The Gold Standard · Archaeology
Make the decision path visible
A bridge model can restate a historical series, but its uncertainty is shared. If the same slope, offset or reference-material relation changes a hundred values, their errors do not behave like a hundred independent coin tosses. Treating them as independent can make a trend or group difference look more precise than it is.
The gold standard traces every component, carries covariance through the reanalysis and judges adequacy at the level of the claim: classification, trend, threshold, provenance group or conservation decision.
Choose the route closest to your decision
Chapter contents
Chapters 1–3
Chapters 4–6
Chapters 7–9
Chapters 10–12
CHAPTER 1 OF 12 · DEFINE THE BUDGET
1. Begin with the claim and measurement model
Write the relation that maps raw or old-era observations to the restated value. Name the output quantity and the archaeological claim that will use it.
An uncertainty budget for one concentration is not automatically sufficient for a slope across decades or a classification boundary. The claim determines which joint uncertainty matters.
CHAPTER 2 OF 12 · DEFINE THE BUDGET
2. List every material uncertainty component
Include original measurement, reference values, lot comparison, bridge parameters, residual variation, matrix mismatch, preparation effects, software and any imputation for missing historical inputs.
Separate uncertainty from known correction. A documented bias should be corrected where defensible; the uncertainty describes what remains incompletely known.
Practical check
Write the decision, evidence and limit in three separate sentences. If one sentence depends on an unstated assumption, name it before continuing.
CHAPTER 3 OF 12 · DEFINE THE BUDGET
3. Mark shared, local and hierarchical components
Local components vary independently by specimen or run. Shared components affect many values together, such as one reference certificate, one bridge slope or one software correction. Hierarchical components may be shared within lot, matrix or period.
Draw the dependency structure before calculation. This prevents a common parameter from being counted once per value as though new information appeared with every record.
| Component | Structure |
|---|---|
| Original reading | Mostly local |
| Bridge slope | Shared across era |
| Lot value | Shared within lot |
| Matrix effect | Shared within class |
CHAPTER 4 OF 12 · PROPAGATE CORRECTLY
4. Estimate covariance as well as variance
Use the joint fit or resampling procedure to retain covariance among slope, intercept and other parameters. Include correlation among reference values or repeated materials when known.
Ignoring covariance can either inflate or shrink uncertainty. The direction depends on the model and where in the measurement range the value sits.
CHAPTER 5 OF 12 · PROPAGATE CORRECTLY
5. Propagate through the whole reanalysis
For linear models, analytic propagation may be transparent. For nonlinear bridges, censored values or complex claims, use Monte Carlo simulation with the documented joint distributions.
Preserve the same sampled bridge parameters across all records within each simulation draw. Sampling a new slope for every specimen destroys the shared structure.
Practical check
Write the decision, evidence and limit in three separate sentences. If one sentence depends on an unstated assumption, name it before continuing.
CHAPTER 6 OF 12 · PROPAGATE CORRECTLY
6. Worked example: a shared slope changes a trend
Suppose restated value equals 0.4 plus 1.035 times the old value. The slope uncertainty is 0.012, the intercept uncertainty 0.25 and their correlation is negative. Twenty historical values share those parameters.
A naïve calculation that treats bridge error independently for each point makes the series trend look extremely precise. Joint simulation shows the whole early era can tilt together, widening the trend interval. Changed-context check: a within-era rank may remain stable because shared shifts cancel, while a comparison across the bridge boundary can remain sensitive.
| Situation | Method |
|---|---|
| Linear scalar | Analytic propagation |
| Nonlinear bridge | Monte Carlo |
| Many records | Joint simulation |
| Hierarchical effects | Structured model |
CHAPTER 7 OF 12 · TEST THE CLAIM
7. Build a contribution and sensitivity view
Report which inputs drive uncertainty for low, middle and high values and for the final claim. A slope can dominate at the upper range while residual scatter dominates near the centre.
Sensitivity is diagnostic, not a licence to delete inconvenient components. It helps decide whether better overlap data, matrix-specific evidence or remeasurement would reduce consequential uncertainty.
CHAPTER 8 OF 12 · TEST THE CLAIM
8. Evaluate threshold and classification stability
Calculate the probability or plausible range that each restated value falls on either side of a decision threshold. For groups, account for common bridge movement before counting classifications.
Avoid hard relabelling when a value straddles the boundary. Report the classification as stable, sensitive or unresolved under the current budget.
Practical check
Write the decision, evidence and limit in three separate sentences. If one sentence depends on an unstated assumption, name it before continuing.
CHAPTER 9 OF 12 · TEST THE CLAIM
9. Test trends, contrasts and provenance groups
Re-run the archaeological analysis within the joint uncertainty model. Check whether trend direction, group separation, clustering or source attribution changes across credible draws.
The relevant limit may be qualitative: the broad pattern survives, but the timing or magnitude does not. State that boundary rather than demanding one binary verdict.
| Claim | Joint question |
|---|---|
| Threshold | Can values cross together? |
| Trend | Can an era tilt? |
| Group contrast | Does shared error cancel? |
| Provenance | Do clusters persist? |
CHAPTER 10 OF 12 · PUBLISH AND MONITOR
10. Version the budget with the bridge
Store parameter estimates, covariance matrix, component definitions, distributions, software, input identifiers and effective date. A revised bridge without a revised budget is incomplete.
Keep earlier budgets so published results can be reconstructed. Name which datasets and claims use each version.
CHAPTER 11 OF 12 · PUBLISH AND MONITOR
11. Validate with new and held-out material
Compare predicted and observed paired results on held-out overlap materials, new runs and relevant matrices. Calibration of uncertainty means intervals behave credibly, not merely that point predictions look close.
Monitor residuals and coverage across range and matrix. A budget that fails prospectively needs revision before it supports stronger claims.
Practical check
Write the decision, evidence and limit in three separate sentences. If one sentence depends on an unstated assumption, name it before continuing.
CHAPTER 12 OF 12 · PUBLISH AND MONITOR
12. Set claim-level decision limits
Approve use only where the joint uncertainty is small enough for the intended claim. A series may support broad chronology while remaining inadequate for a narrow provenance boundary.
When shared uncertainty dominates a high-consequence conclusion, remeasure archived specimens or obtain new overlap evidence. Mathematical completeness cannot replace missing physical information.
| Record | Purpose |
|---|---|
| Budget version | Reproduce result |
| Covariance | Preserve dependence |
| Sensitivity | Target new evidence |
| Use limit | Bound the claim |
Frequently asked questions
Why not add every uncertainty in quadrature?
That shortcut assumes independence. Shared bridge parameters and correlated reference values require covariance or a joint simulation.
Can correlation reduce uncertainty?
Yes in some combinations, while increasing it in others. Preserve the estimated sign and magnitude rather than assuming zero.
Does one budget cover every archaeological claim?
No. The same values can support a stable rank but an unstable cross-era trend or threshold decision.
When is remeasurement preferable?
When shared uncertainty controls a high-consequence conclusion and new overlap evidence cannot resolve it.
Continue through verified routes
Core routes
Archaeology subject root · Gold Standard of Education · Gold Standard directory · How X Works hub
Specialist routes
- Bridge-model revision and versioned reanalysis
- Replacement-panel lot effects and mixed-epoch analysis
- Reference-material lot changes and bridging studies
- Authentic-sample panel depletion and longitudinal comparability