Helen detector

Chain Contradiction detector

The Chain Contradiction detector finds semantic drift inside an existing product chain. It asks whether a descendant’s intent opposes the parent or ancestor that gives it purpose.

Method
Bounded semantic judgment over graph lineage
Implementation policy
Blocking

What it catches

Helen considers pairs that share a direct ancestor-descendant line and fall within a bounded semantic candidate range. An AI judge then classifies whether their intents contradict.

The detector emits a finding only for a contradiction above its confidence floor. Related, unrelated, or merely repetitive statements do not become Chain Contradiction findings.

Evidence Helen records

The finding preserves both artifact identities, their types and titles, the lineage route, semantic similarity, judge verdict, confidence, prompt version, and a plain-language rationale.

Confidence controls severity, but the cited rationale remains available for human review. Semantic judgment is never presented as deterministic proof.

How to reconcile it

Revise the descendant when execution drifted. Revise the ancestor when strategy changed. Dismiss the finding only when the two statements can coexist, and record the reason so detector precision can be reviewed.

Detector boundary

It abstains below its confidence floor and routes near-duplicate text to a separate redundancy check.

Research basis

Kaplan, R. S., & Norton, D. P. (1996). Using the balanced scorecard as a strategic management system. Harvard Business Review, 74(1), 75–85.

Related reading

Give every agent the current product picture.

Join the Helen’s Foundry early-access program.

Join the waitlist