Freelopers Start a conversation

Guard Evidence Engine · Public protocol

Measure first. Change second. Claim last.

The Freelopers redesign is also a Guard case study. This page separates the frozen before state, the intervention, the after measurements, and the limits of what can be concluded.

Before frozen 2026-07-16

Current phase: intervention build.

The theme is being built from scratch. No production after result exists yet.

Before observations

The starting point stays visible.

These are selected observations from the frozen baseline. The research archive retains route-level and engine-level detail.

Surface Observation Context
Homepage mobile performance 47 Fresh Lighthouse lab run; LCP 10.14 seconds and CLS 0.182.
Tracked route accessibility 91–97 Repeated contrast and heading-order findings; Contact included an ARIA issue.
AI assistant mentions 0 / 60 Two successful before windows across Gemini and ChatGPT; combined 95% Wilson interval 0–6.0%.
Homepage corrected AI Readiness 20 / 100 Draft-only LLM information was excluded from the live score.
Homepage corrected GEO 8 / 100 Corrected to use AI Readiness for the analyzed URL rather than an arbitrary page.
Security headers D · 50 / 100 Baseline observation; remediation may require platform-level configuration beyond the theme.

Measurement note: Lab scores are observations, not universal truths. AI visibility is binary within a defined prompt-and-engine panel, not a claim about all AI systems.

Protocol

A case study another team can inspect.

  1. Freeze raw before data

    Store route-level Lighthouse, Guard audits, AI Readiness, GEO, security, and AI-assistant observations.

  2. Define the intervention

    Use a version-controlled, semantic, low-runtime HubSpot theme and record platform changes separately.

  3. Preserve a holdout

    Leave the existing Delight route unchanged to help detect broad scanner or external movement.

  4. Run technical after checks

    Repeat mobile and desktop measurements after staging and production stability.

  5. Repeat AI visibility windows

    Use the frozen prompt set, engines, and success criteria, then report engine-specific observations and intervals.

Intervention hypotheses

Every design choice should predict an observable effect.

Runtime

Remove heavy decorative JavaScript.

Expected effect: lower transfer and main-thread cost, earlier hero rendering, and improved mobile LCP.

Semantics

Clarify entities and page purpose in raw HTML.

Expected effect: stronger structured-content, AI Readiness, and retrieval signals after crawl windows.

Access

Build contrast, headings, focus, and navigation correctly.

Expected effect: removal of repeated accessibility failures and more usable keyboard/mobile interaction.

Limits

What this experiment cannot prove on its own.

  • 01
    It is not a randomized trial.

    The intervention and holdout are different pages with different content and traffic.

  • 02
    Automated accessibility is incomplete.

    Passing scanner checks does not establish full WCAG or EAA compliance.

  • 03
    AI discovery takes time.

    Immediate technical improvements do not guarantee assistant mentions within a short window.

  • 04
    Platform behavior matters.

    HubSpot chat, headers, cookies, and injected scripts may require portal configuration beyond theme code.