test(quick-260610-jrk): assert drag-created edge persists (exact =3)

DOM evidence (instrumented area.addPipe trace) proves the drag-to-connect a→c
connection IS created (connectioncreated, programmatic=0) AND renders with a
real bezier d that persists — drag-to-connect fully works. The prior
"doesn't persist" deferral was a misdiagnosis: Sink's input was a
single-connection input (ClassicPreset multiple:false default), so dropping a→c
onto c's already-occupied input correctly EVICTED b→c, leaving the count at 2
even though a→c persisted.

Give Sink's input multiple:true so a→c ADDS a third edge instead of replacing
b→c, then upgrade the post-drag assertion from a soft >=2 to a strict =3 drawn
paths (the persistence proof). No FlowCanvas source/emitter change — leaves
regenerate byte-identical, dist-parity zero-drift. Passes 6/6 in the container.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: serpentblade

examples/demos/FlowCanvasDemo.rozie

@@ -41,7 +41,15 @@
   nodes: [
     { id: 'a', label: 'Source', x: 20,  y: 40,  data: { kind: 'in' },  outputs: [{ key: 'out', label: 'out' }] },
     { id: 'b', label: 'Filter', x: 260, y: 40,  data: { kind: 'op' },  inputs: [{ key: 'in', label: 'in' }], outputs: [{ key: 'out', label: 'out' }] },
-    { id: 'c', label: 'Sink',   x: 500, y: 120, data: { kind: 'out' }, inputs: [{ key: 'in', label: 'in' }] },
+    // Sink's input is `multiple: true` so a drag-created a→c edge ADDS a third
+    // connection (a→b, b→c, a→c) instead of REPLACING the existing b→c. Rete's
+    // ClassicPreset single-input (multiple:false, the default) holds exactly one
+    // connection, so dropping a→c onto an already-occupied single input correctly
+    // evicts b→c — leaving the edge count at 2 and masking the fact that the new
+    // a→c edge DID persist. A multiple input lets the drag spec assert the
+    // drag-created edge persists as an exact 3rd drawn path (the persistence
+    // proof) rather than a soft ≥2.
+    { id: 'c', label: 'Sink',   x: 500, y: 120, data: { kind: 'out' }, inputs: [{ key: 'in', label: 'in', multiple: true }] },
   ],
   connections: [
     { id: 'e1', source: 'a', sourceOutput: 'out', target: 'b', targetInput: 'in' },

tests/visual-regression/specs/rete-flow.spec.ts

@@ -385,15 +385,24 @@ for (const target of TARGETS) {
         timeout: 10_000,
       });
     }
-    // …and after release the rubber-band preview is torn down, so the drawn-path
-    // count settles back to the committed config-array edges (a→b, b→c) — the
-    // transient preview pseudo no longer inflates the count. We assert the committed
-    // edges persist (≥2) rather than an exact 3: the freshly drag-created a→c edge
-    // exists in the editor model (the event fired) but the rendering of a connection
-    // created *during* a drag gesture is a separate Rete connection-view lifecycle
-    // concern, independent of the preview-line fix under test here.
+    // PERSISTENCE PROOF (260610-jrk continuation): after release the rubber-band
+    // pseudo is torn down, but the freshly drag-created a→c edge is a REAL
+    // connection that renders and PERSISTS with a non-empty `d` — so the drawn-path
+    // count settles to EXACTLY 3 (a→b, b→c committed + a→c drag-created). This
+    // upgrades the earlier soft `≥2` (which a vanishing a→c would have passed) to a
+    // strict `=3` that fails if the drag-created edge does not keep a drawn path.
+    //
+    // Why this is now assertable: the DOM evidence in the 260610-jrk continuation
+    // (an instrumented `area.addPipe` trace) proved the a→c connection IS created
+    // (`connectioncreated`, programmatic=0) AND rendered with a real bezier `d`. The
+    // earlier "doesn't persist" reading was a misdiagnosis — Sink's input was a
+    // single-connection input (ClassicPreset `multiple:false` default), so dropping
+    // a→c onto c's already-occupied input correctly EVICTED b→c (`connectionremoved`
+    // for e2), leaving the count at 2 even though a→c persisted. Sink's input is now
+    // `multiple: true` in FlowCanvasDemo, so a→c ADDS a third edge instead of
+    // replacing b→c — making the persistence directly countable.
     await expect
       .poll(drawnCount, { timeout: 10_000, intervals: [100, 300, 600, 1000] })
-      .toBeGreaterThanOrEqual(2);
+      .toBe(3);
   });
 }