Regulated teams that gate every agent release.
Four teams in healthcare, banking, legal research and insurance, measured at Day 60 and Day 120 against their own baseline.
- Day 0Baseline capturedQA hours, regression incidents, CI catch rate
- Day 14Fast-path gate live20 critical cases on every push
- Day 60First reviewCustomer reports against baseline
- Day 120Case study reviewCustomer-reported results, legal sign-off
- fabricated-finding regressions reaching production between Day 60 and Day 120
- 0
- fabricated-finding regressions reaching production between Day 60 and Day 120
- Customer-reported: Series B clinical documentation company
- to produce the adverse action reason-fidelity evidence pack for model risk review, previously assembled by hand
- 1 export
- to produce the adverse action reason-fidelity evidence pack for model risk review, previously assembled by hand
- Customer-reported: US regional bank, small-business underwriting agent
- when the gate blocked a release with a citation-verification regression; the CTO says the subscription paid for itself there
- Week 2
- when the gate blocked a release with a citation-verification regression; the CTO says the subscription paid for itself there
- Customer-reported: Legal research copilot company
- prompt regressions reaching production between Day 60 and Day 120, from 5 in the prior two quarters
- 0
- prompt regressions reaching production between Day 60 and Day 120, from 5 in the prior two quarters
- Customer-reported: Top-20 US commercial insurer
Filter the stories
By vertical, by the kind of result that mattered, and by who said it.
- Healthcare & ClinicalSeries B clinical documentation companyJudge agreement
How a clinical documentation team gave its safety board the evidence to approve an ambient scribe agent
A Series B company selling ambient documentation to multi-specialty groups needed its clinical safety board to sign off on a new agent that drafts SOAP notes and reconciles medication lists. Golden datasets and trajectory assertions replaced a two-day manual review before every release.
0fabricated-finding regressions reaching production between Day 60 and Day 120Customer-reported - FinTech & BankingUS regional bank, small-business underwriting agentAudit evidence
How a regional bank's model risk team turned CI runs into adverse action evidence
A US regional bank uses an agent to assemble small-business credit memos and draft adverse action reasons. Its model risk and compliance teams now review evidence packs exported from the same CI runs that gate each change.
1 exportto produce the adverse action reason-fidelity evidence pack for model risk review, previously assembled by handCustomer-reported - LegalTech & ContractsLegal research copilot companyRegressions
How a legal research copilot stopped a broken release in week two and wired the gate in an afternoon
A venture-backed company sells a research copilot to litigation teams. Citation verification suites and a fast-path GitHub Actions gate caught a release that would have shipped unverifiable citations.
Week 2when the gate blocked a release with a citation-verification regression; the CTO says the subscription paid for itself thereCustomer-reported - Insurance & ClaimsTop-20 US commercial insurerQA time
How a top-20 US commercial insurer took prompt regressions in claims triage to zero
The claims AI team at a top-20 US commercial insurer runs an agent that triages first notice of loss, interprets coverage and routes claims. Trajectory gates cut manual prompt QA by 80% and kept production regressions at zero through Day 120.
0prompt regressions reaching production between Day 60 and Day 120, from 5 in the prior two quartersCustomer-reported
“We took prompt regressions in production to zero and cut manual QA by 80%. More important, the adjusters stopped re-reading outcomes and started reviewing the failure clusters that actually needed them.”
How we keep these numbers honest.
No modelled savings, no “up to” figures. Each metric is one the customer measured before and after, and approved for publication.
- 01
Baseline before the first suite
We record manual QA hours, regression incidents in the last two quarters and the CI catch rate before any GONATURELLE run, so improvements have a real denominator.
- 02
Same definitions at every review
A production regression means the same thing on Day 0, Day 60 and Day 120: a behaviour change that reached users and was later traced to a prompt, model or tool change.
- 03
Customer-reported, customer-approved
The customer reports each figure and approves the wording. Where a result depended on their own reviewers, as with judge agreement, the story says so.
- 04
Written sign-off before publication
Descriptors like "Top-20 US commercial insurer" are approved in writing. We do not place logos without legal sign-off.
Frequently asked questions
Why are customers anonymous?
Most of our customers sell into hospitals, banks and carriers under enterprise NDAs that prohibit naming vendors in their stack. We use descriptors the customer has approved in writing, such as "Top-20 US commercial insurer". Reference calls with the named teams are available during an Enterprise evaluation.
How are the metrics measured?
We capture a baseline before onboarding: engineering hours on manual QA, production regression incidents over the previous two quarters, and how many regressions CI caught. The customer then reports the same measures at Day 60 and Day 120. Every number on these pages is customer-reported, not an GONATURELLE estimate.
Do these results mean GONATURELLE made the customer compliant?
No. GONATURELLE supplies test evidence. Customers make their own validation decisions, regulatory filings and clinical or credit judgements. Several stories describe how that evidence is used by a safety board or model risk committee, which is the customer's process, not ours.
Can we talk to one of these teams?
Yes, on an Enterprise evaluation. Book a technical demo and tell us which story is closest to your use case.
Measure your own Day 0 baseline.
Start a 14-day trial, run a vertical pack against your agent, and see what your current QA process misses. 5,000 runs, no card.