SALT19 invites independent researchers, engineers and evaluators to reproduce, challenge or falsify published results. Successful replication is useful. A well-documented failure to replicate is also useful. The goal is to separate what the artifacts actually support from what still depends on SALT19's own environment or interpretation.
These targets are selected because they already have public records or machine-readable evidence surfaces. Replication scope should reference a specific version, artifact and claim.
Evaluate architecture claims, benchmark procedures, persistence/memory behavior, planning, governed execution and verified work-product paths using the stated claim boundaries.
Software record DOI ↗Test whether the disclosed transition and verification mechanisms behave as described under reproduced inputs, perturbations and failure cases.
VST DOI ↗Reproduce the conditions under which experience is retained, recalled, accepted or rejected without treating an internal pass as external AGI evidence.
Experience compounding DOI ↗Independently probe discovery, MCP/A2A surfaces, x402 challenge behavior, operation identity, idempotency, receipts and bounded tool outputs. Commercial demand is a separate question.
Open live Grid surface →Record DOI/version, repository revision or live endpoint, test date, environment and the precise claim being tested.
SALT19 may clarify installation or protocol ambiguity, but the independent evaluator should control the test execution, observations and conclusions.
Record failures, deviations, missing dependencies and ambiguous results rather than discarding them because they do not support the original claim.
Where practical, retain scripts, fixtures, logs, environment information, hashes and output artifacts sufficient for another evaluator to inspect the result.
Disclose funding, collaboration, SALT19 assistance and any use of SALT19-controlled infrastructure so readers can judge the strength of independence.
SALT19 will treat credible negative findings as engineering evidence, not as a reason to hide the test.