fix CI API

EVALUATIONS.md

@@ -846,19 +846,19 @@ These four metrics also appear per-model (as `average_bleu_score`/`average_rouge
 
 ### CI gate (self-host)
 
-A finalized run can pass/fail a CI job. `report.gate(...)` checks the run's average rating, prints per-check verdicts into the CI log, and returns a `GateResult` - the caller decides the exit code:
+A finalized run can pass/fail a CI job. `run.gate(...)` (on the run context `.execute()` returns - not on the `Report` model from `get_report()`) checks the run's average rating, prints per-check verdicts into the CI log, and returns a `GateResult` - the caller decides the exit code:
 
 ```python
 import sys
 
-report = (
+run = (
     client.evaluations
     .run(dataset_id="evds_…", subject={"kind": "custom_agent", "framework": "raw_python"})
     .execute(my_agent)   # in CI, this is the PR's version of your agent
     .finalize()
 )
 
-gate = report.gate(fail_under=7, no_regression=True, caller="github-actions")
+gate = run.gate(fail_under=7, no_regression=True, caller="github-actions")
 sys.exit(gate.exit_code)   # 0 = merge, 1 = block
 ```
 

README.md

@@ -129,7 +129,7 @@ client.evaluations.run(
 
 See [Prompt registry](EVALUATIONS.md#prompt-registry) in the full guide, or [self-host's docs](https://docs.agentx.so/improve/prompt-management) for the "Suggest improvement" dashboard flow (self-host only - no hosted-SaaS equivalent yet).
 
-On self-host, a finalized run can also **gate a CI job**: `report.gate(fail_under=7, no_regression=True)` checks the run's average rating against an absolute floor and/or the dataset's previous run, prints per-check verdicts into the CI log, and returns an exit code - `sys.exit(gate.exit_code)` blocks the merge on regression. Recorded gates appear in the dashboard's CI Gates tab. See [self-host's CI docs](https://docs.agentx.so/integrations/self-host-ci) for the GitHub Actions recipe.
+On self-host, a finalized run can also **gate a CI job**: `run.gate(fail_under=7, no_regression=True)` (on the run context `.execute()` returns) checks the run's average rating against an absolute floor and/or the dataset's previous run, prints per-check verdicts into the CI log, and returns an exit code - `sys.exit(gate.exit_code)` blocks the merge on regression. Recorded gates appear in the dashboard's CI Gates tab. See [self-host's CI docs](https://docs.agentx.so/integrations/self-host-ci) for the GitHub Actions recipe.
 
 See **[EVALUATIONS.md](EVALUATIONS.md)** for the full guide - dataset builder, framework adapters, similarity metrics, smoke testing, judge configuration, prompt registry, and the complete API reference.
 

agentx/evaluations/client.py

@@ -37,6 +37,9 @@
 # wait out the whole job on one connection. Matches EvaluationRunContext.analyze()'s own
 # default timeout.
 _SELF_HOST_ANALYZE_TIMEOUT = 1800
+# Batch result submission scores each result synchronously inside the request (one judge call
+# per result on the sync path) - a big batch on a slow judge legitimately takes minutes.
+_SELF_HOST_SCORING_TIMEOUT = 900
 
 
 class AgentXEvaluationsError(Exception):
@@ -344,7 +347,15 @@ def append_results(
             "batchId": batch_id,
             "results": [_result_to_payload(r) for r in results],
         }
-        data = self._request("POST", f"/runs/{run_id}/results", json=payload)
+        # Scoring is synchronous inside this request (a judge call per result - the runner's
+        # spinner says "~60s+" for a reason), so the default 30s timeout + silent backoff loop
+        # re-POSTed the batch WHILE the first submission was still scoring, and the two raced
+        # into the engine's (run_id, idempotency_key) unique constraint. Long timeout, no
+        # transport retry - the runner's own logged retry-once is the retry layer here.
+        data = self._request(
+            "POST", f"/runs/{run_id}/results", json=payload,
+            timeout=_SELF_HOST_SCORING_TIMEOUT, retry=False,
+        )
         return BatchAppendResponse(**data)
 
     def finalize_run(self, run_id: str) -> Dict[str, Any]: