Expand tables into a spreadsheet, and let diagrams be explored

db/seeds/development.rb

@@ -156,12 +156,26 @@ module DevelopmentSeed
     AGENT_ORGANIZE_RUN_ID = "development-seed-organize-run"
 
     CONTENT_FIXTURES = {
+      # Deliberately wider than a document column: this is the fixture that
+      # demonstrates the legibility floor (the diagram keeps its real size
+      # and scrolls rather than shrinking to unreadable) and the pan/zoom
+      # expanded view.
       flowchart: <<~MARKDOWN,
         ```mermaid
         flowchart LR
-          Client --> Gateway
-          Gateway --> API
-          API --> DB[(Database)]
+          Client[Merchant client] --> Gateway[Edge gateway]
+          Gateway --> Auth[Token exchange]
+          Auth --> API[Orders API]
+          API --> Rules[Eligibility rules]
+          Rules --> Engine[Pricing engine]
+          Engine --> Ledger[(Price ledger)]
+          Engine --> Risk[Risk scoring]
+          Risk --> Capture[Capture service]
+          Capture --> Settle[Settlement batch]
+          Settle --> Payout[(Payout store)]
+          API --> Events[Event feed]
+          Events --> Search[(Search index)]
+          Events --> Warehouse[(Warehouse)]
         ```
       MARKDOWN
       state_diagram: <<~MARKDOWN,
@@ -182,11 +196,23 @@ module DevelopmentSeed
           S-->>A: Receipt ID
         ```
       MARKDOWN
+      # Wide and long on purpose — the fixture that shows a table framed and
+      # wrapped in the document, and opened as a spreadsheet.
       table: <<~MARKDOWN,
-        | Metric | Control | Treatment |
-        | --- | ---: | ---: |
-        | Success | 61.2% | 62.9% |
-        | Errors | 18.4% | 17.7% |
+        | Variant | Cohort | Sessions | Success | Errors | p95 latency | Revenue / session | Notes |
+        | --- | --- | ---: | ---: | ---: | ---: | ---: | --- |
+        | Control | All | 41,208 | 61.2% | 18.4% | 412ms | $2.14 | Baseline ranker, unchanged since March |
+        | Treatment A | All | 40,876 | 62.9% | 17.7% | 438ms | $2.31 | Recall boost on long-tail queries |
+        | Treatment B | All | 12,004 | 59.8% | 21.3% | 507ms | $1.96 | Reranker on top; latency cost too high |
+        | Control | New | 8,912 | 54.1% | 22.8% | 401ms | $1.42 | New accounts see fewer personalized results |
+        | Treatment A | New | 8,844 | 58.6% | 19.9% | 430ms | $1.71 | Largest lift in the whole experiment |
+        | Treatment B | New | 2,610 | 55.2% | 24.1% | 512ms | $1.38 | Same latency story as the pooled cohort |
+        | Control | Returning | 32,296 | 63.1% | 17.2% | 414ms | $2.34 | |
+        | Treatment A | Returning | 32,032 | 64.1% | 17.1% | 440ms | $2.47 | Small but consistent across every week |
+        | Treatment B | Returning | 9,394 | 61.1% | 20.5% | 505ms | $2.12 | |
+        | Control | Mobile | 19,441 | 57.8% | 20.6% | 486ms | $1.88 | Mobile is latency-bound before it is ranking-bound |
+        | Treatment A | Mobile | 19,388 | 59.9% | 19.4% | 511ms | $2.02 | Lift holds, but the latency budget is nearly spent |
+        | Treatment B | Mobile | 5,702 | 54.3% | 25.7% | 604ms | $1.61 | Rejected on mobile outright |
       MARKDOWN
       code_walkthrough: <<~'MARKDOWN',
         How the discount engine decides what every line item costs. Each stage is