Start with a fixture you can reproduce.

01 / Choose an explicit experiment

The fresh Earth testnet and regional ground candidates have distinct identities, rules and evidence. The former uses explicit peer bridges. Regional candidates integrate relay and admitted local BFT into their normal fixture lifecycle. A separately started mesh demo tests transport only.

Use the exact guide for the question you want to investigate. The pinned v53 runtime underlies the separate v54 ordinary return cycle and v55 failed finite fault scope. The older v26 guide remains available for segmented history and recovery research.

02 / Verify the named source package

Compare the complete source-manifest inventory and SHA256SUMS before unpacking source.tar.gz. Pin the publication commit and its stated toolchain. GitHub’s automatic publication-repository archive is not the exact runtime source package. A checksum identifies reviewed bytes, not authorized ledger state.

03 / Use fresh private fixture directories

Follow the chosen guide’s actual startup commands, keys and literal neighbor configuration. Keep transport pins, native journals, voting custody, owner custody and separate caller heads in their intended directories. Incompatible candidates require fresh genesis/currency; never convert old custody or migrate value.

Do not learn endpoints, TLS pins or consensus membership from advertisements. The configured neighbor establishes the first contact; farther discovery never grants ledger or signing authority.

04 / Observe the behavior you came to test

Distinguish identity discovery, candidate routes, contact, durable transport receipt, native import and spendability. Inspect the exact declared capacity and verification horizon. Unknown lock-contended observations remain unknown; do not substitute height zero or infer a global balance from one lagging replica.

Useful relay contacts can extend reach and add paths. Node count alone cannot establish independent custody, sustained consensus liveness or physical link capacity.

05 / Stop and retain the evidence

Follow the package’s stop and cold-verification procedure. Preserve failed reports, included owner requests, unresolved exports and interrupted private targets. A stopped authentication audit describes the retained state; it cannot convert a failed live fault campaign into a pass.

Publish only sanitized findings and exact public source bindings. Keys, wallet/caller/signer/replica/node state, TLS material and generated recovery images stay private. Same-host process interruption is not power-loss or independent operation.