fix experiment dataset linking

src/main/java/dev/braintrust/api/BraintrustApiClient.java

@@ -572,7 +572,9 @@ public Experiment getOrCreateExperiment(CreateExperimentRequest request) {
                             request.tags().orElse(List.of()),
                             request.metadata().orElse(Map.of()),
                             "notused",
-                            "notused");
+                            "notused",
+                            request.datasetId(),
+                            request.datasetVersion());
             experiments.add(newExperiment);
             return newExperiment;
         }
@@ -734,7 +736,9 @@ record CreateExperimentRequest(
             Optional<String> description,
             Optional<String> baseExperimentId,
             Optional<List<String>> tags,
-            Optional<Map<String, Object>> metadata) {
+            Optional<Map<String, Object>> metadata,
+            Optional<String> datasetId,
+            Optional<String> datasetVersion) {
 
         public CreateExperimentRequest(String projectId, String name) {
             this(
@@ -743,6 +747,8 @@ public CreateExperimentRequest(String projectId, String name) {
                     Optional.empty(),
                     Optional.empty(),
                     Optional.empty(),
+                    Optional.empty(),
+                    Optional.empty(),
                     Optional.empty());
         }
     }
@@ -755,7 +761,32 @@ record Experiment(
             List<String> tags,
             Map<String, Object> metadata,
             String createdAt,
-            String updatedAt) {}
+            String updatedAt,
+            Optional<String> datasetId,
+            Optional<String> datasetVersion) {
+        /** Convenience constructor */
+        public Experiment(
+                String id,
+                String projectId,
+                String name,
+                Optional<String> description,
+                List<String> tags,
+                Map<String, Object> metadata,
+                String createdAt,
+                String updatedAt) {
+            this(
+                    id,
+                    projectId,
+                    name,
+                    description,
+                    tags,
+                    metadata,
+                    createdAt,
+                    updatedAt,
+                    Optional.empty(),
+                    Optional.empty());
+        }
+    }
 
     record CreateDatasetRequest(String projectId, String name, Optional<String> description) {
         public CreateDatasetRequest(String projectId, String name) {

src/main/java/dev/braintrust/eval/Dataset.java

@@ -25,11 +25,7 @@ public interface Dataset<INPUT, OUTPUT> {
     /** Convenience method to safely iterate all items in a dataset. */
     default void forEach(Consumer<DatasetCase<INPUT, OUTPUT>> consumer) {
         try (var cursor = openCursor()) {
-            Optional<DatasetCase<INPUT, OUTPUT>> cursorCase = cursor.next();
-            while (cursorCase.isPresent()) {
-                consumer.accept(cursorCase.get());
-                cursorCase = cursor.next();
-            }
+            cursor.forEach(consumer);
         }
     }
 
@@ -47,6 +43,17 @@ interface Cursor<CASE> extends AutoCloseable {
 
         /** close all cursor resources */
         void close();
+
+        /** version of the dataset this cursor was opened against */
+        Optional<String> version();
+
+        default void forEach(Consumer<CASE> caseConsumer) {
+            Optional<CASE> c = next();
+            while (c.isPresent()) {
+                caseConsumer.accept(c.get());
+                c = next();
+            }
+        }
     }
 
     /** Create an in-memory Dataset containing the provided cases. */

src/main/java/dev/braintrust/eval/DatasetBrainstoreImpl.java

@@ -2,7 +2,9 @@
 
 import dev.braintrust.api.BraintrustApiClient;
 import java.util.*;
+import javax.annotation.Nonnull;
 import javax.annotation.Nullable;
+import org.jspecify.annotations.NonNull;
 
 /** A dataset loaded externally from Braintrust using paginated API fetches */
 public class DatasetBrainstoreImpl<INPUT, OUTPUT> implements Dataset<INPUT, OUTPUT> {
@@ -39,7 +41,25 @@ public Optional<String> version() {
 
     @Override
     public Cursor<DatasetCase<INPUT, OUTPUT>> openCursor() {
-        return new BrainstoreCursor();
+        return new BrainstoreCursor(null == pinnedVersion ? fetchMaxVersion() : pinnedVersion);
+    }
+
+    private String fetchMaxVersion() {
+        var response =
+                apiClient.btqlQuery(
+                        "SELECT max(_xact_id) as version FROM dataset('%s')".formatted(datasetId));
+        if (response.data().isEmpty()) {
+            throw new RuntimeException(
+                    "Failed to fetch max version for dataset: " + datasetId + " (empty response)");
+        }
+        var version = response.data().get(0).get("version");
+        if (version == null) {
+            throw new RuntimeException(
+                    "Failed to fetch max version for dataset: "
+                            + datasetId
+                            + " (null version — dataset may be empty)");
+        }
+        return String.valueOf(version);
     }
 
     private class BrainstoreCursor implements Cursor<DatasetCase<INPUT, OUTPUT>> {
@@ -48,13 +68,15 @@ private class BrainstoreCursor implements Cursor<DatasetCase<INPUT, OUTPUT>> {
         private @Nullable String cursor;
         private boolean exhausted;
         private boolean closed;
+        private final @Nonnull String cursorVersion;
 
-        BrainstoreCursor() {
+        BrainstoreCursor(@NonNull String cursorVersion) {
             this.currentBatch = new ArrayList<>();
             this.currentIndex = 0;
             this.cursor = null;
             this.exhausted = false;
             this.closed = false;
+            this.cursorVersion = cursorVersion;
         }
 
         @Override
@@ -111,7 +133,7 @@ public Optional<DatasetCase<INPUT, OUTPUT>> next() {
 
         private void fetchNextBatch() {
             var request =
-                    new BraintrustApiClient.DatasetFetchRequest(batchSize, cursor, pinnedVersion);
+                    new BraintrustApiClient.DatasetFetchRequest(batchSize, cursor, cursorVersion);
             var response = apiClient.fetchDatasetEvents(datasetId, request);
 
             currentBatch = new ArrayList<>(response.events());
@@ -129,5 +151,10 @@ public void close() {
             closed = true;
             currentBatch.clear();
         }
+
+        @Override
+        public Optional<String> version() {
+            return Optional.of(cursorVersion);
+        }
     }
 }

src/main/java/dev/braintrust/eval/DatasetInMemoryImpl.java

@@ -44,6 +44,11 @@ public Optional<DatasetCase<INPUT, OUTPUT>> next() {
             public void close() {
                 closed = true;
             }
+
+            @Override
+            public Optional<String> version() {
+                return Optional.empty();
+            }
         };
     }
 }

src/main/java/dev/braintrust/eval/Eval.java

@@ -60,16 +60,28 @@ private Eval(Builder<INPUT, OUTPUT> builder) {
 
     /** Runs the evaluation and returns results. */
     public EvalResult run() {
-        var experiment =
-                client.getOrCreateExperiment(
-                        new BraintrustApiClient.CreateExperimentRequest(
-                                orgAndProject.project().id(),
-                                experimentName,
-                                Optional.empty(),
-                                Optional.empty(),
-                                tags.isEmpty() ? Optional.empty() : Optional.of(tags),
-                                metadata.isEmpty() ? Optional.empty() : Optional.of(metadata)));
-        dataset.forEach(datasetCase -> evalOne(experiment.id(), datasetCase));
+        try (var cursor = dataset.openCursor()) {
+            Optional<String> datasetVersion = Optional.empty();
+            Optional<String> datasetId = Optional.empty();
+            if (dataset instanceof DatasetBrainstoreImpl<INPUT, OUTPUT>) {
+                // TODO: come up with a means of expressing this naturally in the dataset api
+                datasetVersion = cursor.version();
+                datasetId = Optional.of(dataset.id());
+            }
+            var experiment =
+                    client.getOrCreateExperiment(
+                            new BraintrustApiClient.CreateExperimentRequest(
+                                    orgAndProject.project().id(),
+                                    experimentName,
+                                    Optional.empty(),
+                                    Optional.empty(),
+                                    tags.isEmpty() ? Optional.empty() : Optional.of(tags),
+                                    metadata.isEmpty() ? Optional.empty() : Optional.of(metadata),
+                                    datasetId,
+                                    datasetVersion));
+            cursor.forEach(datasetCase -> evalOne(experiment.id(), datasetCase));
+        }
+
         var experimentUrl =
                 "%s/experiments/%s"
                         .formatted(

src/test/java/dev/braintrust/eval/EvalTest.java

@@ -336,4 +336,48 @@ void evalWithExperimentTagsAndMetadata() {
         assertEquals(expectedTags, experiment.tags(), "Experiment should have tags");
         assertEquals(expectedMetadata, experiment.metadata(), "Experiment should have metadata");
     }
+
+    @Test
+    @SneakyThrows
+    void evalLinksToRemoteDataset() {
+        if (TestHarness.getVcrMode() == VCR.VcrMode.REPLAY) {
+            return;
+        }
+
+        var experimentName = "test-dataset-linking";
+        Dataset<String, String> dataset = testHarness.braintrust().fetchDataset("food");
+
+        var eval =
+                testHarness
+                        .braintrust()
+                        .<String, String>evalBuilder()
+                        .name(experimentName)
+                        .dataset(dataset)
+                        .taskFunction(String::toUpperCase)
+                        .scorers(
+                                Scorer.of(
+                                        "exact",
+                                        (expected, result) -> expected.equals(result) ? 1.0 : 0.0))
+                        .build();
+        eval.run();
+        testHarness.awaitExportedSpans();
+
+        // Verify the experiment is linked to the dataset
+        var experiments =
+                testHarness
+                        .braintrust()
+                        .apiClient()
+                        .listExperiments(TestHarness.defaultProjectId());
+        var experiment =
+                experiments.stream()
+                        .filter(e -> e.name().equals(experimentName))
+                        .findFirst()
+                        .orElseThrow(() -> new AssertionError("Experiment not found"));
+
+        assertTrue(experiment.datasetId().isPresent(), "Experiment should be linked to a dataset");
+        assertEquals(dataset.id(), experiment.datasetId().get(), "Dataset ID should match");
+        assertTrue(
+                experiment.datasetVersion().isPresent(),
+                "Experiment should have a dataset version");
+    }
 }

src/test/resources/cassettes/anthropic/__files/v1_messages-71f8c84f-b988-4c20-ab41-8e0f5e20325c.txt → src/test/resources/cassettes/anthropic/__files/v1_messages-14d7ddaf-c4dc-4e41-ae83-1b9b0e7e222c.txt

@@ -1,24 +1,24 @@
 event: message_start
-data: {"type":"message_start","message":{"model":"claude-3-5-haiku-20241022","id":"msg_01VCp1RHSod6yjMZgMhvo63L","type":"message","role":"assistant","content":[],"stop_reason":null,"stop_sequence":null,"usage":{"input_tokens":19,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"cache_creation":{"ephemeral_5m_input_tokens":0,"ephemeral_1h_input_tokens":0},"output_tokens":5,"service_tier":"standard","inference_geo":"not_available"}}  }
+data: {"type":"message_start","message":{"model":"claude-3-5-haiku-20241022","id":"msg_01NhbxNFB9uMvk5gvd7qG2Ve","type":"message","role":"assistant","content":[],"stop_reason":null,"stop_sequence":null,"usage":{"input_tokens":19,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"cache_creation":{"ephemeral_5m_input_tokens":0,"ephemeral_1h_input_tokens":0},"output_tokens":5,"service_tier":"standard","inference_geo":"not_available"}}         }
 
 event: content_block_start
-data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}           }
+data: {"type":"content_block_start","index":0,"content_block":{"type":"text","text":""}               }
 
 event: ping
 data: {"type": "ping"}
 
 event: content_block_delta
-data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"The capital of France is"}  }
+data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":"The capital of France is"}}
 
 event: content_block_delta
-data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":" Paris."}   }
+data: {"type":"content_block_delta","index":0,"delta":{"type":"text_delta","text":" Paris."}      }
 
 event: content_block_stop
-data: {"type":"content_block_stop","index":0}
+data: {"type":"content_block_stop","index":0    }
 
 event: message_delta
-data: {"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"input_tokens":19,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":10}               }
+data: {"type":"message_delta","delta":{"stop_reason":"end_turn","stop_sequence":null},"usage":{"input_tokens":19,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"output_tokens":10}           }
 
 event: message_stop
-data: {"type":"message_stop"            }
+data: {"type":"message_stop" }
 

src/test/resources/cassettes/anthropic/__files/v1_messages-850b9b3b-255f-4688-8cbc-b975114bb35d.json → src/test/resources/cassettes/anthropic/__files/v1_messages-c848dbe0-8923-4e08-bb1e-892a6fb643f3.json

@@ -1 +1 @@
-{"model":"claude-3-5-haiku-20241022","id":"msg_01QZ2Kog1CjtsKvGZnAbYYSQ","type":"message","role":"assistant","content":[{"type":"text","text":"The capital of France is Paris."}],"stop_reason":"end_turn","stop_sequence":null,"usage":{"input_tokens":19,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"cache_creation":{"ephemeral_5m_input_tokens":0,"ephemeral_1h_input_tokens":0},"output_tokens":10,"service_tier":"standard","inference_geo":"not_available"}}
+{"model":"claude-3-5-haiku-20241022","id":"msg_01SdXWAvhUNSc6wMD8ii2Ma7","type":"message","role":"assistant","content":[{"type":"text","text":"The capital of France is Paris."}],"stop_reason":"end_turn","stop_sequence":null,"usage":{"input_tokens":19,"cache_creation_input_tokens":0,"cache_read_input_tokens":0,"cache_creation":{"ephemeral_5m_input_tokens":0,"ephemeral_1h_input_tokens":0},"output_tokens":10,"service_tier":"standard","inference_geo":"not_available"}}

src/test/resources/cassettes/anthropic/mappings/v1_messages-71f8c84f-b988-4c20-ab41-8e0f5e20325c.json → src/test/resources/cassettes/anthropic/mappings/v1_messages-14d7ddaf-c4dc-4e41-ae83-1b9b0e7e222c.json

@@ -1,5 +1,5 @@
 {
-  "id" : "71f8c84f-b988-4c20-ab41-8e0f5e20325c",
+  "id" : "14d7ddaf-c4dc-4e41-ae83-1b9b0e7e222c",
   "name" : "v1_messages",
   "request" : {
     "url" : "/v1/messages",
@@ -17,35 +17,35 @@
   },
   "response" : {
     "status" : 200,
-    "bodyFileName" : "v1_messages-71f8c84f-b988-4c20-ab41-8e0f5e20325c.txt",
+    "bodyFileName" : "v1_messages-14d7ddaf-c4dc-4e41-ae83-1b9b0e7e222c.txt",
     "headers" : {
-      "Date" : "Fri, 06 Feb 2026 01:36:11 GMT",
+      "Date" : "Wed, 18 Feb 2026 04:40:37 GMT",
       "Content-Type" : "text/event-stream; charset=utf-8",
       "Cache-Control" : "no-cache",
       "anthropic-ratelimit-requests-limit" : "10000",
       "anthropic-ratelimit-requests-remaining" : "9999",
-      "anthropic-ratelimit-requests-reset" : "2026-02-06T01:36:11Z",
+      "anthropic-ratelimit-requests-reset" : "2026-02-18T04:40:36Z",
       "anthropic-ratelimit-input-tokens-limit" : "5000000",
       "anthropic-ratelimit-input-tokens-remaining" : "5000000",
-      "anthropic-ratelimit-input-tokens-reset" : "2026-02-06T01:36:11Z",
+      "anthropic-ratelimit-input-tokens-reset" : "2026-02-18T04:40:36Z",
       "anthropic-ratelimit-output-tokens-limit" : "1000000",
       "anthropic-ratelimit-output-tokens-remaining" : "1000000",
-      "anthropic-ratelimit-output-tokens-reset" : "2026-02-06T01:36:11Z",
+      "anthropic-ratelimit-output-tokens-reset" : "2026-02-18T04:40:36Z",
       "anthropic-ratelimit-tokens-limit" : "6000000",
       "anthropic-ratelimit-tokens-remaining" : "6000000",
-      "anthropic-ratelimit-tokens-reset" : "2026-02-06T01:36:11Z",
-      "request-id" : "req_011CXqxWWdE6Kb91K1dfu7Zt",
+      "anthropic-ratelimit-tokens-reset" : "2026-02-18T04:40:36Z",
+      "request-id" : "req_011CYEvHDdWoj2q5sbR7CzQ9",
       "strict-transport-security" : "max-age=31536000; includeSubDomains; preload",
       "anthropic-organization-id" : "27796668-7351-40ac-acc4-024aee8995a5",
       "Server" : "cloudflare",
-      "x-envoy-upstream-service-time" : "388",
-      "cf-cache-status" : "DYNAMIC",
+      "x-envoy-upstream-service-time" : "414",
       "X-Robots-Tag" : "none",
       "Content-Security-Policy" : "default-src 'none'; frame-ancestors 'none'",
-      "CF-RAY" : "9c96ea441d1eba4e-SEA"
+      "cf-cache-status" : "DYNAMIC",
+      "CF-RAY" : "9cfad8ee5a4ab9f4-SEA"
     }
   },
-  "uuid" : "71f8c84f-b988-4c20-ab41-8e0f5e20325c",
+  "uuid" : "14d7ddaf-c4dc-4e41-ae83-1b9b0e7e222c",
   "persistent" : true,
   "insertionIndex" : 2
 }