Artifact Scope

Reproduction starts with scope.

Public artifacts verify public claims, provenance, boundaries, and reproducible public behavior. They are not requests for customer data, live finance execution, private agent orchestration, credentials, commercial schedulers, or anonymous conference packages.

Evidence has a shape visual.

Scope Rule

A public artifact can support only the claim boundary it actually exposes.

The public layer is designed for bounded inspection: claim, formula or protocol, public evidence, explicit limitations, and an attack route. If a result requires private logs or protected systems to be understood, the public claim must be narrowed rather than inflated.

Artifact Classes

What each public object can and cannot prove.

ArtifactCan checkCannot prove by itselfRepair route
Paper PDF / Zenodo recordTimestamped claim, method, authorship, and boundaryDeployment readiness, universal truth, private product capabilityBoundary update or versioned preprint
Dataset / benchmark artifactTasks, scoring, raw rows, CI, negative resultsGeneral intelligence, real deployment, untested transferScorer patch, new split, stronger baseline
Code / public demoProtocol interface, threshold behavior, toy executionProduction safety, customer workflow, live financial performanceFailing test or counterexample packet
Registry rowPointer from claim to evidence, boundary, and challenge classEvidence by itselfRegistry correction
Public noteRoadmap, explanation, and public-facing synthesisDecisive evidence without linked artifactLink to DOI, code, data, or manifest
R0

Inspect

Read the claim, scope, evidence pointer, and protected boundary.

R1

Verify Package

Check DOI, file names, hashes, manifest rows, and versioned public links.

R2

Run Procedure

Run public scripts, demos, or scorer commands without private credentials.

R3

Reproduce Table

Regenerate a public table, interval, or no-go decision from public artifacts.

R4

Independent Replication

Use an independent dataset, baseline, simulator, or external lab procedure.

Challenge Packet

A useful attack must be reproducible.

A public counterexample should include:

  • Claim id or public URL.
  • Artifact URL, command, screenshot, input, or table row.
  • Expected behavior versus observed behavior.
  • Failure class: formula counterexample, leakage, stronger baseline, reproduction failure, boundary too broad, credit leak, or authority leak.
  • Proposed repair: narrow the claim, patch the artifact, add a negative result, or update the registry.

Flagship Scope Notes

Three public cards, three bounded claims.

ProgramPublic scopeBoundaryBest attack
WisdomBenchTask templates, scorers, raw scores, CI, negative resultsNot human-like wisdom or universal deployment proofScorer bug, leakage, stronger baseline
Proof-Carrying ActionSchema, demo, no-go, no-credit repair, receipt disciplineNot live trading or product-grade deployment proofFalse pass, missing receipt, false no-go
Relational ObservabilitySystems protocol, relation debt, evidence half-life, claim mapNot a universal law without domain validationRelation deletion or stale evidence case

Protected Boundary

Public verification does not require private exposure.

Do not request API keys, credentials, customer data, raw private logs, live financial execution records, private agent orchestration, commercial schedulers, or anonymous conference packages as a condition for evaluating public claims. A public claim that depends on those materials should be moved to a weaker public tier or marked private-only.