artifacts/standard-named

Rendered from markdown source. Open raw source on GitHub.

Withdrawal, Unlearning, Derivative Propagation, and Benefit Matrix

Status: candidate pre-production Content canon status: unset

| Action or claim | What it can establish | What it cannot establish automatically | |---|---|---| | Stop new collection | Future acquisition ends | Stored copies are deleted | | Delete source copy | Known source storage is removed | Model influence is gone | | Block future training | Source will not enter future declared runs | Existing models forgot | | Block runtime retrieval | Retrieval system stops serving the source | Base weights changed | | Restrict release | Named model or checkpoint cannot be released | External copies complied | | Approximate unlearning | Declared behavior changed under tests | Complete forgetting | | Verification | Declared tests passed | All possible influence is absent | | Retrain without source | New model excludes declared source set | Every ancestor restriction was resolved | | Retire model | Future declared use stops | Repair and residual duties disappear | | Update derivative profile | Known successor receives restriction | Unknown successors are reached | | Benefit payment | Concrete economic return occurred | Unlimited future consent | | Community license | Collective governance and use rights exist | Every individual concern is resolved | | Public access | Wider benefit is available | Source extraction is repaired | | Repair fund | Resources exist for correction and harm | No further accountability is needed |

Withdrawal ladder

1. stop new collection
2. delete source copy
3. block future training
4. block runtime retrieval
5. restrict release or use
6. apply approximate unlearning
7. verify against declared tests
8. retrain without source
9. retire or replace model

Derivative propagation minimum

source change
origin record
affected datasets
affected models
affected deployments
propagation rules
performed changes
unreachable descendants
verification
residual risk

Benefit test

A benefit mechanism should identify:

contributors or communities
contribution class
concrete mechanism
governance
distribution rule
nonfinancial benefit
limitations
review

Governing principles

Withdrawal is a governance right. Complete model unlearning is a technical capability that must not be promised beyond evidence.

A vague promise that innovation benefits everyone is not a benefit-sharing mechanism.