The Gold Standard · Experimental Design
Make the decision path visible
A baseline can return because someone needed to reproduce an older analysis, test a disputed preference weight or recover work during an outage. Those are legitimate purposes, but an unmanaged fork can inherit current credentials, data and publication paths. Deleting the discovered worker leaves the same need and the same conditions ready to create another copy.
The gold standard governs the fork as a lifecycle. It registers purpose, isolates historical work, reconciles current outputs and retires obsolete templates from every authorised build and recovery path while preserving a transparent route for genuine replication.
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 · MAP THE FORK LANDSCAPE
1. Define a fork as governed analytical lineage
Treat copied repositories, images, notebooks, manifests and scheduled environments as forks when they can reproduce or alter the baseline. Record parent, digest, owner and purpose.
A fork is not made safe by being unofficial; lack of registration is itself part of the risk.
CHAPTER 2 OF 12 · MAP THE FORK LANDSCAPE
2. Inventory effective access, not just files
Map which forks can reach current data, identities, compute, exports and decision gates. Include personal namespaces, caches and emergency recovery sources.
The authority question depends on what the fork can do, not where its files are stored.
Evidence check
Check whether the scope, owner and decision consequence are explicit before continuing.
CHAPTER 3 OF 12 · MAP THE FORK LANDSCAPE
3. Classify purpose and required lifespan
Separate current analysis, validated development, historical replication, evidence preservation and one-time investigation. Give each a boundary, expiry and accountable owner.
One environment should not quietly serve every purpose.
| Field | Purpose |
|---|---|
| Parent digest | Lineage |
| Owner | Accountability |
| Use | Boundary |
| Expiry | Review point |
CHAPTER 4 OF 12 · SEPARATE VALID USES
4. Design historical isolation
Use frozen data, separate identity, restricted network, non-current outputs and logged approval for any export. Preserve packages and documentation needed for reproducibility.
Isolation should protect current decisions without destroying the ability to understand older work.
CHAPTER 5 OF 12 · SEPARATE VALID USES
5. Define how experimental changes re-enter current work
Require a registered proposal, explicit estimand and preference-weight change, validation evidence, review and a new baseline decision. Do not merge from an historical fork by convenience.
Re-entry is a scientific and authority decision, not merely a code operation.
Evidence check
Ask a second reviewer to reconstruct the example from the preserved evidence and stated assumptions.
CHAPTER 6 OF 12 · SEPARATE VALID USES
6. Worked example: a fork tests a disputed preference weight
A statistician copies an obsolete template to investigate whether one stakeholder weight drove a sensitivity conclusion. The fork begins on frozen data, but its inherited token can read the current repository and its exporter can reach the publication queue.
Governance revokes the inherited token, recreates the study in an isolated historical environment and records the comparison. Any proposed weight change enters a fresh registered analysis. Changed-context check: if the fork already uses frozen data, a separate identity and a non-current export store, preserve it under the historical register with a review date rather than deleting useful replication evidence.
| Purpose | Required boundary |
|---|---|
| Current | Approved baseline |
| Development | Test data and review |
| Historical | Frozen data and identity |
| Evidence | Read-only preservation |
CHAPTER 7 OF 12 · RETIRE OBSOLETE TEMPLATES
7. Retire obsolete sources and descendants
Remove old baselines from parent images, templates, build caches, bootstrap scripts, schedulers and emergency bundles. Register or destroy local copies according to purpose and retention.
A deleted repository can be rebuilt from an old image or package cache unless the whole lineage is addressed.
CHAPTER 8 OF 12 · RETIRE OBSOLETE TEMPLATES
8. Enforce the approved baseline at admission
Require signed digests, registered lineage, least-privilege identities and decision-version receipts before current data or gates are available. Reject unknown forks by default.
A register without enforcement records failure after it happens.
Evidence check
Verify that the control still reaches the intended decision under a changed but plausible context.
CHAPTER 9 OF 12 · RETIRE OBSOLETE TEMPLATES
9. Reconcile current outputs from recurrent forks
Trace analyses, exports and decisions created during exposure. Compare analytical definitions and recompute material current uses on the approved baseline.
A governance breach does not automatically negate every result; consequence follows the changed design and how the output was used.
| Source | Action |
|---|---|
| Parent image | Replace digest |
| Build cache | Purge obsolete layer |
| Scheduler | Remove manifest |
| Emergency bundle | Exercise new version |
CHAPTER 10 OF 12 · PROVE PERMANENCE
10. Exercise permanent retirement
Restore backups, build clean workers, use emergency runbooks, rotate identities and attempt deployment from obsolete digests. Verify that historical environments remain isolated and current gates reject them.
Retirement proof must test the pathways that once recreated the fork.
CHAPTER 11 OF 12 · PROVE PERMANENCE
11. State limits and misconceptions
No central register can guarantee discovery of every personal copy. Signed code also cannot establish that the experimental design is scientifically valid. Preservation duties may require some obsolete environments to remain.
Record residual repositories, compensating controls and triggers for renewed reconciliation.
Evidence check
Confirm that versions, notices and unresolved limits remain visible in the final record.
CHAPTER 12 OF 12 · PROVE PERMANENCE
12. Monitor lineage, purpose and authority
Track active digests, repository access, build provenance, identity grants, decision receipts, export destinations and expired historical registrations.
The gold standard retires obsolete templates permanently from current authority while keeping a disciplined, intelligible route for historical replication and future scientific change.
| Test | Expected result |
|---|---|
| Old digest | Current gate rejects |
| Historical run | No current access |
| Clean rebuild | Approved receipt |
| One-time export | Named approval and log |
Frequently asked questions
Must every obsolete fork be destroyed?
No. A historically necessary fork may remain when it is registered, isolated and unable to confer current authority.
Can useful code move from an historical fork to current work?
Yes, through a fresh registered proposal, review and baseline decision rather than an informal merge.
Why test emergency recovery?
Emergency bundles and caches are common paths by which retired templates regain identities and current access.
What makes retirement permanent?
Obsolete sources and descendants are removed from authorised builds, admission rejects unknown lineage, affected outputs are reconciled and realistic rebuild tests pass.
Continue through verified routes
Core routes
Experimental Design subject root · Gold Standard of Education · Gold Standard directory · How X Works hub
Specialist routes
- Final enforcement closure surveillance, unauthorised baseline recurrence and reconciliation reopening
- Reconciliation reopening governance, template baseline appeals and final enforcement closure
- Enforcement-repair regression, recovery-template drift and reconciliation reopening
- Final decision propagation failures, stale interim permissions and authority reconciliation