Our CI fails the build if the marketing lies.
Every page on this site passes machine-checked honesty gates before it can deploy. This page publishes the gates, the proof ladder behind every number, and the builds those gates have already failed.
Why publish this
AI companies ship pages full of animated counters, invented dashboards, and testimonials nobody can trace. Buyers have learned the smell. We went the other way: we wrote our marketing rules as code, wired them into our build pipeline, and made the build fail whenever a page breaks them. This page is the firewall itself, published.
The proof ladder behind every number
Every claim about work or money on a RelayLaunch surface is classified on an 8-rung ladder, and copy may never claim a higher rung than the data underneath it. Climbing a rung requires a recorded event, not a sprint deadline.
- Sample
Carries a visible "Sample" badge. Demo data, never presented as real.
- Projected
A modeled number. Always carries the "(projected)" suffix.
- Scaffold
A real database row exists, outcome still pending.
- Owner-approved empty today
A business owner approved this action. Requires a recorded approval event.
- Delivered empty today
The message was actually sent. Requires a delivery event.
- Responded empty today
The customer replied. Requires a response event.
- Recovered empty today
A verified win. The count is fact; the dollar value stays projected.
- Revenue confirmed empty today
A real, measured dollar. No "(projected)" suffix allowed before this rung.
Where we sit today: rung 2. Rungs 3 through 7 are empty. They fill only with real, attributable events from a live pilot. When the first measured dollar exists, this page will say so, and that claim will have survived an adversarial refutation pass before it ships.
The gates that run on every build
- Anti-slop gate: scans the rendered HTML of every built page for kill-on-sight writing patterns: em-dashes in buyer copy, negation reframes, significance inflation, sycophantic openers, hedge stacking, and percentages with no source. One hit fails the build.
- Contradiction gate: key surfaces, including our pricing page and the machine-readable feed AI assistants read, must agree on pricing, product claims, and proof status. A contradiction fails the build.
- PII gate: our pilot clients are real small businesses. Their names are hard-banned from this codebase, and the gate scans every change.
- Brand and contrast gates: WCAG AA contrast and design-token discipline, machine-checked so the calm, readable design cannot quietly drift.
- Anti-theater gate (in the product): inside the RelayLaunch console, no animation may run without a real event behind it, no status indicator may be driven by a timer pretending to be activity, and no counter may render a number it cannot cite to a database row. This gate blocks every console build.
Builds this firewall has already failed
- 2026-06-10: the contradiction gate's first production run flagged a missing proof-status disclosure in our machine-readable AI feed. The fix was written, and the gate then flagged the fix itself for over-broad language. Both were corrected before anything deployed.
- 2026-06-10: the anti-theater gate's first full sweep of our console found 46 violations: fabricated activity tokens, a fake timer driving a status indicator, and placeholder metrics. All 46 were removed the same day, and the gate now blocks every console build.
Our own adversarial scoring system currently rates the company 30 out of 100, with demand and revenue scored at zero. We publish that here because the instrument only works if we cannot opt out of its verdicts.
What you will not find on this site
- No testimonials. We run two pilot programs; both are unpaid and we say so wherever they appear.
- No inflated customer counters. One paid engagement (a website + console build); $0 AI-product (SaaS) revenue.
- No recovered-dollar claims. AI-recovered revenue today: $0. This line changes only when a Pulse recovery receipt exists.
- No fake liveness. If something moves on a RelayLaunch surface, a real recorded event moved it.
Why this matters for your business
The same firewall runs inside the product. Relay drafts revenue-recovery messages for your business, and you approve every send. When a rule stops an action, a client who never opted in, or a send while you are offline, the system writes a brake receipt: what was held, which rule fired, and why. Your AI keeps receipts, including for the things it refused to do.