Omnis HelixGenome IDE
← Docs
Helix CLI docs
Browse Helix CLI docs

Genome Editing Evidence Runtime v0.1

Claim

Helix can build a deterministic, replayable evidence bundle for a synthetic genome-editing scenario. The runtime produces canonical software artifacts, provenance hashes, policy receipts, an offline verifier, and a reviewer summary without producing real-world execution instructions.

Scope

This is not a new design UI and not a broader simulator. It is a narrow evidence runtime for software-only, in-silico review of a proposed edit scenario.

The v0.1 demo uses a synthetic Wilson ATP7B scenario because it is easy to explain and review. The sequence inputs are synthetic demonstration data.

Inputs

The runtime consumes one JSON spec with:

  • run_id
  • safety_scope = in_silico_only
  • reference_sequence
  • mutation
  • proposed_edit
  • editing_modality
  • simulation_parameters

The policy gate fails closed when required fields are missing, when safety_scope is not in_silico_only, or when real-world execution fields such as sample_prep, delivery_method, or dosage appear in the spec.

Runtime

The runtime performs:

  • canonical JSON serialization
  • deterministic edit artifact generation
  • deterministic toy off-target screening over declared synthetic candidates
  • confidence scoring for review display
  • policy receipt generation
  • provenance hashing
  • environment fingerprinting
  • evidence contract generation
  • deterministic zip assembly

The off-target model is explicitly labeled deterministic_toy_off_target_v1. It is a software proof model, not a biological validity claim.

Outputs

PYTHONPATH=src python -m helix.evidence_runtime --out <dir> --verify writes from a source checkout. Installed environments can omit PYTHONPATH=src.

  • evidence_bundle.zip
  • manifest.json
  • evidence_contract.json
  • input_spec.json
  • edit_artifact.json
  • off_target_report.json
  • policy_receipt.json
  • provenance.json
  • reproducibility_receipt.json
  • verify.py
  • reviewer_summary.md

evidence_contract.json records the required artifact set, canonicalization rules, replay semantics, policy fail-closed conditions, verifier targets, and non-claims. manifest.json records every artifact path, byte size, SHA-256 digest, bundle digest, and manifest digest. verify.py performs offline verification against either the unpacked bundle directory or evidence_bundle.zip and fails closed when any recorded artifact is missing, changed, or inconsistent with the evidence contract.

Demo Path

Build the synthetic demo bundle:

PYTHONPATH=src python -m helix.evidence_runtime --out /tmp/helix-evidence-demo --verify

Verify offline:

cd /tmp/helix-evidence-demo
python verify.py .

Verify the archive directly:

python verify.py evidence_bundle.zip

Replay determinism:

PYTHONPATH=src python -m helix.evidence_runtime --out /tmp/helix-evidence-demo-a --verify
PYTHONPATH=src python -m helix.evidence_runtime --out /tmp/helix-evidence-demo-b --verify
sha256sum /tmp/helix-evidence-demo-a/evidence_bundle.zip
sha256sum /tmp/helix-evidence-demo-b/evidence_bundle.zip

Tamper proof:

python - <<'PY'
from pathlib import Path
p = Path('/tmp/helix-evidence-demo/off_target_report.json')
p.write_text(p.read_text() + '\n', encoding='utf-8')
PY
python /tmp/helix-evidence-demo/verify.py /tmp/helix-evidence-demo

The final command must fail.

Non-Claims

  • Helix v0.1 evidence runtime does not optimize edit design.
  • It does not provide real-world execution guidance.
  • It does not replace expert review.
  • It does not claim biological validity for the toy off-target model.
  • It proves deterministic evidence packaging, replay, and fail-closed verification for a constrained synthetic scenario.

Validation

Targeted regression coverage:

python -m pytest tests/test_genome_editing_evidence_runtime.py

Covered behavior:

  • identical replay hashes for repeated bundle builds
  • offline verifier passes on the original bundle
  • offline verifier fails after tampering
  • evidence contract records replay, policy, verifier, and non-claim semantics
  • verifier fails closed when the evidence contract is inconsistent
  • policy gate blocks real-world execution fields
  • policy gate requires in_silico_only