The Gold Standard · Archaeology
Make the decision path visible
A bridge model is not finished when the historical series is restated. New overlap specimens, reference values, software corrections or matrix classes can reveal that the model behaves differently from the evidence available at release. The new result may matter only near one threshold—or it may change an entire historical claim.
The gold standard monitors the bridge against predeclared decision limits, keeps new evidence independent where possible, and reopens only the claims whose support has materially changed.
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 · SET SURVEILLANCE
1. Monitor the claim, not only the calibration fit
List the conclusions supported by the bridge: classification, chronology, group contrast, trend, provenance assignment or conservation threshold. Link each claim to the values and bridge region it uses.
A small parameter change may be inconsequential for rank order but decisive at a category boundary. Surveillance must therefore test outputs that matter to interpretation.
CHAPTER 2 OF 12 · SET SURVEILLANCE
2. Predeclare decision limits and action zones
Set limits for prediction residuals, uncertainty coverage, parameter movement and claim instability. Use warning and action zones rather than one automatic pass/fail line.
Limits should reflect intended use and measurement capability. A control limit describing process behaviour is not automatically an archaeological importance threshold.
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 · SET SURVEILLANCE
3. Choose independent and representative check evidence
Reserve overlap specimens, reference materials or later paired measurements across range, period and matrix. Independence from the original fit makes a check more capable of revealing bias.
Do not let convenient central-range samples dominate. A bridge used at high values or in a rare matrix needs checks in that region.
| Element | Example |
|---|---|
| Claim output | Boundary classification |
| Check evidence | Held-out overlap specimen |
| Warning limit | Investigate signal |
| Action limit | Reopen affected claim |
CHAPTER 4 OF 12 · DETECT CHANGE
4. Keep the surveillance process stable enough to interpret
Record procedure, instrument, software, reference lot, operator and environment for every check. Changes in the monitoring process can mimic a change in the bridge.
When a method change is unavoidable, bridge the surveillance process itself or mark the break. Otherwise the alarm has no clear source.
CHAPTER 5 OF 12 · DETECT CHANGE
5. Separate random alarms from structural evidence
Examine control-chart signals, residual pattern, matrix concentration, direction and persistence. One unusual check may request confirmation; a systematic range-dependent shift suggests model inadequacy.
Use investigation rules decided before the result is known. Repeatedly widening limits after an alarm converts monitoring into reassurance.
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 · DETECT CHANGE
6. Worked example: new overlap evidence near a provenance boundary
Suppose a bridge was fitted from 24 paired specimens and used to classify historical objects at a value of 50. Six later overlap specimens near 48–54 show the restated old method is 1.8 units lower than the current method in that range, while central checks remain stable.
The team does not reject the whole bridge. It freezes the new evidence, confirms the procedures and reruns the classifications with a local-bias component. Three boundary objects become unresolved; the broad chronology is unchanged. Changed-context check: if the same shift also appears in low and high checks, a local correction is no longer adequate and the full series must be reassessed.
| Pattern | Possible cause |
|---|---|
| Single extreme | Error or rare event |
| Persistent shift | Drift or reference change |
| Range pattern | Model form |
| Matrix cluster | Commutability failure |
CHAPTER 7 OF 12 · REOPEN PROPORTIONATELY
7. Open an investigation before reopening a claim
Verify data entry, specimen identity, reference values, software versions and check-method stability. Then test alternative explanations such as matrix mismatch, drift or an unmodelled nonlinear relation.
Document the investigation even when the bridge is retained. A cleared alarm contributes evidence about the monitoring system and future thresholds.
CHAPTER 8 OF 12 · REOPEN PROPORTIONATELY
8. Map new evidence to affected records and claims
Use dependency records to identify which historical values occupy the affected range, matrix or epoch and which analyses consume them. Avoid recalling every result merely because the bridge version changed.
Also look for compound effects. Several individually small revisions can align in one trend or cluster and become consequential together.
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 · REOPEN PROPORTIONATELY
9. Reanalyse with the old and revised models side by side
Preserve shared uncertainty and rerun the actual downstream analysis. Compare classifications, effect sizes, intervals, trend direction and narrative conclusion.
The question is not whether any number changed. It is whether the evidential support for a published claim crossed a meaningful boundary.
| Finding | Response |
|---|---|
| No claim change | Document only |
| Boundary instability | Limited reclassification |
| Series-wide shift | Revised release |
| Unsupported conclusion | Correction or withdrawal |
CHAPTER 10 OF 12 · RELEASE AND LEARN
10. Choose a proportionate reopening outcome
Possible outcomes include no change, clarification, limited reclassification, revised dataset, corrected publication or withdrawal of a claim. Name the rule and authority for each level.
Do not hide a consequential change to protect continuity, and do not treat a minor metadata correction as if every conclusion collapsed.
CHAPTER 11 OF 12 · RELEASE AND LEARN
11. Version the bridge, data and claim together
Release the revised parameters, covariance, effective range, software, affected records and relationship to the prior version. Preserve the earlier state for reconstruction.
Where a new DOI or release is warranted, link both directions. A reader of the old conclusion should be able to find the revision without guessing.
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 · RELEASE AND LEARN
12. Update surveillance from what the alarm taught
Add checks where evidence was sparse, refine limits where false alarms were common and shorten review intervals when instability emerged faster than expected.
The gold standard keeps the bridge open to correction without making every new datum a crisis. Monitoring earns trust by defining when evidence is strong enough to reopen a historical claim.
| Record | Purpose |
|---|---|
| Old and new model | Reconstruct change |
| Affected records | Bound scope |
| Claim comparison | Show consequence |
| Notification | Reach dependent users |
Frequently asked questions
Does every new overlap specimen require a new bridge?
No. It enters the surveillance process; revision depends on confirmed model change and consequence for intended use.
Can a bridge remain valid for some claims?
Yes. A boundary classification can fail while rank order or a broad chronology remains supported.
Why reserve independent check evidence?
Evidence not used to fit the model is better able to reveal overfitting, drift or an unrecognised range effect.
Should earlier conclusions be deleted?
No. Preserve versions and issue a linked clarification, correction, revision or withdrawal according to the impact.
Continue through verified routes
Core routes
Archaeology subject root · Gold Standard of Education · Gold Standard directory · How X Works hub
Specialist routes
- Bridge-model uncertainty budgets and claim-level decision limits
- Bridge-model revision and versioned reanalysis
- Reference-material commutability drift after method updates
- Reference-material lot changes and long-term comparability