Real anonymised engagement · Case study
The Release Gate That Held an Enterprise AI Launch
Captivolt put a release gate in front of a listed IT services company’s enterprise intelligence platform: a golden-question set, repeat-run evaluation and a go/no-go readout before launch.
The gate found unstable evidence binding and ranking that ignored question intent. Launch was held for root-cause fixes before any client saw them.
What the work produced
- Release gate
- Golden-question set
- Defect register
- Go/no-go readout
CLIENT CONTEXT
A listed IT services company was preparing to launch an enterprise intelligence platform built on agents, RAG and a knowledge graph.
BUSINESS PROBLEM
The platform performed well in demonstrations. No one had tested whether it gave the same evidence-backed answer every time it was asked.
CONSTRAINTS
- A launch planned for a platform that performed well in demonstrations
- Answers drawn from agents, RAG and a knowledge graph together
- No existing test of whether an answer repeated with the same evidence
- Clients who would meet the platform from its first release
ARCHITECTURE & APPROACH
A release gate between the platform and its launch. Real questions from the business formed a golden set; each was run repeatedly through the platform and checked for whether its evidence stayed bound to the same sources and whether its ranking followed what the question asked.
WHAT CAPTIVOLT DELIVERED
Golden-question set · repeat-run evaluation harness · evidence-binding and ranking checks · ranked defect register · go/no-go readout.
EVIDENCE
The gate’s artefacts are available for an anonymised discussion under NDA where client permission allows.
The public case study describes the system and the work. It does not publish the client’s identity, source data, commercial information or unapproved performance figures.
OUTCOME
The defects were found by the gate rather than by a client: launch waited for root-cause fixes to evidence binding and ranking.
WHAT THE CLIENT OWNS NOW
- The golden-question set
- The repeat-run evaluation harness
- The ranked defect register and its root causes
- The go/no-go readout
RELATED SOLUTION
Explore the capabilities behind the engagement.
AI Release Readiness Review · VeriCore · AI Quality, Governance & Security
Want to see the artefacts?
Start with the work most relevant to your initiative.
Anonymised artefacts and reference discussions are available under NDA where client permission allows.