Full @graph architecture
One consolidated JSON-LD @graph per page, designed across templates, with @id and sameAs chains stable enough for AI engines to merge entities cleanly.
What it actually is
Before any code ships, I design the @graph. One consolidated JSON-LD block per page emitted as a single @graph array. Every entity gets a stable @id (URL-fragment-style, not page-URL-style). Every relationship between entities is expressed as a reference, not a duplicated nested object. sameAs chains point at canonical external identifiers (Wikidata, Wikipedia, official social profiles).
Most agencies ship multiple separate JSON-LD blocks per page (Organization here, Product there, BreadcrumbList over there) and hope Google stitches them together. Google does, sometimes. AI engines mostly don't. The consolidated @graph is the architectural decision that makes entity resolution deterministic instead of accidental.
Deliverables
- Architecture decision record documenting the @graph design
- Entity-relationship diagram showing every @type, @id pattern, and reference
- @id naming convention spec (the rule your team will use forever)
- sameAs canonical-identifier list for the brand and its core entities
- Validator output confirming the design parses clean across all engines
What breaks without it
The @graph design is the decision the rest of the Sprint depends on. Get it right and every template implementation slots in cleanly. Get it wrong and you spend the next two years patching entity resolution bugs as new templates ship and the graph fragments again.
I've inherited Sprint scopes where the previous vendor shipped per-template JSON-LD with no shared @graph design. Every template invented its own @id pattern. The fix wasn't more schema. It was deleting the existing schema and starting from a graph design. Two weeks of design work would have saved them six months of remediation.
How it fits the Sprint
Architecture is the spine of the Sprint. Per-template implementation, validator suite, and entity resolution work all reference the architecture decision record. If you go into a Retainer afterward, the regression monitoring rules trace back to this same document.
The full Sprint breakdown
Stop pouring budget into a broken foundation.
If your SEO retainer hasn’t compounded, your AI citations have stalled, or your last technical audit ended in a deck nobody read, that’s not a content problem. It’s an engineering problem. The same engineer who diagnoses ships the fix.