The Gold Standard · Archaeology
Make the decision path visible
A bridge model connects results across an old and a new reference-material lot, method or instrument era. It may begin as a single offset or ratio, then fail when more overlap data reveal curvature, matrix effects, heteroscedasticity or time dependence. The new evidence does not make the earlier work dishonest. It changes what can responsibly be claimed.
The gold standard preserves the original reported series, builds a versioned reanalysis from recoverable inputs, distinguishes corrected values from newly measured values and shows which historical conclusions change under the revised bridge.
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 · BOUND THE REVISION
1. Define the bridge and the claim it supported
Name the old and new eras, measurand, materials, procedures, matrix population, range and equation. A bridge may align control values without proving that every archaeological sample is comparable.
List the decisions that used the bridged series: trend detection, site comparison, provenance grouping or conservation thresholds. Revision matters because those claims may respond differently to the same numerical change.
CHAPTER 2 OF 12 · BOUND THE REVISION
2. State why revision is being considered
Triggers include a larger overlap set, a replacement-lot effect, nonlinearity, residual structure, changed uncertainty or discovery that reference materials were not commutable for part of the matrix population.
Write the diagnostic before choosing the replacement model. A revision prompted by a preferred historical conclusion is vulnerable to hindsight. Preserve the date, evidence and investigator who opened the review.
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 · BOUND THE REVISION
3. Inventory recoverable historical evidence
Locate raw signals, calibration files, blank corrections, preparation records, reference-material certificates, software versions and original reported values. Classify each period by what can actually be reconstructed.
Do not infer missing raw values from the published answer and then call the result reanalysis. Where only final values survive, a bridge may be applied at that level with a wider uncertainty and an explicit limit.
| Element | Record |
|---|---|
| Eras | Old and new procedures |
| Population | Matrices and range |
| Equation | Bridge version |
| Uses | Claims depending on it |
CHAPTER 4 OF 12 · REBUILD THE MODEL
4. Diagnose how the original bridge fails
Plot old-versus-new paired results and residuals across value, matrix, day and time since transition. Test whether a constant offset, proportional factor or linear model leaves structure that matters for the decision.
A more complex model is not automatically better. Sparse overlap can make a curve look persuasive while worsening prediction. Use scientific plausibility, cross-validation and uncertainty, not fit alone.
CHAPTER 5 OF 12 · REBUILD THE MODEL
5. Compare candidate bridge models
Predefine candidates such as offset, ratio, linear errors-in-variables, matrix-specific effects or a hierarchical model. Account for uncertainty on both axes and repeated preparations from the same material.
Choose using predictive performance, residual checks, parameter stability and decision relevance. Report credible alternatives when several models fit nearly equally well.
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 · REBUILD THE MODEL
6. Worked example: an offset becomes a slope
An original bridge added 0.8 units to old-era values. Twelve new overlap materials spanning 5 to 80 units show that the difference grows with concentration. A revised relation maps old values as 0.4 plus 1.035 times the old result.
A historical value of 20 changes from 20.8 under bridge v1 to 21.1 under v2; a value of 70 changes from 70.8 to 72.85. The revision barely affects a low-range classification but crosses a conservation threshold at the high end. Changed-context check: if a ceramic matrix lies outside the overlap population, neither version earns an automatic transfer; label it unbridged or obtain matrix evidence.
| Candidate | Risk |
|---|---|
| Offset | Misses proportional change |
| Ratio | Misses additive bias |
| Linear | Can overfit sparse range |
| Matrix effects | Needs adequate strata |
CHAPTER 7 OF 12 · RESTATE RESPONSIBLY
7. Create a restated series without erasing the original
Keep original reported values immutable. Add restated values as a new field or release with bridge version, reanalysis date and status. Users should be able to recover what was known and reported at each time.
Use clear labels: observed, originally reported, bridge-v1 adjusted and bridge-v2 restated. Avoid presenting restated values with false measurement dates or implying that historical specimens were remeasured.
CHAPTER 8 OF 12 · RESTATE RESPONSIBLY
8. Propagate revision uncertainty
Carry parameter uncertainty, residual variation, original measurement uncertainty and relevant matrix effects into each restated value. Values derived from the same bridge share correlated uncertainty.
A pointwise interval is not enough for a trend built from many restated values. Report how common bridge uncertainty affects slopes, group contrasts and threshold counts.
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 · RESTATE RESPONSIBLY
9. Re-run conclusions and disclose decision changes
Recompute the analyses that materially depended on the series. Record which conclusions remain, weaken, strengthen or reverse. A restatement is incomplete if it updates numbers but leaves obsolete interpretation in place.
Use a change log that links each affected table, figure, dataset and public claim. When the original remains historically important, retain it with a visible supersession notice.
| Field | Meaning |
|---|---|
| Original value | Historical report |
| Restated value | Versioned reanalysis |
| Bridge version | Transformation used |
| Status | Supported, limited or excluded |
CHAPTER 10 OF 12 · GOVERN THE SERIES
10. Publish versioned data and executable lineage
Archive code, model specification, input identifiers, software environment, checksums and output schema. Give each release a persistent version and a machine-readable relationship to the prior release.
Executable code is useful only when inputs and assumptions are also preserved. Include a human-readable method note explaining why the revision occurred and where it should not be used.
CHAPTER 11 OF 12 · GOVERN THE SERIES
11. Validate the revised bridge prospectively
Hold out overlap materials or collect new paired samples after model selection. Monitor residuals by matrix, range and time. A bridge that fits the development set but fails new material should remain provisional.
Set triggers for another review: new lot, method update, range extension, residual trend or arrival of a matrix absent from the original bridge.
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 THE SERIES
12. Know when remeasurement is safer than restatement
A bridge cannot recover information lost through saturation, undocumented preparation, selective retention or a noncommutable reference. High-consequence claims may require remeasurement of archived specimens.
The defensible endpoint may be a partial series: restated where evidence supports it, unchanged with caveats elsewhere and excluded where reconstruction would be speculation. Completeness is not a substitute for traceability.
| Check | Evidence |
|---|---|
| Lineage | Inputs and code archived |
| Uncertainty | Shared components carried |
| Impact | Claims rerun |
| Prospective test | New paired data |
Frequently asked questions
Does restating a series mean the original was fraudulent?
No. A model can be reasonable with earlier evidence and later need revision. Preserve the original and explain the evidence that changed.
Should every historical value be restated?
Only where the bridge applies and inputs are recoverable. Unsupported periods or matrices should remain clearly limited.
Can the best-fitting model be selected automatically?
No. Fit must be weighed with plausibility, uncertainty, prediction and the decisions the series supports.
Must every published conclusion be rewritten?
Reassess all material conclusions, then issue corrections or supersession notices where the revision changes interpretation.
Continue through verified routes
Core routes
Archaeology subject root · Gold Standard of Education · Gold Standard directory · How X Works hub
Specialist routes
- Replacement-panel lot effects and mixed-epoch analysis
- Authentic-sample panel depletion and preserving comparability
- Authentic-sample panel renewal and archived specimen reuse
- Reference-material lot changes and bridging studies