Trust
Reliability & Architecture
Verdict is engineered as a robust, deterministic growth intelligence engine rather than a brittle prompt wrapper.
Reliability Guarantees
Strict Bounded Execution
Every audit executes within deterministic character and page gathering budgets. This ensures bounded investigation times (typically one to a few minutes) and prevents runaway scraping loops.
Structured Schema Validation
All LLM outputs are validated against strict JSON schemas. If a model output deviates from the expected schema, it is retried and validated before persisting, preventing corrupt records.
Deterministic Scoring Authority
The overall Growth Readiness Score is calculated via deterministic weighted aggregation applied to structured dimension evaluations.
Fail-Closed Relevance Gating
If a supporting page's relevance to the audited company cannot be verified due to network errors or ambiguity, Verdict fails closed and discards the page from scoring.
Model Failover & Timeout Discipline
The investigation pipeline enforces rigid timeout budgets across all HTTP extraction and model execution tasks. If an individual stage encounters transient model latency, the system degrades gracefully, maintaining audit progress without stalling the consumer.