Phase 4 handoff

External validation package

Everything needed to reproduce the internal numerical claims, freeze hashes before data reveal, publish failures and submit an independently reviewed evidence record.

Validation cannot be self-awarded. A clean run demonstrates that the supplied software reproduces on one environment. It does not prove experimental truth, institutional independence, ignition, gain or reactor feasibility.

Validation readiness8 / 8schemas, commands, fixtures, manifests, templates and issue path

Independent validation0 / 3third-party reproduction, external benchmark, institutional sign-off

Current internal phases15/15 · 21/21Phase 2 and scoped Phase 3 acceptance gates

Reproduce

1 · Freeze

Record the archive and source-tree hashes and publish the preregistration before viewing expected outputs.

2 · Execute

Run the complete suite on infrastructure not controlled by HBF Network. Keep raw stdout and failures.

3 · Review

Have a second named human review the residuals, hashes, deviations and declared domain.

python3 external-validation/run_validation.py \
  --institution "YOUR INSTITUTION" \
  --operator "NAMED OPERATOR" \
  --output external-validation/validation-run.json

The full run requires Python ≥3.10, NumPy and Node ≥18; SciPy is optional. Use --quick for the new phase gates and release checks only, or --describe to inspect the command manifest without executing it.

Files

Check a reproduction-run file locally

This browser-only precheck reads no file off your device. It verifies structure and pass/fail state; it cannot verify identity, signatures or independence.

Select validation-run.json

What can change status

A passing reproduction can support CROSS_CODE_BENCHMARKED only after the evidence schema is complete and accepted by two distinct named humans. EXPERIMENTALLY_VALIDATED additionally requires blinded, preregistered physical evidence with frozen model, code and data hashes.