks to bounded attribution

Back to contents

An assemblage has 120 observable cortical fragments. Twenty-four carry linear grooves. After blind rescoring, 15 meet the preregistered cut-mark feature bundle, 5 are ambiguous between cut and trampling, and 4 fit modern excavation damage.

The defensible result is not “20 cut-marked bones.” A lower bound is 15/120 = 12.5%; an inclusive sensitivity bound is 20/120 = 16.7%. Report both, plus the five ambiguous specimens and the criteria that would move them.

7. Measure observer agreement without confusing it with truth

Back to contents

Blind duplicate scoring can reveal whether a category is reproducible. Agreement on 108 of 120 fragments is 90%, but raw agreement can be inflated when most fragments are unmodified. A confusion table by category is more revealing.

High agreement means analysts apply the rule similarly. It does not prove that the rule identifies the true agent; experimental references and contextual tests address validity.

8. Changed-context transfer check

Back to contents

Move the problem from faunal remains to a human skeletal collection cleaned decades ago. The same differential workflow transfers, but the candidate set changes: medical handling, curation labels, consolidants and preparation tools may become more important.

The transfer succeeds when the observation-attribution separation and certainty scale remain intact while the reference set and site-history evidence change.

9. Limits and common misconceptions

Back to contents

Microscopy does not eliminate equifinality. Machine-learning accuracy on an experimental image set does not automatically transfer to a different substrate, preservation state or imaging system. Absence of diagnostic marks is not proof that an action never occurred.

Do not turn an assemblage-level pattern into a specimen-level certainty. Preserve raw images, ambiguous classes and the full denominator of observable surfaces.

10. Audit images, labels and denominators

Back to contents

A practical audit links each specimen ID to photographs at declared scales, scored features, observer decisions, preparation history and final certainty category. Randomly re-image a subset and review disagreements without showing analysts the preferred interpretation.

Denominators matter: 15 diagnostic marks among 120 observable cortical surfaces is not the same as 15 among all 260 fragments, many of which may be unobservable.

11. Frequently asked questions

Back to contents

Does a V-shaped groove prove cutting?

No. It is one feature within a differential diagnosis and can be mimicked under some conditions.

Should ambiguous marks be discarded?

No. Keep them as an explicit class and show sensitivity bounds with and without them.

Can artificial intelligence settle the attribution?

It can improve reproducibility on validated data, but transfer to a new assemblage still requires testing.

12. Certainty should be earned specimen by specimen

Back to contents

The strongest archaeological account is often narrower than the most dramatic one. It distinguishes observed trace, plausible agent, ruled-out alternatives and remaining uncertainty, then aggregates only compatible categories.

Before publication, ask whether another team could reconstruct every attribution from the stored images, feature definitions and context. If not, the story has outrun the evidence trail.