Week 9: The product starts breathing

This was the densest week since launch. 133 commits. Not refactors or dependency bumps, but real product surface. The theme: Corexi should feel alive. Not a tool you run once and file away, but something that watches, learns, and keeps working after you close the tab.

The dashboard breathes

We added ambient animations to the Analysis Layers card. The three status dots pulse slowly, a faint heartbeat line scrolls along the bottom edge, and a barely-visible gradient sweeps across the card like a radar scan. None of it demands attention. All of it signals: this thing is running. Continuously.

It sounds cosmetic. It isn't. The difference between "here's your report" and "we're watching your product right now" is the difference between an audit tool and a continuous UX layer. That distinction lives in details like these.

Annotated screenshots

Findings now pin directly to the screenshot. Each issue gets a numbered marker on the exact element it refers to. Hover a finding in the list, the corresponding bounding box pulses on the screenshot. Click a pin, the finding card scrolls into view. No more reading "Button contrast too low" and wondering which button.

This works on both the overview page and the full report, with synced hover/click state between the screenshot and the findings panel.

UX Assistant

A new AI chat sidebar lives inside scan reports. Ask it "What should I fix first?" or "Why is my score below 70?" and it answers in context, aware of your scan data, findings, and score breakdown. It's not a generic chatbot. It reads your report and gives specific, actionable guidance.

Product Canvas

A new visual page that maps how everything connects: your site, Corexi Snippet, analytics providers, AI engine, findings, IDE delivery. Built with React Flow with draggable nodes, zoom, and a dark theme. It's the "how does this all work together?" answer in one glanceable diagram.

Overview is the report

We killed the friction. Previously, the site overview showed a summary and you had to click "Open full report" to see anything useful. Now the overview is the report. Inline findings with severity badges, page-level scores, annotated screenshots, UX category breakdowns, Neurodiversity Lens, Color Vision Accessibility, all right there. No extra click.

The full report view still exists for deep dives, but the overview now stands on its own.

Findings got practical

Select findings with checkboxes. Bulk copy as IDE prompts for Cursor, Claude Code, Windsurf, VS Code, or Replit. Download as markdown. A floating action bar appears at the bottom when you select, letting you copy prompts, copy fix code, or download .md. The goal: findings should flow into your editor in under 5 seconds.

AI engine quality

Three changes that matter:

1. Source-based fingerprinting. Findings are now deduplicated by their root cause, not surface text. Same contrast issue on 5 pages = 1 finding, not 5.
2. Anti-hallucination prompt. The AI is now explicitly instructed to report only what it can verify in the screenshot. No invented metrics, no assumed behaviors.
3. Mandatory behavioral correlation. When behavioral data exists (snippet, GA4, Clarity), the engine must reference it. No more visual-only findings when we have session data.

Demo for everyone

The live demo now works properly for everyone. Sign in as the demo account, get a real dashboard with real scan data (Bolt as the demo product), all in read-only mode. Settings page is visible but disabled. "Run new analysis" is grayed out. 100 people can use it simultaneously, each gets their own session, nobody can break anything.

If you're already logged into your own account and click "Try live dashboard," we sign you out first, log into demo, and when you exit, you're back on the marketing site. No session conflicts.

Behavioral Snippet: eating our own cooking

We installed the Corexi Snippet on corexi.ai itself. Every pageview, scroll depth, rage click, dead click, and form abandonment on our own site now flows into the scoring engine. The snippet shows as "Active" in settings, events aggregate hourly, and the next scan will include real behavioral data from real visitors.

If we're going to tell customers to install a snippet, we should be running it ourselves first.

Landing page rebuilt

The homepage now shows real dashboard screenshots instead of placeholder mockups. Industry-specific cards (SaaS, E-commerce, Fintech, Healthcare) with actual scan examples. CRO-optimized section order. The video moved to a modal to reduce scroll depth. Quick visual scan CTA for instant gratification, Request Demo for serious conversations.

Journey becomes a blog

Journey got a blog section. Weekly build logs stay in their own "Weekly Journey Log" area (latest featured, rest collapsed). Blog posts live below, covering announcements, milestones, and deep dives. First blog post: Esade eWorks acceptance. The navigation moved Journey from the Company dropdown to a standalone link: Product | Developers | Journey | Company.

Esade eWorks 2026

We got accepted into the Esade eWorks Acceleration Program, one of Europe's most selective university-backed accelerators, run by Esade Business School in Barcelona. Mentorship, network, validation. The selection committee saw what we see: enterprise and scale-up teams need a continuous UX layer, and nobody else is building it.

Nordics are calling

Demo requests landed from Finland and Denmark this week, from scale-up teams building products for European markets. The pattern continues: teams that ship fast need something watching UX quality continuously, not quarterly audits that arrive after the damage is done.

Everything else

• PX Score became UX Score. Renamed across the entire product. Clearer, more universal.
• Security page. /security for IT and compliance teams evaluating Corexi.
• Proactive AI recommendations. Clickable insights that link directly to filtered findings.
• Sitemap + JSON-LD. Proper SEO infrastructure for all pages and blog posts.
• Session handling. Fixed marketing header showing "Dashboard" after session expires.
• Full-page screenshots. Scans now capture the entire page, not just the viewport.
• MCP auto-create. trigger_scan with a URL automatically creates the site if it doesn't exist.
• Auto-scan on product creation. Create a product, first scan starts immediately.

The number

133 commits. One week. The product went from "good foundation" to "this feels like something real." It breathes, it pins findings to pixels, it chats with you about your UX, it maps its own architecture, and it caught the attention of one of Europe's top accelerators.

Try the live demo or set up the MCP server in your IDE. It takes 60 seconds.

Week 10 starts tomorrow.