# A remediation claim is worth what its method shows, so read the denominator, the clock and the customer first

Guide, 8 Oct 2026, read 10 Nov 2026

**The short answer.** Averlon states 110,000 critical and high issues became 260 pull requests and mean time to remediate fell from 45 days to 6, for an unnamed customer. Backline states 5 weeks to 2 hours. Both are vendor claims read 10 Nov 2026. Check the denominator, the clock and the customer before using either.

## Ask three questions of every number

Remediation claims come as percentages, time reductions and ratios. Each hides three choices: what was counted, when the clock started and stopped, and who the customer was. A claim that states all three can be tested. A claim that states none is marketing copy, and the register labels it that way.

The denominator is the first thing to find. A 95% reduction of what: raw scanner output, critical findings, findings on internet-facing assets? The clock is second. Mean time to remediate can start at detection, at disclosure or at ticket creation, and end at deploy, at merge or at verified closure. The customer is third: size, stack and whether it is named.

## Four claims read for the register

The claims below are quoted as the vendors state them, and the third column notes what the page gives. All were read on 10 Nov 2026 and all are the vendor's claim unless marked.

| Vendor and claim | Customer | What the page gives |
| --- | --- | --- |
| Averlon: 110,000 critical and high issues, 3,400 tickets, 260 pull requests; mean time to remediate 45 days to 6 | Unnamed | Counts and a before and after; the register records no clock definition |
| Backline: average time to fix 5 weeks to 2 hours | Unnamed data-analytics company | A case study; the register records no clock definition |
| Cogent Security: 57.9% of 190 LiveCVEBench tasks, against 45.3% for two coding tools (third-party report) | Not a customer measure | Method notes: outbound network blocked, grading tests hidden, inference budgets differed |
| Seemplicity: unknown asset ownership 20% to 2% in six months | Hertz (named) | A named customer case study; the measure is ownership, not remediation time |

## Read the shape of the claim

- Named customers can be called. Unnamed ones cannot, and the register records that fact.
- A benchmark on a public task set can be repeated. A figure from a customer's private data cannot.
- An average over all customers includes the easy ones. Ask for the median and the range.
- A claim of time saved is not a claim of risk removed. Ask what happened to the findings that were not fixed.
- A homepage figure with no method on the page, such as a stated 90% faster remediation, is a headline. Ask for the method in writing.

## Turn the claim into a trial test

Every claim implies a test. If a vendor says mean time to remediate fell from 45 days to 6, ask for the definition, then measure the same quantity on a sample of your own findings before and during a trial. If it says a share of findings are not exploitable, take 20 of them and check the evidence shown. The guide on trial questions lists the tests by topic.

Where a page gives no method, the register says so. It does not fill the gap with an estimate.

## Know what a third-party label means

The register labels a fact third-party source when an outside party reported it. That label means the outside party published the statement, not that it measured the result. SiliconANGLE reports that Cogent's VR-1.1 passed 57.9% of 190 LiveCVEBench tasks, and notes the test was run by the vendor. SecurityWeek confirmed Astelia's $35 million seed and Series A, which is a funding fact, not a performance one.

Read the label with the content. A funding fact confirmed by an outside outlet is well supported. A performance figure reported by an outlet from a vendor's own test is the vendor's claim, carried by a second page.

## Keep a claims sheet

During evaluation, keep a sheet with one row per claim: the vendor, the figure, the page and date, the three answers (denominator, clock, customer), and the test you will run. Rows with no answers go to the bottom and stay out of the decision.

The sheet also gives you a record for procurement. When a contract names a figure, such as a reduction in findings, the row shows where it came from and what was agreed about its definition. Figures that appear only on a homepage should not appear in a contract without a written definition.

## Related pages

- [Guides](/guides/)
- [The remediation SLA that holds](/guides/the-remediation-sla-that-holds/)
- [Ownership mapping: finding who can fix a finding](/guides/ownership-mapping/)
- [Mean time to remediate, measured honestly](/guides/mean-time-to-remediate-measured-honestly/)

## Sources

- Averlon blog on automating remediation: https://www.averlon.ai/blog/automating-remediation-application-infrastructure-security-issues (read 10 Nov 2026)
- Backline case study: https://backline.ai/case-studies/five-weeks-to-under-a-day/ (read 10 Nov 2026)
- SiliconANGLE on Cogent VR-1.1, 3 Nov 2026: https://siliconangle.com/2026/11/03/exclusive-cogent-securitys-vr-1-1-model-outscores-claude-code-and-codex-on-vulnerability-repair/ (read 10 Nov 2026)
- Seemplicity, Hertz case study: https://seemplicity.ai/resource/hertz-case-study/ (read 10 Nov 2026)

Last reviewed 10 Nov 2026. Canonical: https://automatedvulnerabilityremediation.com/guides/reading-a-vendors-remediation-claim/
