cmpnd 0.70.0__tar.gz → 0.72.0__tar.gz

{cmpnd-0.70.0 → cmpnd-0.72.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: cmpnd
-Version: 0.70.0
+Version: 0.72.0
 Summary: DSPy observability and deployment SDK for cmpnd
 Project-URL: Homepage, https://cmpnd.ai
 Author-email: cmpnd <hello@cmpnd.ai>

{cmpnd-0.70.0 → cmpnd-0.72.0}/cmpnd/cli/__init__.py

@@ -171,15 +171,8 @@ def main() -> None:
     evals.register(subparsers)
     datasets.register(subparsers)
     deployments.register(subparsers)
-
-    # Assessments and valsets CLIs are still in development, hidden behind
-    # CMPND_CLI_DEV=1 (mirroring the CMPND_CLI_ADMIN gate above). The env var
-    # only hides the commands from customers; the server still gates each action
-    # on the assessments:* / valsets:* scope, so this is discoverability, not
-    # enforcement.
-    if os.environ.get("CMPND_CLI_DEV") == "1":
-        assessments.register(subparsers)
-        valsets.register(subparsers)
+    assessments.register(subparsers)
+    valsets.register(subparsers)
 
     # Top-level Docker-style shortcuts (human-readable by default; --json opts in)
     top.register(subparsers)

{cmpnd-0.70.0 → cmpnd-0.72.0}/cmpnd/cli/commands/admin.py

@@ -14,8 +14,7 @@ default limits and feature flags, and `cmpnd admin signup …` sets who may sign
 and the plan a new account's org lands on. `cmpnd admin feature-flags …` reads and
 toggles the server's runtime flags. These hit
 /api/v1/admin/* and require an API key carrying the admin:* scopes — the server
-is the real authority. The CMPND_CLI_ADMIN env gate (mirroring CMPND_CLI_DEV)
-only keeps these out of customers' --help; it grants nothing.
+is the real authority. The CMPND_CLI_ADMIN env gate only keeps these out of customers' --help; it grants nothing.
 """
 
 from __future__ import annotations

{cmpnd-0.70.0 → cmpnd-0.72.0}/cmpnd/cli/commands/deployments.py

@@ -191,6 +191,9 @@ def submit_cli_run(
         raise
     except httpx.HTTPStatusError as exc:
         raise _http_refusal(exc) from exc
+    trace_id = settled.get("trace_id")
+    if isinstance(trace_id, str) and trace_id:
+        print(f"Trace: {trace_id}", file=sys.stderr)
     if settled["status"] == "failed":
         return settled, None
     result = settled.get("result")

{cmpnd-0.70.0 → cmpnd-0.72.0}/cmpnd/client.py

@@ -17,10 +17,6 @@ and propagates to its own eval / parallel threads. It registers nothing global
 and holds no OS resource: exporters are pooled by credentials at module scope
 (:mod:`cmpnd._exporter_pool`), so a client may fall out of scope with nothing to
 release. ``close()`` is an optional alias for end-of-scope flushing.
-
-The assessment and valset calls are in development. Each raises unless
-``CMPND_CLI_DEV=1`` is set, the switch that hides the in-development
-``cmpnd assessments`` and ``cmpnd valsets`` CLI groups. Tracing needs no flag.
 """
 
 from __future__ import annotations
@@ -111,15 +107,6 @@ class Client:
         if not self._active:
             raise RuntimeError("client is closed; create a new cmpnd.Client")
 
-    def _check_in_development(self, method: str) -> None:
-        """Refuse an assessment or valset call unless ``CMPND_CLI_DEV=1`` is set."""
-        if os.getenv("CMPND_CLI_DEV") != "1":
-            raise RuntimeError(
-                f"cmpnd.Client.{method}() is in development and not yet generally available. "
-                "Set CMPND_CLI_DEV=1 to enable it."
-            )
-        self._check_active()
-
     def _scoped_callbacks(self) -> list[Any]:
         """This client's callback, plus the caller's own non-cmpnd callbacks.
 
@@ -403,6 +390,19 @@ class Client:
 
         return execution.deployment_run_id(result_or_error)
 
+    def get_trace_id(self, result_or_error: Any) -> str | None:
+        """The root trace id of a ``run_deployment`` prediction or exception.
+
+        It reads the id serve returned with the settled run and sends no
+        request. ``get_trace`` opens the trace.
+        The id lives outside the prediction's output fields. A run that stored no
+        trace, a server that names none, and a failure before serve accepted a run
+        all return ``None``.
+        """
+        from . import execution
+
+        return execution.get_trace_id(result_or_error)
+
     def submit_deployment_run(
         self,
         deployment_or_id: Any,
@@ -918,7 +918,7 @@ class Client:
         e.g. ``annotations=[("output", "/trends/2", "the third trend"),
         ("span", "a1b2c3", "wrong field")]``.
         """
-        self._check_in_development("create_assessment")
+        self._check_active()
         _exporter_pool.flush(self._config)  # the referenced trace must be durable first
         return _assessments.create(
             trace_id,
@@ -937,7 +937,7 @@ class Client:
         Pass exactly one of ``program`` (all of that program's assessments, named by
         its name or ``p_`` id) or ``trace_id`` (one trace's assessments).
         """
-        self._check_in_development("list_assessments")
+        self._check_active()
         if program is not None and trace_id is not None:
             raise ValueError("pass exactly one of program or trace_id")
         page = _assessments.list(
@@ -952,24 +952,24 @@ class Client:
 
     def get_assessment(self, assessment_id: str) -> dict[str, Any]:
         """Fetch one assessment with its annotations."""
-        self._check_in_development("get_assessment")
+        self._check_active()
         return _assessments.get(assessment_id, api_key=self._config.api_key, endpoint=self._config.endpoint)
 
     def delete_assessment(self, assessment_id: str) -> None:
         """Delete one assessment. Raises if it does not exist."""
-        self._check_in_development("delete_assessment")
+        self._check_active()
         _assessments.delete(assessment_id, api_key=self._config.api_key, endpoint=self._config.endpoint)
 
     def create_valset(self, program: str, *, name: str | None = None) -> dict[str, Any]:
         """Create a valset owned by ``program``, a program name or ``p_`` id; name auto-assigned when omitted."""
-        self._check_in_development("create_valset")
+        self._check_active()
         return _valsets.create(
             self._program_id(program), name=name, api_key=self._config.api_key, endpoint=self._config.endpoint
         )
 
     def list_valsets(self, program: str, *, limit: int = 50, offset: int = 0) -> list[dict[str, Any]]:
         """The valsets of ``program``, a program name or ``p_`` id, newest first (paged by limit/offset)."""
-        self._check_in_development("list_valsets")
+        self._check_active()
         page = _valsets.list(
             self._program_id(program),
             limit=limit,
@@ -981,12 +981,12 @@ class Client:
 
     def get_valset(self, valset_id: str) -> dict[str, Any]:
         """Fetch one valset with its field shape and current version."""
-        self._check_in_development("get_valset")
+        self._check_active()
         return _valsets.get(valset_id, api_key=self._config.api_key, endpoint=self._config.endpoint)
 
     def list_valset_entries(self, valset_id: str, *, limit: int = 50, offset: int = 0) -> list[dict[str, Any]]:
         """A valset's active entries (paged by limit/offset)."""
-        self._check_in_development("list_valset_entries")
+        self._check_active()
         page = _valsets.list_entries(
             valset_id, limit=limit, offset=offset, api_key=self._config.api_key, endpoint=self._config.endpoint
         )
@@ -1002,7 +1002,7 @@ class Client:
     ) -> dict[str, Any]:
         """Add one entry. Pass exactly one of ``from_assessment`` (the server
         derives inputs from its trace) or ``inputs``."""
-        self._check_in_development("create_valset_entry")
+        self._check_active()
         return _valsets.create_entry(
             valset_id,
             expected_outputs=expected_outputs,
@@ -1014,7 +1014,7 @@ class Client:
 
     def delete_valset_entry(self, entry_id: str) -> dict[str, Any]:
         """Retire an entry (versioned, not a hard delete)."""
-        self._check_in_development("delete_valset_entry")
+        self._check_active()
         return _valsets.delete_entry(entry_id, api_key=self._config.api_key, endpoint=self._config.endpoint)
 
     def _run_gepa(

{cmpnd-0.70.0 → cmpnd-0.72.0}/cmpnd/execution.py

@@ -33,6 +33,21 @@ def deployment_run_id(result_or_error: Any) -> str | None:
     return getattr(result_or_error, "_cmpnd_deployment_run_id", None)
 
 
+def get_trace_id(result_or_error: Any) -> str | None:
+    """Return the root trace id serve named for a run's prediction or exception.
+
+    The id lives outside the prediction's program output fields. A run that
+    stored no trace, a server that names none, and a failure before serve
+    accepted a run all return ``None``.
+    """
+    return getattr(result_or_error, "_cmpnd_deployment_trace_id", None)
+
+
+def _settled_trace_id(settled: dict[str, Any]) -> str | None:
+    trace_id = settled.get("trace_id")
+    return trace_id if isinstance(trace_id, str) and trace_id else None
+
+
 def _with_run_id(error: Exception, run_id: str | None) -> Exception:
     error._cmpnd_deployment_run_id = run_id  # pyright: ignore[reportAttributeAccessIssue]
     return error
@@ -247,14 +262,20 @@ def run_deployment(
         raise
 
     settled_id: str = settled["deployment_run_id"]
+    trace_id = _settled_trace_id(settled)
     result = settled["result"]
     if settled["status"] == "failed":
-        _raise_stored_failure(settled_id, result, config.endpoint)
+        try:
+            _raise_stored_failure(settled_id, result, config.endpoint)
+        except Exception as error:
+            error._cmpnd_deployment_trace_id = trace_id  # pyright: ignore[reportAttributeAccessIssue]
+            raise
     if not isinstance(result, dict) or "output" not in result:
         raise _with_run_id(RuntimeError("Deployment run result has no program output"), settled_id)
     output = result["output"]
     prediction = Prediction(**output) if isinstance(output, dict) else Prediction(**{"output": output})
     setattr(prediction, "_cmpnd_deployment_run_id", settled_id)
+    setattr(prediction, "_cmpnd_deployment_trace_id", trace_id)
     return prediction
 
 

{cmpnd-0.70.0 → cmpnd-0.72.0}/docs/cli.md

@@ -228,6 +228,10 @@ seconds, unless `--timeout` sets the budget. The CLI prints the accepted run ID
 on stderr as soon as serve replies, which can take up to serve's five-second
 wait. A failed run exits nonzero. Human mode prints the stored error code,
 message and provider status. `--json` prints the whole settled run on stdout.
+When a settled run names its root trace, the CLI prints `Trace: <id>` on
+stderr, completed or failed. `cmpnd traces show <id>` opens that trace. The
+run's `trace_id` field holds the same ID, and serve leaves it out when the run
+stored no trace.
 The CLI retries status reads that time out or answer `408`, `429` or `5xx`
 within the budget.
 
@@ -418,10 +422,7 @@ down or neutral conclusion, an optional prose body and annotations. An
 annotation attaches a note to a span or to a piece of the output addressed by
 a JSON pointer.
 
-The group is in development. The CLI registers it only when
-`CMPND_CLI_DEV=1` is set, so it does not appear in `cmpnd --help`. The
-variable only hides the commands. Serve still checks each action against the
-`assessments:*` scopes.
+Serve checks each action against the `assessments:*` scopes.
 
 - `cmpnd assessments create <trace-id> [--up|--down|--neutral] [-m "prose"]
   [--annotate KIND REF NOTE …]` records one judgement.
@@ -534,9 +535,7 @@ that lookup, so it creates a valset by the program's name. On a server from befo
 name fails with an error that says to upgrade the server. Entries live in an
 `entries` subgroup, the same shape as `cmpnd datasets examples`.
 
-The group is in development. The CLI registers it only when
-`CMPND_CLI_DEV=1` is set. Serve still checks each action against the
-`valsets:*` scopes.
+Serve checks each action against the `valsets:*` scopes.
 
 - `cmpnd valsets create <program> [--name <name>]` creates a validation
   set for a program. The valset ID is its identity, and the name is a label

{cmpnd-0.70.0 → cmpnd-0.72.0}/docs/client.md

@@ -10,11 +10,6 @@ client methods too, and have no form outside a client. A client holds its own ta
 the global config that `auto_instrument()` reads, so both can run in one
 process.
 
-The client's assessment and valset calls are in development. Each one raises
-`RuntimeError` unless `CMPND_CLI_DEV=1` is set. The same variable reveals the
-in-development `cmpnd assessments` and `cmpnd valsets` CLI groups. Tracing
-through a client needs no flag.
-
 ## Entry point
 
 ```python
@@ -152,7 +147,9 @@ endpoint. None of these reads the global config:
 - `sdk.submit_deployment_run`, `list_deployment_runs`, `get_deployment_run`,
   `get_deployment_run_result` and `wait_deployment_run` submit, find and resume
   runs without waiting in one call. `sdk.deployment_run_id(result)` reads the run
-  id of a prediction or an exception.
+  id of a prediction or an exception. `sdk.get_trace_id(result)` reads the
+  id of the run's root trace the same way, or returns `None` when serve named no
+  trace.
 - `sdk.optimize_deployment(deployment, optimizer="gepa", ...)` runs the
   platform's optimization on a deployment and returns the best
   `DeployedProgram`, with `.score`. It never deploys. The program's model comes
@@ -218,12 +215,12 @@ methods:
   file the deployment stores, or the archive-relative file `path` names.
   `sdk.get_deployment_signature(deployment_id)` returns the selected
   artifact's typed metadata. It reads no worker status or executions.
-- Assessments, in development behind `CMPND_CLI_DEV=1`:
+- Assessments:
   `sdk.create_assessment(trace_id, conclusion=..., body=...,
   annotations=...)`, `sdk.list_assessments(program=... | trace_id=...,
   limit=..., offset=...)`, `sdk.get_assessment(assessment_id)` and
   `sdk.delete_assessment(assessment_id)`.
-- Valsets, in development behind `CMPND_CLI_DEV=1`:
+- Valsets:
   `sdk.create_valset(program, name=...)`, `sdk.list_valsets(program,
   limit=..., offset=...)`, `sdk.get_valset(valset_id)`,
   `sdk.list_valset_entries(valset_id, limit=..., offset=...)`,

{cmpnd-0.70.0 → cmpnd-0.72.0}/docs/sdk-instrumentation.md

@@ -432,7 +432,12 @@ command, and re-raises.
 `client.deployment_run_id(result)` reads the ID from
 private prediction metadata, so a program can use `output` or
 `deployment_run_id` as an output field. The accessor also accepts a stored
-execution error, a refused read, `RunWaitTimeout` and `RunResultExpired`. A
+execution error, a refused read, `RunWaitTimeout` and `RunResultExpired`.
+`client.get_trace_id(result)` reads the id of the run's root trace from
+the same metadata, for a completed run's prediction and a failed run's error.
+`cmpnd traces show` opens that trace. It returns `None` when the run stored no
+trace, when the server names none, and when the run failed before serve
+accepted it. A
 retained provider failure keeps its structured `RetryableProviderError` type but
 does not start another run. Trace delivery can finish after the result is
 available.
@@ -454,8 +459,10 @@ header readable.
 try:
     result = client.run_deployment(deployed, question="What changed?")
     run_id = client.deployment_run_id(result)
+    trace_id = client.get_trace_id(result)
 except Exception as error:
     run_id = client.deployment_run_id(error)
+    trace_id = client.get_trace_id(error)
     raise
 ```
 

{cmpnd-0.70.0 → cmpnd-0.72.0}/pyproject.toml

@@ -5,7 +5,7 @@ name = "cmpnd"
 # it's new) — no git tags. Bump with `uv version --bump {minor,patch}` or by
 # hand. Runtime `cmpnd.__version__` reads installed metadata, so it tracks
 # this automatically. See docs/plans/2026-07-02-sdk-pypi-tag-free-publish.md.
-version = "0.70.0"
+version = "0.72.0"
 description = "DSPy observability and deployment SDK for cmpnd"
 readme = "README.md"
 requires-python = ">=3.11"

{cmpnd-0.70.0 → cmpnd-0.72.0}/tests/test_cli/test_deployment_runs.py

@@ -537,3 +537,37 @@ def test_json_run_commands_do_not_import_dspy(argv: list[str], replies: list):
     assert "JSON run command imported dspy" not in result.stderr
     out = json.loads(result.stdout)
     assert RUN_ID in result.stdout or "yes" in json.dumps(out)
+
+
+TRACE_ID = "0b9c6f2e-4f1d-4c3a-9e2b-7a1d5c3e8f00"
+
+
+def test_failed_run_names_its_trace_on_stderr(monkeypatch: pytest.MonkeyPatch, capsys):
+    failure = {"error": {"type": "ProgramError", "message": "bad input"}, "execution_status": 500}
+
+    def answer(req):
+        return httpx.Response(200, json={**_envelope("failed", failure), "trace_id": TRACE_ID}, request=req)
+
+    with pytest.raises(SystemExit):
+        _run_cli(monkeypatch, ["run", DEPLOYMENT_ID, "{}"], answer)
+    assert f"Trace: {TRACE_ID}" in capsys.readouterr().err
+
+
+def test_completed_run_names_its_trace_on_stderr_and_keeps_stdout_the_output(monkeypatch: pytest.MonkeyPatch, capsys):
+    def answer(req):
+        return httpx.Response(
+            200, json={**_envelope("completed", {"output": {"answer": "yes"}}), "trace_id": TRACE_ID}, request=req
+        )
+
+    _run_cli(monkeypatch, ["run", DEPLOYMENT_ID, "{}", "--json"], answer)
+    out = capsys.readouterr()
+    assert json.loads(out.out) == {"answer": "yes"}
+    assert f"Trace: {TRACE_ID}" in out.err
+
+
+def test_run_without_a_trace_names_none(monkeypatch: pytest.MonkeyPatch, capsys):
+    def answer(req):
+        return httpx.Response(200, json=_envelope("completed", {"output": {"answer": "yes"}}), request=req)
+
+    _run_cli(monkeypatch, ["run", DEPLOYMENT_ID, "{}", "--json"], answer)
+    assert "Trace:" not in capsys.readouterr().err

{cmpnd-0.70.0 → cmpnd-0.72.0}/tests/test_cli/test_main.py

@@ -169,6 +169,28 @@ class TestDevGate:
         output = json.loads(captured.out)
         assert "families" in output
 
+    def test_assessments_works_without_flag(self, capsys):
+        env = {k: v for k, v in os.environ.items() if k != "CMPND_CLI_DEV"}
+        env["CMPND_API_KEY"] = "test-key"
+        mock_response = {"assessments": [], "total": 0}
+        with mock.patch("sys.argv", ["cmpnd", "assessments", "list", "--trace", "t_1"]):
+            with mock.patch.dict(os.environ, env, clear=True):
+                with mock.patch("cmpnd.cli.api_client.ApiClient.get", autospec=True, return_value=mock_response) as get:
+                    main()
+        assert get.call_args.args[1] == "/api/v1/assessments"
+        assert json.loads(capsys.readouterr().out) == mock_response
+
+    def test_valsets_works_without_flag(self, capsys):
+        env = {k: v for k, v in os.environ.items() if k != "CMPND_CLI_DEV"}
+        env["CMPND_API_KEY"] = "test-key"
+        mock_response = {"id": "v_1", "name": "valset-2026-11-02"}
+        with mock.patch("sys.argv", ["cmpnd", "valsets", "show", "v_1"]):
+            with mock.patch.dict(os.environ, env, clear=True):
+                with mock.patch("cmpnd.cli.api_client.ApiClient.get", autospec=True, return_value=mock_response) as get:
+                    main()
+        assert get.call_args.args[1] == "/api/v1/valsets/v_1"
+        assert json.loads(capsys.readouterr().out) == mock_response
+
 
 class TestAdminGate:
     """CMP-126: `admin` commands gated behind CMPND_CLI_ADMIN=1; the

{cmpnd-0.70.0 → cmpnd-0.72.0}/tests/test_cli/test_parity.py

@@ -16,7 +16,6 @@ from __future__ import annotations
 
 import argparse
 import itertools
-import os
 import re
 from pathlib import Path
 from unittest import mock
@@ -26,6 +25,9 @@ from cmpnd.cli.api_client import ApiClient
 from cmpnd.cli.commands import (
     admin as _admin,
 )
+from cmpnd.cli.commands import (
+    assessments as _assessments,
+)
 from cmpnd.cli.commands import (
     auth as _auth,
 )
@@ -59,6 +61,9 @@ from cmpnd.cli.commands import (
 from cmpnd.cli.commands import (
     traces as _traces,
 )
+from cmpnd.cli.commands import (
+    valsets as _valsets,
+)
 
 #: Every module registering a resource-tier noun group. `login` is excluded
 #: deliberately: `login` / `logout` are task-tier top-level commands with no
@@ -75,6 +80,8 @@ _RESOURCE_MODULES = (
     _evals,
     _datasets,
     _deployments,
+    _assessments,
+    _valsets,
 )
 
 
@@ -281,10 +288,8 @@ class TestOptimizeBinding:
     partial revival (CLI but not REPL) is the failure worth catching."""
 
     def test_optimize_is_bound_on_both_surfaces(self):
-        env = {k: v for k, v in os.environ.items() if k != "CMPND_CLI_DEV"}
-        with mock.patch.dict(os.environ, env, clear=True):
-            assert "optimize" in _argparse_subcommands()
-            assert "optimize" in _shell_namespace()
+        assert "optimize" in _argparse_subcommands()
+        assert "optimize" in _shell_namespace()
 
 
 class TestResourceVocabulary:

{cmpnd-0.70.0 → cmpnd-0.72.0}/tests/test_client.py

@@ -327,11 +327,11 @@ class TestMisuseGuard:
             sdk5.run_program(a_program(), question="What is the capital of France?")
 
 
-class TestDevelopmentGate:
-    """The client's tracing needs no flag. Its assessment and valset calls are in
-    development, gated behind CMPND_CLI_DEV=1 as their CLI groups are."""
+class TestAssessmentsAndValsetsNeedNoFlag:
+    """The client's assessment and valset calls need no environment flag. A closed
+    client still refuses each one, as it refuses every other call."""
 
-    GATED = {
+    CALLS = {
         "create_assessment": ("t_1",),
         "list_assessments": ("p_0123456789Ab",),
         "get_assessment": ("a_1",),
@@ -344,23 +344,19 @@ class TestDevelopmentGate:
         "delete_valset_entry": ("e_1",),
     }
 
-    def test_construction_needs_no_flag(self, monkeypatch: pytest.MonkeyPatch):
-        monkeypatch.delenv("CMPND_CLI_DEV", raising=False)
-        sdk = cmpnd.Client(project_tags=["context-Z"], api_key="k", endpoint="http://x")
-        assert sdk._active
-
-    @pytest.mark.parametrize("method", sorted(GATED))
-    def test_an_assessment_or_valset_call_refuses_without_the_flag(self, method, monkeypatch: pytest.MonkeyPatch):
+    @pytest.mark.parametrize("method", sorted(CALLS))
+    def test_a_closed_client_refuses_an_assessment_or_valset_call(self, method, monkeypatch: pytest.MonkeyPatch):
         monkeypatch.delenv("CMPND_CLI_DEV", raising=False)
         sdk = cmpnd.Client(api_key="k", endpoint="http://x")
+        sdk.close()
         kwargs = {"expected_outputs": {}} if method == "create_valset_entry" else {}
-        with pytest.raises(RuntimeError, match=r"CMPND_CLI_DEV=1"):
-            getattr(sdk, method)(*self.GATED[method], **kwargs)
+        with pytest.raises(RuntimeError, match=r"client is closed"):
+            getattr(sdk, method)(*self.CALLS[method], **kwargs)
 
-    def test_every_assessment_and_valset_method_is_gated(self):
+    def test_every_assessment_and_valset_method_is_listed(self):
         public = {name for name in dir(cmpnd.Client) if not name.startswith("_")}
         data = {name for name in public if "assessment" in name or "valset" in name}
-        assert data == set(self.GATED)
+        assert data == set(self.CALLS)
 
 
 # --- Case 7: exceptions and cleanup --------------------------------------
@@ -794,7 +790,7 @@ class TestClientDataSurface:
     client's own credentials, list_* unwrap to lists, and client.traces warns."""
 
     def _client(self, monkeypatch):
-        monkeypatch.setenv("CMPND_CLI_DEV", "1")
+        monkeypatch.delenv("CMPND_CLI_DEV", raising=False)
         monkeypatch.setattr(cmpnd.client._exporter_pool, "flush", lambda cfg: None)
         return cmpnd.Client(api_key="k", endpoint="http://x")
 
@@ -811,6 +807,21 @@ class TestClientDataSurface:
         assert seen["trace_id"] == "t-1" and seen["conclusion"] == "down"
         assert seen["api_key"] == "k" and seen["endpoint"] == "http://x"
 
+    @pytest.mark.parametrize(
+        ("method", "module", "function", "ref"),
+        [
+            ("get_assessment", "_assessments", "get", "as_1"),
+            ("get_valset", "_valsets", "get", "v_1"),
+            ("delete_valset_entry", "_valsets", "delete_entry", "e_1"),
+        ],
+    )
+    def test_a_one_id_call_delegates_with_client_creds(self, method, module, function, ref, monkeypatch):
+        fake = mock.create_autospec(getattr(getattr(cmpnd, module), function), return_value={"id": ref})
+        monkeypatch.setattr(getattr(cmpnd, module), function, fake)
+
+        assert getattr(self._client(monkeypatch), method)(ref) == {"id": ref}
+        fake.assert_called_once_with(ref, api_key="k", endpoint="http://x")
+
     def test_get_deployment_signature_delegates_with_client_creds(self, monkeypatch):
         seen: dict = {}
         answer = object()

{cmpnd-0.70.0 → cmpnd-0.72.0}/tests/test_client_program_names.py

@@ -126,7 +126,6 @@ def _provider():
 
 @pytest.fixture(autouse=True)
 def _clean_slate(monkeypatch: pytest.MonkeyPatch):
-    monkeypatch.delenv("CMPND_CLI_DEV", raising=False)  # a client traces with no dev flag
     monkeypatch.delenv("CMPND_PROGRAM_ID", raising=False)
     configuration._config = None
     saved_callbacks = list(dspy.settings.callbacks or [])
@@ -328,7 +327,6 @@ def minting_server(monkeypatch: pytest.MonkeyPatch):
     def _new_client(**kw: Any) -> httpx.Client:
         return real_new_client(transport=httpx.MockTransport(srv.handler), **kw)
 
-    monkeypatch.setenv("CMPND_CLI_DEV", "1")
     with (
         mock.patch("cmpnd._http.new_client", autospec=True, side_effect=_new_client),
         mock.patch("cmpnd.exporter_http.BatchExporter", autospec=True, side_effect=_factory),

{cmpnd-0.70.0 → cmpnd-0.72.0}/tests/test_deployment_runs.py

@@ -505,3 +505,71 @@ def test_failure_summary_names_code_message_and_provider_status():
         f"Deployment run {RUN_ID} failed: EgressFailed: provider failed (provider status 429)"
     )
     assert deployment_runs.failure_summary(_run("failed", {"error": "boom"})) == f"Deployment run {RUN_ID} failed: boom"
+
+
+TRACE_ID = "0b9c6f2e-4f1d-4c3a-9e2b-7a1d5c3e8f00"
+
+
+def test_a_completed_run_names_its_trace_beside_its_run_id(monkeypatch: pytest.MonkeyPatch):
+    def answer(req: httpx.Request) -> httpx.Response:
+        body = {**_run("completed", {"output": {"answer": "yes"}}, available=True), "trace_id": TRACE_ID}
+        return httpx.Response(200, json=body, request=req)
+
+    _transport(monkeypatch, answer)
+    prediction = _client().run_deployment(DEPLOYMENT_ID, question="hi")
+    assert prediction.answer == "yes"
+    assert _client().get_trace_id(prediction) == TRACE_ID
+    assert _client().deployment_run_id(prediction) == RUN_ID
+    assert "trace_id" not in prediction.keys()
+
+
+def test_a_failed_run_names_its_trace_on_the_raised_error(monkeypatch: pytest.MonkeyPatch):
+    failure = {"error": {"type": "ProgramError", "message": "bad input"}, "execution_status": 500}
+
+    def answer(req: httpx.Request) -> httpx.Response:
+        return httpx.Response(200, json={**_run("failed", failure, available=True), "trace_id": TRACE_ID}, request=req)
+
+    _transport(monkeypatch, answer)
+    with pytest.raises(httpx.HTTPStatusError) as caught:
+        _client().run_deployment(DEPLOYMENT_ID)
+    assert _client().get_trace_id(caught.value) == TRACE_ID
+    assert _client().deployment_run_id(caught.value) == RUN_ID
+
+
+def test_a_retryable_provider_failure_names_its_trace(monkeypatch: pytest.MonkeyPatch):
+    failure = {
+        "error": {"type": "EgressFailed", "message": "provider refused"},
+        "execution_status": 503,
+        "retryable": True,
+        "failure": {"domain": "provider", "upstream_status": 429, "attempts": 3},
+    }
+
+    def answer(req: httpx.Request) -> httpx.Response:
+        return httpx.Response(200, json={**_run("failed", failure, available=True), "trace_id": TRACE_ID}, request=req)
+
+    _transport(monkeypatch, answer)
+    with pytest.raises(cmpnd.RetryableProviderError) as caught:
+        _client().run_deployment(DEPLOYMENT_ID)
+    assert _client().get_trace_id(caught.value) == TRACE_ID
+
+
+@pytest.mark.parametrize("trace_field", [{}, {"trace_id": None}, {"trace_id": ""}, {"trace_id": 7}])
+def test_a_run_serve_names_no_trace_for_has_none(monkeypatch: pytest.MonkeyPatch, trace_field: dict):
+    def answer(req: httpx.Request) -> httpx.Response:
+        body = {**_run("completed", {"output": {"answer": "yes"}}, available=True), **trace_field}
+        return httpx.Response(200, json=body, request=req)
+
+    _transport(monkeypatch, answer)
+    prediction = _client().run_deployment(DEPLOYMENT_ID, question="hi")
+    assert _client().get_trace_id(prediction) is None
+
+
+def test_a_failure_before_acceptance_has_no_trace(monkeypatch: pytest.MonkeyPatch):
+    def answer(req: httpx.Request) -> httpx.Response:
+        return httpx.Response(404, json={"error": "not found"}, request=req)
+
+    _transport(monkeypatch, answer)
+    with pytest.raises(httpx.HTTPStatusError) as caught:
+        _client().run_deployment(DEPLOYMENT_ID)
+    assert _client().get_trace_id(caught.value) is None
+    assert _client().get_trace_id(object()) is None

{cmpnd-0.70.0 → cmpnd-0.72.0}/tests/test_program_id_deprecation.py

@@ -44,7 +44,6 @@ def _clean_configuration(monkeypatch):
     monkeypatch.delenv("CMPND_PROGRAM_ID", raising=False)
     monkeypatch.delenv("CMPND_API_KEY", raising=False)
     monkeypatch.delenv("CMPND_ENDPOINT", raising=False)
-    monkeypatch.delenv("CMPND_CLI_DEV", raising=False)  # a client traces with no dev flag
     dspy.settings.configure(lm=None)
     yield
     configuration._config = None

{cmpnd-0.70.0 → cmpnd-0.72.0}/uv.lock

@@ -505,7 +505,7 @@ wheels = [
 
 [[package]]
 name = "cmpnd"
-version = "0.70.0"
+version = "0.72.0"
 source = { editable = "." }
 default-groups = ["dspy-3-4-1"]
 dependencies = [