Read the evidence before you book the call.

This page shows what an INNOVIZZ engagement actually delivers: an evidence packet from a TLF QC run, end to end, on synthetic data. A candidate adverse-event summary goes in for review; a discrepancy report, an execution score, and a hashed, tool-logged evidence record come out. A reviewer accepts or rejects every finding.

One table, submitted for QC, with defects planted in it

A candidate adverse event summary is submitted for QC with its inputs: synthetic ADAE and ADSL datasets, the table shell, and the SAS program that produced it.

The system interprets the shell, re-derives the expected output deterministically from source, and compares expected against candidate at schema, row and cell level. It finds the defects we seeded.

Every finding carries the expected value, the candidate value, the variance, a likely root cause and a suggested fix. A human reviewer disposes of each one.

This is the artifact. Not a dashboard, a record.

Every run exports a structured evidence record: the hash of every artifact it read, every tool event it executed, the score it earned, and the human review state of every finding. It is built to be read by QA, not by us.

The L0 to L5 execution score

Every run earns a score against a fixed execution model: how much of the workflow executed under control, with evidence, before a human took over. The score and its basis travel inside the packet, so two runs can be compared and a corridor’s progress can be audited over time. The scoring model is published with the demonstration, not summarized here.