[harlan-zw]

🔗 Linked issue

Experiment B in the SEO recovery brief.

❓ Type of change

• 📖 Documentation
• 🐞 Bug fix
• 👌 Enhancement
• ✨ New feature
• 🧹 Chore
• ⚠️ Breaking change

📚 Description

Experiment B needs ten eligible destinations and dated baselines before any README links start its measurement windows.
Add a candidate manifest and activation rules, preserving the original 41 URLs and the 11 November gate.
Clarify that the five-of-ten threshold is an observational signal, with a separate 14-day window for each live link.

🤖 AI disclosure: Harlan Agent Kit modified this description. My AI open-source policy.

[harlan-github-agent]

🤖 REVIEW SKIPPED

Harlan Agent Kit posted this automated triage. AI open source policy. Last updated: 2026-10-21 06:44 UTC.

• Reason: model: classification chose skip with confidence 0.72.
• Override: Add the harlan-agent-review label to force an adversarial Review.

[harlan-zw]

🤖 Agent evidence for the ten-candidate preparation:

On 21 October, all ten destinations returned direct HTTP 200, allowed indexing, declared their own canonical URLs, and appeared in the Skills sitemap.
Each Skill detail reported resolved source and current admission. Both badge themes returned HTTP 200 SVG for each destination.
The checked default-branch READMEs contained no skilld.dev string; the manifest records their source URLs and blob SHAs.
This bounded README check does not establish absence of other outside links.

NuxtSEO's retained Google inspections reported unknown to Google, with check dates from 20 August to 13 October.
These are dated pre-recruitment records, not fresh or adoption-day verdicts. No fresh inspection was requested for this preparation.
The existing panel's groups object matches the base exactly, including all 41 original URLs and exclusions.

[harlan-zw]

🤖 Harlan Agent Kit Agent assessment of head 2f9f483.

The Service Review was queued behind capacity: 12 Review or Repair Tasks, with four free host slots.
The exact-head stop returned Stopped before an independent read-only review began.
That assessment found no confirmed blocker. Current-head Actions CI passed.
This assessment is separate from Service READY; its Review remains stopped for this head.

• Logged, 40/100: The 14-day qualification upper bound is implicit in docs/work/seo-link-experiment.md:63. A day-20 indexed inspection cannot establish success within 14 days. Optional follow-up: require qualifying evidence within the window explicitly, keeping later observations separate.

The independent reviewer confirmed all ten candidate records, exact badge embeds, and preservation of the original panel groups.
Review confidence: 95/100, with five deducted for late-evidence classification uncertainty.