Datasets:

AfterQuery
/

CrossSearch

+---
+language:
+- en
+pretty_name: CrossSearch (100-task paper set)
+configs:
+- config_name: tasks
+  data_files: tasks.jsonl.gz
+- config_name: rollouts
+  data_files: rollouts-*.jsonl.gz
+---
+# CrossSearch: 100-task paper set
+This resource contains 100 tasks and 4,269 selected, completed evaluation runs: 3,000 native-search runs (10 model configurations, three final attempts per task) and 1,269 search-backend ablation runs (three models, 47 shared tasks, three alternative backends, three attempts). The 600 Opus 5.5/GPT-6.1 Sol runs are included in the 3,000.
+## Files
+- tasks.jsonl.gz: expert questions, frozen rubrics and evidence maps.
+- tasks.tar.gz: starting workspace assets, separated from grader-only reference material.
+- rollouts-*.jsonl.gz: saved agent trace files, API request/response captures, final answers, original rewards and criterion judgments. The 669 encrypted source runs are exported as readable derivatives. Source archives remain unchanged; no private key is distributed.
+- grading-protocol.md: scoring protocol and judge-version uncertainty.
+- judge-model-audit.json: recorded judge fields for the 415 encrypted extension records.
+- analysis-summary.json: frozen paper/blog results.
+- release-manifest.json: final counts, shard checksums, redaction totals and verification status. Check this manifest for completeness.
+## Reading the data
+    from datasets import load_dataset
+    tasks = load_dataset('AfterQuery/CrossSearch', 'tasks', split='train')
+    runs = load_dataset('AfterQuery/CrossSearch', 'rollouts', split='train', streaming=True)
+Each rollout retains the original archive hash and source-member hashes. Operational credentials and local author/account paths are removed from publication derivatives. Available reasoning text is retained; opaque provider reasoning cannot be converted to visible thoughts. Per-run trace files preserve evaluated prompt wrappers. Native and ablation conditions remain separate; valid low and zero rewards are included. Historical failed attempts are not relabeled or retried.
+## Publisher and rights
+AfterQuery is the visible dataset owner; this URL is not an anonymous publisher link. Third-party evidence retains its source attribution and rights information. No blanket new license is asserted for third-party documents.
+## Judge identity
+All 415 audited encrypted extension records save google/gemini-3.7-flash as requested/model/resolved identifiers and lack a separate provider-version field. Google's October 8 routing change means these labels alone do not establish which underlying Gemini version served the judge. See grading-protocol.md.

collection-protocol-check.json ADDED Viewed

	@@ -0,0 +1 @@

+ [{"claim":"At least one of four pilots must fail","confirmed_in_draft":true,"where":"Section3.3 p5 lines221-235; Figure2 p4","qualification":"A completed answer must miss a required criterion. Infrastructure errors/refusals/broken-tool runs are labeled and rerun, not counted as capability failures; the reference answer must pass."},{"claim":"Maximum three task submissions per author","confirmed_in_draft":true,"where":"Section3.2 p4 line206"},{"claim":"Five to ten hours per task","confirmed_in_draft":true,"where":"Section3.2 p4 line206","qualification":"Author-reported time, not instrumented timing or a mandatory task duration."},{"claim":"AI-text score blocks above30% and rejects at40%","confirmed_in_draft":true,"where":"Section3.3 p4 lines213-215; AppendixC p17 lines892-893","qualification":"Text states >30% blocks next gate until rewrite; >=40% rejects. AppendixC says an earlier50% threshold was lowered during collection. Figure2 says under30%, so behavior at exactly30% is not fully consistent in the draft; verify portal before claiming exact equality."},{"claim":"Fifteen structured reviewer send-back reasons","confirmed_in_draft":true,"where":"Section3.3 p5 lines241-242; AppendixD p18 lines930-935","qualification":"The separate review checklist has12items;15 is the number of return reasons."},{"claim":"Frontier pilot models GPT-5.6 and Claude Opus4.8","confirmed_in_draft":true,"where":"Section3.3 p5 lines221-223","qualification":"Two pilots each under provider defaults. Older difficulty probe GPT-5.3 and reconciliation judge GPT-5.6 high are separate roles."}]

evidence-source-accuracy.json ADDED Viewed

	@@ -0,0 +1 @@

+ {"scope":{"tasks":100,"models":10,"native_runs":3000,"attempts_per_task_model":3,"criterion_count":663,"criterion_decisions":19890,"categories":{"both":390,"web":121,"local":152}},"method":{"classification":"Union of frozen supportingClaimIds categories: workspace_fact=local; web_fact=web; union=both. No inferred labels or model-generated labels.","accuracy":"Passed criterion decisions divided by all criterion decisions within each source group. Every required criterion has unit weight.","perfect_runs":"Original validated reward1.0 implies all positive-weight criteria passed; frozen contract and validation consumed.","confidence_interval":"Percentile 95% task-cluster bootstrap; 10000 resamples of 100 tasks with replacement, same draw for each model and category; three final attempts kept together.","seed":20261013,"new_model_calls":0,"new_grading_calls":0,"raw_archive_reads":0},"models":[{"model":"anthropic/claude-opus-5.5","name":"Opus 5.5","categories":{"local":{"passed":442,"total":456,"accuracy":0.9692982456140351,"ci95":[0.9406938958455893,0.9901960784313726],"task_count":63,"task_mean_accuracy":0.9594356261022927},"web":{"passed":355,"total":363,"accuracy":0.977961432506887,"ci95":[0.951280487804878,0.997289972899729],"task_count":59,"task_mean_accuracy":0.9755178907721281},"both":{"passed":1140,"total":1170,"accuracy":0.9743589743589743,"ci95":[0.9581195462478186,0.9873741851135468],"task_count":100,"task_mean_accuracy":0.9720185185185186}}},{"model":"gpt-6-astra","name":"GPT-6 Astra","categories":{"local":{"passed":442,"total":456,"accuracy":0.9692982456140351,"ci95":[0.9401197604790419,0.993103448275862],"task_count":63,"task_mean_accuracy":0.9653880070546736},"web":{"passed":355,"total":363,"accuracy":0.977961432506887,"ci95":[0.9534307136833229,0.9971014492753624],"task_count":59,"task_mean_accuracy":0.9759887005649718},"both":{"passed":1123,"total":1170,"accuracy":0.9598290598290599,"ci95":[0.935357383642535,0.9808027923211169],"task_count":100,"task_mean_accuracy":0.9469166666666666}}},{"model":"gpt-6.1-sol","name":"GPT-6.1 Sol","categories":{"local":{"passed":437,"total":456,"accuracy":0.9583333333333334,"ci95":[0.9288582732525438,0.9832134292565947],"task_count":63,"task_mean_accuracy":0.9543650793650794},"web":{"passed":354,"total":363,"accuracy":0.9752066115702479,"ci95":[0.946031746031746,0.9971264367816092],"task_count":59,"task_mean_accuracy":0.9675141242937854},"both":{"passed":1105,"total":1170,"accuracy":0.9444444444444444,"ci95":[0.9154923961632635,0.9698536890428281],"task_count":100,"task_mean_accuracy":0.9302777777777778}}},{"model":"anthropic/claude-opus-5","name":"Opus 5","categories":{"local":{"passed":434,"total":456,"accuracy":0.9517543859649122,"ci95":[0.9171717171717172,0.9791666666666666],"task_count":63,"task_mean_accuracy":0.9545855379188712},"web":{"passed":329,"total":363,"accuracy":0.90633608815427,"ci95":[0.8526309822174484,0.9538475177304965],"task_count":59,"task_mean_accuracy":0.8973634651600753},"both":{"passed":1074,"total":1170,"accuracy":0.9179487179487179,"ci95":[0.8908923346923231,0.941871869728818],"task_count":100,"task_mean_accuracy":0.9108108465608464}}},{"model":"x-ai/grok-4.6","name":"Grok 4.6","categories":{"local":{"passed":432,"total":456,"accuracy":0.9473684210526315,"ci95":[0.9125665983606558,0.9750011140819964],"task_count":63,"task_mean_accuracy":0.9320987654320988},"web":{"passed":325,"total":363,"accuracy":0.8953168044077136,"ci95":[0.847348779854621,0.9391370223978919],"task_count":59,"task_mean_accuracy":0.8894538606403013},"both":{"passed":1070,"total":1170,"accuracy":0.9145299145299145,"ci95":[0.8847176533101466,0.9410257985257985],"task_count":100,"task_mean_accuracy":0.9074457671957671}}},{"model":"gpt-5.6-sol","name":"GPT-5.6 Sol","categories":{"local":{"passed":433,"total":456,"accuracy":0.9495614035087719,"ci95":[0.9185171862509992,0.9761904761904762],"task_count":63,"task_mean_accuracy":0.9506172839506174},"web":{"passed":349,"total":363,"accuracy":0.9614325068870524,"ci95":[0.9338619289955168,0.9838709677419355],"task_count":59,"task_mean_accuracy":0.9576271186440678},"both":{"passed":1050,"total":1170,"accuracy":0.8974358974358975,"ci95":[0.8622446521961386,0.9289347036143242],"task_count":100,"task_mean_accuracy":0.8769880952380952}}},{"model":"google/gemini-3.8-flash","name":"Gemini 3.8 Flash","categories":{"local":{"passed":424,"total":456,"accuracy":0.9298245614035088,"ci95":[0.8941441441441441,0.9612403100775194],"task_count":63,"task_mean_accuracy":0.9243197278911566},"web":{"passed":311,"total":363,"accuracy":0.8567493112947658,"ci95":[0.8042318792318792,0.9066666666666666],"task_count":59,"task_mean_accuracy":0.8516949152542372},"both":{"passed":1031,"total":1170,"accuracy":0.8811965811965812,"ci95":[0.8472213476070529,0.9122807017543859],"task_count":100,"task_mean_accuracy":0.8623981481481482}}},{"model":"moonshotai/kimi-k3","name":"Kimi K3","categories":{"local":{"passed":414,"total":456,"accuracy":0.9078947368421053,"ci95":[0.8719361147327249,0.9396135265700483],"task_count":63,"task_mean_accuracy":0.9066830435878054},"web":{"passed":303,"total":363,"accuracy":0.8347107438016529,"ci95":[0.764102564102564,0.8948717948717949],"task_count":59,"task_mean_accuracy":0.8244821092278719},"both":{"passed":919,"total":1170,"accuracy":0.7854700854700855,"ci95":[0.737835695926474,0.8280106405106405],"task_count":100,"task_mean_accuracy":0.7562984607984609}}},{"model":"anthropic/claude-sonnet-5","name":"Sonnet 5","categories":{"local":{"passed":409,"total":456,"accuracy":0.8969298245614035,"ci95":[0.8524579735535811,0.9346846846846847],"task_count":63,"task_mean_accuracy":0.8816137566137567},"web":{"passed":293,"total":363,"accuracy":0.8071625344352618,"ci95":[0.7418540223106787,0.8661629045557616],"task_count":59,"task_mean_accuracy":0.7684557438794727},"both":{"passed":857,"total":1170,"accuracy":0.7324786324786324,"ci95":[0.6759516589683016,0.7837439052371938],"task_count":100,"task_mean_accuracy":0.6883015873015873}}},{"model":"thinkingmachines/inkling","name":"Inkling","categories":{"local":{"passed":322,"total":456,"accuracy":0.706140350877193,"ci95":[0.6293327586206896,0.7783791461916463],"task_count":63,"task_mean_accuracy":0.7074829931972789},"web":{"passed":142,"total":363,"accuracy":0.39118457300275483,"ci95":[0.31578615098351936,0.46264367816091956],"task_count":59,"task_mean_accuracy":0.34651600753295664},"both":{"passed":441,"total":1170,"accuracy":0.3769230769230769,"ci95":[0.3139412208449366,0.43792640790493415],"task_count":100,"task_mean_accuracy":0.3424011544011544}}}]}

grading-protocol.md ADDED Viewed

	@@ -0,0 +1,11 @@

+# Grading protocol
+The final answer is assessed against each task's frozen rubric, reference answer, supporting evidence and accepted-alternative policy. Each required rubricId receives a boolean verdict, rationale and candidate-answer evidence. Full-credit pass@1 counts reward exactly 1.0. Mean reward averages the saved fractional rewards, including valid zeros. No answer is regraded for this release.
+The sweep requested google/gemini-3.7-flash at temperature 0.0 and high reasoning effort through an OpenAI-compatible gateway. All 415 encrypted Opus 5.5/GPT-6.1 Sol records save that identifier in requested_model, model and resolved_model. None includes a separate provider model version. Google's official October 8, 2026 changelog says Gemini API requests using gemini-3.7-flash route to gemini-3.8-flash: https://ai.google.dev/gemini-api/docs/changelog#october-8-2026 . The saved gateway labels therefore do not establish the actual provider version. Archive dates are not judge-call timestamps. No new model call was made to resolve this uncertainty.
+Each run contains its original reward and full criterion judgments. The exported task question comes from the pinned expert task context. Per-run trace files retain the actual evaluated prompt wrappers and messages. Current working instruction.md and verifier.py were not substituted for historical evaluated sources, because they differ from the old task manifest.
+Task workspace files are under workspace/. Rubrics, reference answers, evidence maps and archived reference sources are under grading/ and must be withheld from solvers. Source rights and attribution remain in source_context; no blanket new license is asserted for third-party documents.
+The collection-stage pilots and reconciliation judge are distinct from the sweep judge. The supplied draft names two GPT-5.6 and two Opus 4.8 pilots. A completed answer-quality failure in at least one pilot is required; infrastructure failures do not count. See collection-protocol-check.json for all six verified draft claims and the exactly-30-percent ambiguity.

judge-model-audit.json ADDED Viewed

	@@ -0,0 +1 @@

+ {"scope":415,"model_counts":{"gpt-6.1-sol":190,"anthropic/claude-opus-5.5":225},"recorded_model_field_counts":[{"model_under_test":"gpt-6.1-sol","requested_model":"google/gemini-3.7-flash","model":"google/gemini-3.7-flash","resolved_model":"google/gemini-3.7-flash","provider_model_version":null,"runs":190},{"model_under_test":"anthropic/claude-opus-5.5","requested_model":"google/gemini-3.7-flash","model":"google/gemini-3.7-flash","resolved_model":"google/gemini-3.7-flash","provider_model_version":null,"runs":225}],"existing65decrypted_derivatives_consumed":65,"new350authenticated_member_decryptions":350,"all_selected_judge_hashes_match":true,"original_native_bytes_and_fields_unchanged":true,"no_decrypted_archive_or_key_or_plaintext_credentials_written":true,"new_models_grading_API_calls_or_acceptance_release":0,"existing_ledger_mutations":0,"archive_dates_not_promoted_to_judge_call_times":true,"provider_routing_may_differ_from_gateway_model_field":true,"record_count":415,"interpretation":"Saved gateway model labels do not establish actual upstream version after the announced rerouting."}

rollouts-00000.jsonl.gz ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:d071cba951991a9ba8d7b2e6ccb64720f4eb61e98ce3c3f2d6d84600bf981592
+size 2137750

rollouts-00001.jsonl.gz ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:460d5761da13437ac0a5dca226cee58dea68eb359af01f1940d3cd0097331e60
+size 38177687

tasks.jsonl.gz ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:c98dc1d64ba27ecc2c4993b5ae978552daedd0876a23931ec26d58b2e0e8cb4c
+size 688788

tasks.tar.gz ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:6cb279974de60c77b6fb8503c9e15845ac8ccb86af630cbc5e6239c1dd11a752
+size 458546816