How Helix Proves ROI

Helix can report “ROI” in a way that is truthful, auditable, and policy-grounded:

• No claims about biological counterfactuals (“would have failed biologically”).
• Only claims about policy-defined interpretability requirements (“could not have produced interpretable data under policy”).

1) Blocked experiments (hard numbers)

Every helix helixspec-lint bundle now includes an immutable counter artifact:

• helix.metrics.run.v1.json

It records (at minimum) the policy_id, policy_hash, verdict (PASS/FAIL), and the emitted WLP codes.

For ROI reporting:

• Blocked experiments = count of verdict == "FAIL" within the selected time window and policy.

This is mechanical, reproducible, and defensible.

2) Structural non-interpretability (policy grounded)

Helix does not claim “these experiments would have failed.”

Helix claims:

• “These experiments violated policy requirements that are necessary for interpretability.”

Each WLP code maps to a stable failure mode and a declared interpretability impact (see the Failure Mode Registry in code).

For ROI reporting:

• Structurally non interpretable = blocked experiments with ≥1 hard error whose interpretability impact is impossible.

3) Estimated avoided cost and time (declared ranges)

Cost and duration are estimates, and must be declared by policy (not predicted by Helix).

Policy packs may include:

• roi.typical_cost_usd.min / roi.typical_cost_usd.max
• roi.typical_duration_days.min / roi.typical_duration_days.max

For ROI reporting:

• Estimated avoided cost = sum of the per-policy cost ranges for blocked experiments
• Estimated time saved = sum of the per-policy duration ranges for blocked experiments

Helix always reports ranges (never a single point estimate).

Outputs you can use in reports

For a given window and policy, Helix can generate:

• helix.metrics.summary.v1.json

This is intended to feed pilot reports, renewal justification, and procurement-friendly summaries.

To generate a summary from a directory of lint bundles:

helix metrics summarize --in <DIR> --out helix.metrics.summary.v1.json --policy-id <POLICY_ID> --start YYYY-MM-DD --end YYYY-MM-DD

Sales-safe phrasing (recommended)

Avoid:

• “We prevented X failed experiments”

Use:

• “We prevented X experiments that violated your policy’s requirements for interpretability.”
• “Based on your declared experiment cost ranges, this avoided approximately $Y–$Z in spend.”