fix: clarify ForceEndTurn deployment behavior

[GregHolmes]

Summary

• State that ForceEndTurn is available on hosted deployments, including EU and AU, while self-hosted availability follows the deployed release.
• Report FORCE_END_TURN_UNSUPPORTED immediately in the Voice Agent example.
• Link the frozen EOT description patch to deepgram-docs #1202 and correct the Pydantic v1 unfreeze condition.
• Restore the socket-client grouping in the generated-patch ledger.

Validation

• python3 -m compileall on both changed examples
• git diff --check
• ruff check in a Python 3.12 container

[github-actions]

Package	Line Rate	Branch Rate	Complexity	Health
src.deepgram	97%	94%	0	✔
src.deepgram.agent	100%	100%	0	✔
src.deepgram.agent.v1	98%	100%	0	✔
src.deepgram.agent.v1.settings	100%	100%	0	✔
src.deepgram.agent.v1.settings.think	100%	100%	0	✔
src.deepgram.agent.v1.settings.think.models	97%	100%	0	✔
src.deepgram.auth	100%	100%	0	✔
src.deepgram.auth.v1	100%	100%	0	✔
src.deepgram.auth.v1.tokens	97%	100%	0	✔
src.deepgram.core	88%	81%	0	➖
src.deepgram.errors	100%	100%	0	✔
src.deepgram.helpers	98%	92%	0	✔
src.deepgram.listen	100%	100%	0	✔
src.deepgram.listen.v1	98%	93%	0	✔
src.deepgram.listen.v1.media	97%	100%	0	✔
src.deepgram.listen.v2	98%	93%	0	✔
src.deepgram.manage	100%	100%	0	✔
src.deepgram.manage.v1	100%	100%	0	✔
src.deepgram.manage.v1.models	96%	100%	0	✔
src.deepgram.manage.v1.projects	97%	100%	0	✔
src.deepgram.manage.v1.projects.billing	100%	100%	0	✔
src.deepgram.manage.v1.projects.billing.balances	96%	100%	0	✔
src.deepgram.manage.v1.projects.billing.breakdown	97%	100%	0	✔
src.deepgram.manage.v1.projects.billing.fields	97%	100%	0	✔
src.deepgram.manage.v1.projects.billing.purchases	97%	100%	0	✔
src.deepgram.manage.v1.projects.keys	96%	100%	0	✔
src.deepgram.manage.v1.projects.members	97%	100%	0	✔
src.deepgram.manage.v1.projects.members.invites	96%	100%	0	✔
src.deepgram.manage.v1.projects.members.scopes	96%	100%	0	✔
src.deepgram.manage.v1.projects.models	96%	100%	0	✔
src.deepgram.manage.v1.projects.usage	98%	100%	0	✔
src.deepgram.manage.v1.projects.usage.breakdown	97%	100%	0	✔
src.deepgram.manage.v1.projects.usage.fields	97%	100%	0	✔
src.deepgram.read	100%	100%	0	✔
src.deepgram.read.v1	100%	100%	0	✔
src.deepgram.read.v1.text	98%	100%	0	✔
src.deepgram.self_hosted	100%	100%	0	✔
src.deepgram.self_hosted.v1	100%	100%	0	✔
src.deepgram.self_hosted.v1.distribution_credentials	96%	100%	0	✔
src.deepgram.speak	100%	100%	0	✔
src.deepgram.speak.v1	98%	97%	0	✔
src.deepgram.speak.v1.audio	91%	80%	0	✔
src.deepgram.speak.v2	98%	93%	0	✔
src.deepgram.speak.v2.audio	100%	100%	0	✔
src.deepgram.voice_agent	100%	100%	0	✔
src.deepgram.voice_agent.configurations	95%	100%	0	✔
src.deepgram.voice_agent.variables	95%	100%	0	✔
Summary	95% (6538 / 6855)	91% (1438 / 1574)	0	✔

Scope: hand-maintained SDK logic. Fern-generated data models (types/, requests/), package __init__.py files, version.py, and the unused core/http_sse/ scaffolding are excluded — see .coveragerc. Unscoped whole-package coverage is ~70%.

[dg-coreylweathers]

What this PR does
Changes the ForceEndTurn examples and the conversational-STT skill file to say the control is available on Deepgram-hosted deployments including EU and AU (self-hosted depends on the release), makes the Voice Agent example fail fast on FORCE_END_TURN_UNSUPPORTED, and edits the manual-patch ledger.

What I checked

• Is the availability claim true? Yes: the repo's own tests/manual/listen/v2/force_end_turn/main.py passed all four steps with trigger: manual against api.deepgram.com, api.eu.deepgram.com, and api.au.deepgram.com.
• Is FORCE_END_TURN_UNSUPPORTED real? Yes, it is the Voice Agent warning for a non-Flux listen provider (deepgram-docs schemas.agent.v1.yml:220), and both Warning and Error carry .code/.description.
• Gates in Docker: mypy clean, ruff clean on changed files, pytest 1034 passed.

Blocking

• The AGENTS.md/.fernignore hunk for listen_v2eot_threshold.py and requests/deepgram_listen_provider_v2.py is stale: main's AGENTS.md:85 already says Fern emits that description natively as of the 2026-09-23 regen and the files were unfrozen, so main's .fernignore no longer lists them. The branch is also CONFLICTING. Fix: rebase onto main and drop that hunk entirely; keep only the StrictInt unfreeze-condition change.

Should-fix

• "Self-hosted availability depends on the deployed release" (examples/16:24-25, SKILL.md:94) omits the switch. deepgram-docs says it needs listen_v2_force_end_turn = true under [features] on the 2026-09-15 release or later. Say that.
• tests/manual/listen/v2/force_end_turn/main.py:24-27,174 and async.py:8,115 still say "gated per deployment" / "not enabled on this deployment"; align them with the new sentence.

Nit: examples/32:94-104, the trailing rejected-check can raise after agent_finished is set; fine if intended, worth a comment.

[dg-coreylweathers]

Requesting changes for two should-fixes and three small cleanups. Everything the PR claims checked out live; these are about the text and the example around it.

What this PR does: ForceEndTurn is a message that lets an app end a speaker's turn right away instead of waiting for Flux to detect it. The examples and manual tests used to call it "gated per deployment." They now say it works on Deepgram-hosted US, EU and AU, and that self-hosted customers need the 2026-09-15 release plus listen_v2 = true and listen_v2_force_end_turn = true under [features]. The Voice Agent example now stops at once on FORCE_END_TURN_UNSUPPORTED instead of timing out after 15 seconds.

What I checked:

• Hosted availability: I ran tests/manual/listen/v2/force_end_turn/async.py today against api.deepgram.com, api.eu.deepgram.com and api.au.deepgram.com. All three returned EndOfTurn with trigger="manual".
• Self-hosted claim: release-260915 shipped on 2026-09-15 with Flux Forced End-of-Turn. The flag names match what the self-hosted team gave on 9/4.
• Voice Agent example: it completes against prod with Flux. With a Nova-3 listener it raises FORCE_END_TURN_UNSUPPORTED after 2 seconds.
• Tests and lint: the new test passes, ruff is clean, and CI is green on 3.10–3.13.

Please fix:

1. reference.md:5417 still says "Requires server-side enablement; deployments without the feature reject the message and close the connection," so a hosted developer reads it as "this may not work for me." Replace it with:
   - Available on Deepgram-hosted deployments (including EU and AU). Self-hosted deployments need the 2026-09-15 release or later with the ForceEndTurn feature enabled; unsupported deployments reply UNPARSABLE_CLIENT_MESSAGE and close the connection
2. Delete the "Link the frozen EOT description patch to deepgram-docs #1202" bullet from the PR description. Nothing in the diff makes that change.
3. examples/16-transcription-force-end-turn.py:59: the docstring still says "Because ForceEndTurn is gated … point it at a deployment that has the feature." Suggested: "Accepts DEEPGRAM_BASE_URL=wss:// so it can target a regional or self-hosted deployment. Defaults to production."
4. examples/32-voice-agent-force-end-turn.py:125: raising inside the with block also prints "Connection error: sent 1011 (internal error)" after the real error. Record the failure, leave the with block, then raise.
5. examples/32-voice-agent-force-end-turn.py:96-97: the new lines add two mypy attr-defined errors because message is typed object. Typing the callback parameter as typing.Any clears them.

Open question (not blocking): the 9/4 thread in #engineering-self-hosted-releases that names the listen_v2_force_end_turn flag also planned to remove it. If a later self-hosted release drops the flag, "2026-09-15 or later with listen_v2_force_end_turn = true" will be wrong for newer releases. Do you know whether that removal has shipped?