docs: fix dangling issue-evaluation.md references

CHANGELOG.md

@@ -54,8 +54,8 @@
   connection, indefinitely. Default is now 30s (overridable, including
   back to `0`, via the existing `WithSessionWriteTimeouts`). `sendPacket`
   also now logs a warning when a single write exceeds 250ms — previously
-  there was no observable signal for this at all. Root-cause writeup and
-  regression tests in `issue-evaluation.md`. (#19)
+  there was no observable signal for this at all. See PR #19 for the full
+  root-cause writeup and regression tests. (#19)
 - `relay/store.EventStore.FindEventBytes` returned a bbolt-transaction-
   scoped byte slice after its own read transaction had already closed —
   invalid per bbolt's own contract, and reproducibly served corrupted

relay/config.go

@@ -14,8 +14,8 @@ const (
 	// may block on conn.WriteJSON before the session gives up and closes
 	// the connection. Every subscription on a connection shares one
 	// outgoing pipe, so an unbounded deadline let one stuck reader stall
-	// delivery to all of them indefinitely (see issue-evaluation.md). 30s
-	// is a deliberately generous default, not a tuned optimum. Override
+	// delivery to all of them indefinitely (see PR #19). 30s is a
+	// deliberately generous default, not a tuned optimum. Override
 	// via WithSessionWriteTimeouts (0 restores the old unbounded behavior).
 	defaultDataWriteTimeout = 30 * time.Second
 

relay/session.go

@@ -32,9 +32,9 @@ const (
 
 	// slowSendWarnThreshold is how long a single sendPacket call may take
 	// before it's logged as a warning -- surfaces a backed-up outgoing
-	// pipe (see issue-evaluation.md), which previously had no observable
-	// signal at all. Several multiples of the poll interval, to avoid
-	// firing on ordinary jitter.
+	// pipe (see PR #19), which previously had no observable signal at
+	// all. Several multiples of the poll interval, to avoid firing on
+	// ordinary jitter.
 	slowSendWarnThreshold = 250 * time.Millisecond
 )
 

relay/session_backpressure_test.go

@@ -97,7 +97,7 @@ func readUntil(t testing.TB, conn *websocket.Conn, deadline time.Time, fn func(w
 }
 
 // TestSessionSlowReaderStallsEveryOtherSubscriptionOnThatConnection
-// reproduces the delivery stall (issue-evaluation.md) at report-realistic
+// reproduces the delivery stall fixed in PR #19 at report-realistic
 // scale: 3 fresh events on the filter under test, not the 55+
 // TestSubscriptionBackpressureDelaysButNeverLosesEvents needs.
 //

relay/subscription_test.go

@@ -180,9 +180,9 @@ func TestSubscriptionCombinedKindBurstDoesNotStarveFreshEvent(t *testing.T) {
 }
 
 // TestSubscriptionBackpressureDelaysButNeverLosesEvents reproduces the
-// no-timeout blocking-send mechanism behind the delivery stall in
-// issue-evaluation.md: storeScan.handleEvents' send into a subscription's
-// outgoing channel (eventBufferCapacity, 55 slots) has no timeout, and
+// no-timeout blocking-send mechanism behind the delivery stall fixed in
+// PR #19: storeScan.handleEvents' send into a subscription's outgoing
+// channel (eventBufferCapacity, 55 slots) has no timeout, and
 // Subscription.Start's poll loop calls Fetch synchronously on every tick --
 // so once the channel is full, the next poll tick blocks inside Fetch until
 // the consumer drains it, stalling delivery of any new matching event.