Carry evidence through built-in generation retries

internal/observation/recorder.go

@@ -28,7 +28,7 @@ type Recorder struct {
 	capacityWait         time.Duration
 	generationDuration   time.Duration
 	retryDelay           time.Duration
-	usage                usageTotals
+	usage                UsageAccumulator
 }
 
 // Selection contains resolved identities only. Unresolved fields stay empty.
@@ -79,8 +79,7 @@ func New(observer domain.ExecutionObserver) (*Recorder, error) {
 }
 
 func newRecorder(runID string, now func() time.Time, observer domain.ExecutionObserver) *Recorder {
-	return &Recorder{runID: runID, startedAt: now(), now: now, observer: observer,
-		usage: newUsageTotals()}
+	return &Recorder{runID: runID, startedAt: now(), now: now, observer: observer}
 }
 
 // NewRunID creates the identity shared by observations and successful results.
@@ -119,7 +118,7 @@ func (r *Recorder) Attempt(data AttemptData) {
 			r.lastRepairGeneration = data.Generation
 		}
 		r.generationDuration = addDuration(r.generationDuration, data.GenerationDuration)
-		r.usage.add(data.Usage, data.UsagePresence)
+		r.usage.Add(data.Usage, data.UsagePresence)
 	}
 	event := r.event(domain.ExecutionAttempt)
 	event.Generation = data.Generation
@@ -171,7 +170,7 @@ func (r *Recorder) Finish(data FinishData) {
 	if event.Duration < 0 {
 		event.Duration = 0
 	}
-	event.Usage = r.usage.snapshot(r.attempts)
+	event.Usage = r.usage.Observed()
 	r.deliver(event)
 }
 
@@ -243,7 +242,29 @@ type usageTotals struct {
 	prompt, completion, total, cached, cacheWrite usageField
 }
 
-func newUsageTotals() usageTotals {
+// UsageAccumulator combines presence-aware usage across dispatched attempts.
+// Its zero value represents no dispatched attempts.
+type UsageAccumulator struct {
+	totals   usageTotals
+	attempts uint64
+}
+
+// Add includes the usage reported by one dispatched attempt.
+func (u *UsageAccumulator) Add(usage domain.TokenUsage, presence *domain.TokenUsagePresence) {
+	if u.attempts == 0 {
+		u.totals = completeUsageTotals()
+	}
+	u.attempts++
+	u.totals.add(usage, presence)
+}
+
+// Observed returns aggregate counts, marking every field known and complete
+// when no attempt has been added.
+func (u UsageAccumulator) Observed() domain.ObservedTokenUsage {
+	return u.totals.snapshot(u.attempts)
+}
+
+func completeUsageTotals() usageTotals {
 	return usageTotals{prompt: usageField{complete: true}, completion: usageField{complete: true},
 		total: usageField{complete: true}, cached: usageField{complete: true}, cacheWrite: usageField{complete: true}}
 }

internal/observation/usage_accumulator_test.go

@@ -0,0 +1,27 @@
+package observation
+
+import (
+	"testing"
+
+	"gitea.maximumdirect.net/eric/promptkit/internal/domain"
+)
+
+func TestUsageAccumulatorPreservesPresenceAndNoAttemptSemantics(t *testing.T) {
+	var empty UsageAccumulator
+	zero := empty.Observed()
+	if !zero.PromptTokens.Known || !zero.PromptTokens.Complete || zero.PromptTokens.Value != 0 ||
+		!zero.CacheWriteTokens.Known || !zero.CacheWriteTokens.Complete || zero.CacheWriteTokens.Value != 0 {
+		t.Fatalf("zero-attempt usage = %+v", zero)
+	}
+
+	var usage UsageAccumulator
+	usage.Add(domain.TokenUsage{PromptTokens: 0}, &domain.TokenUsagePresence{PromptTokens: true})
+	usage.Add(domain.TokenUsage{PromptTokens: 3}, &domain.TokenUsagePresence{PromptTokens: true})
+	observed := usage.Observed()
+	if observed.PromptTokens.Value != 3 || !observed.PromptTokens.Known || !observed.PromptTokens.Complete {
+		t.Fatalf("reported prompt usage = %+v", observed.PromptTokens)
+	}
+	if observed.CompletionTokens.Known || observed.CompletionTokens.Complete {
+		t.Fatalf("unreported completion usage = %+v", observed.CompletionTokens)
+	}
+}