Eight field artifacts. Each one is a class of problem solved — what the agent saw before, what it saw after, what stayed durable. Names and numbers stay sealed; the shape of the work does not.
Site lost the index after a platform shift. Canonical layer rebuilt from log signal, crawl budget redirected, structured data re-anchored. Traffic curve back online inside one cycle.
canonical :: 4,212 paths rewritten
crawl-budget :: +180% reclaimed
Property data restructured for agent parsing across four locales. Schema survives translation without semantic drift. Cross-market entity continuity preserved.
@type :: RealEstate
hreflang :: 4 locales · stable
Custom MCP server exposing crawl diagnostics, schema validity, redirect chains, and Core Web Vitals to AI agents. Site became something a model could query, not just scrape.
mcp:// audit · 7 tools
returns :: machine-readable
Large reorganization with canonical, schema, and entity continuity preserved across the rebuild. Internal linking architecture migrated without breaking the equity graph.
301 :: 12,800 paths mapped
loss :: < 2% within cycle
Log file analysis revealing where crawler budget was bleeding into faceted noise. Surface re-routed, parameter handling rewritten, agent attention restored to canonical paths.
noise paths :: identified
parameter rules :: 17 new
LCP, INP, server response trimmed to the thresholds that decide whether agents bother reading. Critical path slimmed; agent recommendation probability rose with it.
LCP :: 4.2s → 1.6s
INP :: 320ms → 110ms
Mapped a brand into the knowledge layer — sameAs, identifiers, author trail, structured provenance. Agents stopped confusing the brand with similarly named entities.
sameAs :: 11 anchors
identifier :: 4 surfaces
Content architecture refactored into topic clusters with entity-aware internal linking. Pages stopped competing with themselves; the cluster became the unit agents recommend.
clusters :: 6 pillars
internal links :: graph rebuilt