The Gold Standard · Archaeology
Make the decision path visible
Replacing depleted authentic specimens solves a material problem but creates an analytical one. The new panel may cover the same categories and concentration range yet differ in matrix behaviour, preservation history or preparation response. That panel-era difference can be mistaken for instrument drift—or hidden by an overconfident correction.
Longitudinal comparability depends on a visible bridge between epochs. The laboratory needs retained anchors, balanced overlap measurements, models that allow panel effects and a rule for when historical harmonisation is justified. Not every discontinuity should be corrected away.
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 PANEL EPOCHS
1. Treat each panel as a versioned lot
Assign a panel version whenever membership, aliquot preparation or intended use changes. A convenient label such as replacement panel is not enough; record the specimen list, preparation state and effective date.
Keep reference-material lots, authentic-sample panels and routine controls distinct. They may change at different times and contribute different effects to the measurement chain.
Link every reported result to the panel version used. Without that field, later analysts cannot distinguish a historical shift in specimens from a shift in method or instrument.
CHAPTER 2 OF 12 · DEFINE THE PANEL EPOCHS
2. State the historical comparison being protected
Comparability may mean stable laboratory proficiency, stable bias relative to authentic specimens or continuity of an archaeological trend. These are related but not identical claims.
Name the measurand, method family, matrix range and decision. A correction that is acceptable for a broad proficiency score may be unsafe for a threshold near an important interpretation.
Do not promise universal harmonisation. The bridge applies only to the ranges, matrices and procedures represented in the overlap evidence.
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 PANEL EPOCHS
3. Map possible epoch effects
A panel change can alter the mean, spread, concentration range, matrix interactions and proportion of difficult specimens. It can also coincide with analysts, software, digestion batches or calibration lots changing.
Build a dated change map for panel membership, method releases, instruments and environmental conditions. Coincidence does not prove cause, but hidden coincidence makes cause almost impossible to investigate.
Preserve specimen-level covariates such as material class, preservation and treatment history. Aggregate panel means can conceal a shift concentrated in one matrix.
| Field | Purpose |
|---|---|
| Panel version | Identify membership |
| Method release | Separate process change |
| Matrix class | Expose interactions |
| Effective date | Locate the discontinuity |
CHAPTER 4 OF 12 · MEASURE LOT EFFECTS
4. Design a balanced overlap study
Measure surviving old members and candidate replacements across the same runs, operators and procedures. Randomise or balance order so panel membership is not confounded with run position.
Include stable controls and enough replicates to estimate repeatability. One side-by-side run cannot separate a persistent panel effect from ordinary measurement noise.
Pre-specify exclusions, transformations and comparison limits. Removing an awkward replacement after seeing its result may improve the bridge while making the panel less representative.
CHAPTER 5 OF 12 · MEASURE LOT EFFECTS
5. Diagnose level, slope and matrix interaction
Begin with paired plots and residuals. A constant offset suggests a different remedy from a concentration-dependent slope or a method-by-matrix interaction.
Use retained anchors to test whether the measurement process itself moved. If anchors and replacements shift together, investigate method drift; if only replacements differ, investigate selection, preparation and commutability.
Avoid reducing every difference to one correction factor. Stratified or nonlinear relationships may require a narrower comparison claim rather than a more elaborate adjustment.
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 · MEASURE LOT EFFECTS
6. Work a mixed-epoch panel case
Suppose Panel A supported eight years of elemental measurements. Panel B replaces 10 depleted specimens. In a balanced overlap, retained anchors differ by 0.3 percent between periods, but Panel B's high-carbonate specimens read 4.8 percent higher under one digestion.
The laboratory keeps the raw epochs separate, estimates a matrix-specific bridge for the affected digestion and reports a larger uncertainty for adjusted historical comparisons. It does not shift results from unaffected matrices.
Changed-context transfer check: if the new panel intentionally includes weathered material absent from Panel A, the difference may represent a broader target population rather than a lot artefact. That subgroup needs a new stratum or series, not forced harmonisation.
| Element | Control |
|---|---|
| Retained anchors | Connect epochs |
| Balanced runs | Avoid time confounding |
| Stable controls | Detect method drift |
| Replicates | Estimate repeatability |
CHAPTER 7 OF 12 · HARMONISE CAREFULLY
7. Choose a harmonisation model that matches evidence
Options include no adjustment with an epoch indicator, a constant offset, regression calibration, matrix-specific factors or a hierarchical model. Select the simplest model supported by the overlap design.
Test predictions on held-out bridge observations where feasible. A model fitted and judged on the same few specimens can appear exact while transferring poorly.
Define the reference epoch deliberately. Adjusting old results to the new panel and adjusting new results to the old panel answer different operational questions.
CHAPTER 8 OF 12 · HARMONISE CAREFULLY
8. Carry bridge uncertainty into conclusions
Adjusted values are estimates, not repaired observations. Propagate uncertainty from repeatability, panel assignment, model coefficients and residual matrix variation.
Show when interpretation changes because of the bridge. If a trend is visible only after adjustment and is similar in size to bridge uncertainty, report that fragility directly.
Do not improve apparent precision by treating estimated correction factors as fixed. Historical harmonisation usually adds uncertainty even when it reduces discontinuity.
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 · HARMONISE CAREFULLY
9. Preserve raw, epoch and harmonised layers
Keep original results immutable, store panel version and produce harmonised values as a derived layer with equation, software version and date. Never overwrite the historical record.
A user should be able to reproduce both the contemporary report and the later comparison. This protects provenance when the bridge model is revised.
Publish a clear flag for values outside the validated bridge range. Extrapolated harmonisation must not look identical to interpolation supported by overlap specimens.
| Pattern | Possible treatment |
|---|---|
| No detectable shift | Epoch flag only |
| Constant offset | Additive bridge |
| Range-dependent shift | Regression bridge |
| Matrix interaction | Stratify or narrow claim |
CHAPTER 10 OF 12 · GOVERN LONG SERIES
10. Monitor the replacement panel as it ages
Panel B can itself drift through depletion, storage change or repeated opening. Continue stability checks and reserve material for future overlap with Panel C.
Use control charts as prompts for investigation, not proof that the panel is stable or unstable. Review specimen-specific patterns and the state of retained anchors.
Update the quantity budget after every bridge study. Comparability work consumes the scarce material needed for the next transition.
CHAPTER 11 OF 12 · GOVERN LONG SERIES
11. Control mixed-epoch analysis
Analysis plans should name whether panel epoch is a covariate, stratification factor or basis for sensitivity analysis. Omitting it assumes away the discontinuity.
Provide both unharmonised and harmonised sensitivity results for important claims. Agreement supports robustness; disagreement directs attention to the bridge rather than being hidden.
When sharing data, include the panel dictionary and correction metadata. A derived value without its epoch and method can be reused beyond the evidence that justified it.
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 · GOVERN LONG SERIES
12. Know when histories should remain segmented
Do not join epochs when overlap is absent, panel coverage changed materially, method interactions are unresolved or the correction dominates the signal of interest.
A segmented series can still support within-epoch trends, qualitative comparison and future reanalysis. Honest discontinuity is more useful than false continuity.
The final assurance record should connect depletion decisions, panel versions, overlap design, bridge model, uncertainty and interpretive limits. Long-term comparability is an evidence chain, not a cosmetic alignment.
| Layer | Rule |
|---|---|
| Raw | Never overwrite |
| Epoch-labelled | Attach panel and method |
| Harmonised | Store model and uncertainty |
| Extrapolated | Flag outside validation range |
Frequently asked questions
Is a replacement panel a new lot even if the method is unchanged?
Yes. Changed specimen membership can create a panel-era effect and must be versioned and tested.
Can one correction factor harmonise every specimen?
Only when overlap evidence supports a common relationship across the relevant ranges, matrices and methods.
Should historical results be overwritten?
No. Preserve raw results and add harmonised values as a reproducible derived layer.
When should epochs remain separate?
When overlap is weak, coverage changed, interactions remain unresolved or bridge uncertainty is too large for the intended conclusion.
Continue through verified routes
Core routes
Archaeology subject root · Gold Standard of Education · Gold Standard directory · How X Works hub
Specialist routes
- Authentic-sample panel depletion
- Authentic-sample panel renewal
- Reference-material lot changes
- Reference-material commutability