diff --git a/docs/plans/2026-02-22-backend-result-type.md b/docs/plans/completed/2026-02-22-backend-result-type-original.md
similarity index 100%
rename from docs/plans/2026-02-22-backend-result-type.md
rename to docs/plans/completed/2026-02-22-backend-result-type-original.md
diff --git a/docs/plans/completed/backend-result-type.md b/docs/plans/completed/backend-result-type.md
new file mode 100644
index 0000000..6f96eda
--- /dev/null
+++ b/docs/plans/completed/backend-result-type.md
@@ -0,0 +1,38 @@
+# Backend Result Type — Completion Summary
+
+**Completed:** 22 February 2026
+**Module:** `forge.lthn.ai/core/go-ml`
+**Status:** Complete — unified Result struct across all backends
+
+## What Was Built
+
+Refactored Generate/Chat return types across all ML backends (HTTP, Llama,
+MLX adapter) to use a unified `Result` struct carrying both generated text
+and inference metrics.
+
+### Result struct
+
+```go
+type Result struct {
+    Text    string
+    Metrics Metrics
+}
+
+type Metrics struct {
+    PromptTokens     int
+    CompletionTokens int
+    TotalTokens      int
+    LatencyMs        float64
+    TokensPerSecond  float64
+}
+```
+
+### Backends updated
+
+- HTTP backend (Ollama, OpenAI-compatible endpoints)
+- Llama backend (llama.cpp via CGo)
+- MLX adapter (delegates to go-mlx)
+
+All tests updated to use the new return type. No breaking changes to the
+public `ml.Generate()` / `ml.Chat()` API — the Result struct is returned
+where previously only a string was.