docs: align contributor fallback guidance

Author: rlippmann

CONTRIBUTING.md

@@ -56,10 +56,11 @@ This package does own the human-input drafting boundary. Documentation and
 specification changes in this repository should treat the drafter as responsible
 for converting messy user input into one of three non-authoritative outcomes:
 
-- `directive`: propose one canonical directive string
-- `no_directive`: classify the message as not requesting a directive
-- `unknown`: preserve uncertainty, malformed recovery failure, or unresolved
-  directive-like intent without guessing
+- `CanonicalDirective`: propose one canonical directive
+- `RejectedDirective`: classify the message as not requesting a directive or
+  reject an unsafe or malformed candidate
+- `UnknownDirective`: preserve uncertainty or unresolved directive-like intent
+  without guessing
 
 The authoritative acquisition contract lives in [docs/DrafterAcquisitionSpec.md](docs/DrafterAcquisitionSpec.md).
 

docs/PythonFallbacks.md

@@ -34,8 +34,9 @@ drafter = DirectiveDrafter(fallback=fallback, fallback_source="openai")
 ```
 
 Use `create_async_openai_fallback(...)` with
-`DirectiveDrafter(async_fallback=...)` and
-`async_draft_directive(...)` for an asynchronous host path. OpenAI-compatible
+`DirectiveDrafter(async_fallback=..., async_fallback_source="openai")` and
+`async_draft_directive(...)` for an asynchronous host path. Set the source
+explicitly because the async source is configured separately. OpenAI-compatible
 endpoints can provide a custom `base_url`.
 
 ## LiteLLM