What Corund is not, and what closes the gap
Three honest comparisons. The claim on every one of them is the same, and it is deliberately narrow: no CI product enforces red-on-revert, skip-audit, gate-fold, and approval-binding at merge time -- not that nobody has thought of the underlying idea. Where a mechanism exists unenforced, this page says so.
Agent-side discipline vs merge-side enforcement
Open-source verification skills already exist for coding agents (built for
Claude Code, Codex, and other Agent Skills runtimes) that ask an agent to check a test fails
for the intended bug before trusting it. make-no-mistakes (MIT-licensed) is one:
"frozen specs, tamper-detected tests, an independent blind verifier, hard-blocking gates,"
with a named failure state it calls GAMING-DETECTED -- "a check was
weakened/edited to pass. Halts, never silently retried." That is real, and worth doing. But a
check the agent itself can choose to run, or skip, is exactly the kind of check reward hacking
routes around: an agent under pressure to reach green has every incentive to skip the one
skill that would catch it faking green.
Corund's answer keeps both layers. The planned Claude Code plugin runs the same discipline agent-side, as a complement, early and cheap. The GitHub App runs the check merge-side, outside the agent's own control, as the gate. Agent-side discipline catches early; merge-side enforcement is the piece no CI product enforces today.
Source: make-no-mistakes README and LICENSE, quoted verbatim,
fetched 2026-09-05. Full citation: /docs#sources
(/marketing/sources.md).
The three comparisons
vs mutation testing
Synthetic mutants measure how strong a whole suite is, in general. Corund reverts this PR's actual fix and proves this PR's test detects this PR's bug, in seconds -- plus three checks mutation testing doesn't touch.
vs LLM code reviewers
An LLM reviewer gives you an opinion about a diff. Corund gives you a deterministic proof about one specific test. We don't fight for the word "verification" -- we own proof, gate-integrity, and fake-green.
vs nothing
The simplest alternative is trusting the green checkmark the way CI has always meant it. The dossier is the argument for why that trust has already broken, with dates and commit SHAs.
Pricing, compared
We don't publish a number for a competitor's product until it's been checked against their own current pricing page, with a fetch date. Everything below is quoted verbatim from that page; a row with no public price says so instead of guessing.
| mutant (Ruby mutation testing) | $30/dev/mo or $250/dev/yr, commercial · free for open source |
| Arcmutate (Java/Kotlin mutation testing) | UNMEASURED -- no public price list; /pricing returns 404 (fetched 2026-09-05) |
| CodeRabbit | $24/dev/mo annual (Essentials tier) |
| Greptile | $30/seat/mo + credits (Pro tier) |
| Qodo | $30/mo per 2,500-credit pack, pooled (Pro Team tier) |
| Corund | see /pricing |
Every figure above is quoted from the vendor's own pricing page, fetched
2026-09-05, with the full quote and URL in marketing/sources.md. "Diff-scoped
mutest" is retracted from this comparison -- the 2026-09-05 sweep found no distinct commercial
product under that name.