Add SMARD DE/LU 2024 evaluation pack

.gitattributes

@@ -60,3 +60,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
 online-retail-ii/1.0.0/database/online_retail_ii.duckdb filter=lfs diff=lfs merge=lfs -text
 online-retail-ii-derivation/1.0.0/database/online_retail_ii.duckdb filter=lfs diff=lfs merge=lfs -text
+smard-de-lu-2024/1.0.0/database/smard_de_lu_2024.duckdb filter=lfs diff=lfs merge=lfs -text

smard-de-lu-2024/1.0.0/ATTRIBUTION.md

@@ -0,0 +1,10 @@
+# Attribution and license
+
+Source: [SMARD market data](https://www.smard.de/), published by the German Federal
+Network Agency (Bundesnetzagentur).
+
+Attribution: `Bundesnetzagentur | SMARD.de`
+
+The source data is licensed under the [Creative Commons Attribution 4.0 International
+license](https://creativecommons.org/licenses/by/4.0/). This distribution preserves
+the attribution and does not change the source license.

smard-de-lu-2024/1.0.0/README.md

@@ -0,0 +1,20 @@
+# SMARD DE/LU 2024 evaluation pack
+
+Version `1.0.0` contains twelve completely specified quantitative questions, their
+Draft 2020-12 answer schemas, exact public reference answers, and the content-pinned
+SMARD DE/LU 2024 DuckDB database used to answer them. Eight cases form the core set and
+four form the generalization set.
+
+The database covers the 2024 `Europe/Berlin` calendar year at native 15-minute
+resolution (35,136 intervals). It combines realized generation and load, load and
+renewable forecasts, day-ahead prices, and commercial net exports. Historical source
+responses are checksum-pinned; this release does not query mutable live data.
+
+`source_level` preserves the predecessor task difficulty label; it is not a normalized
+cross-dataset difficulty rating. Reference answers are evaluator inputs. DSA keeps them
+outside model-visible requests, tools, traces, and run records. The source database
+remains immutable; each analysis receives a private writable copy.
+
+See `ATTRIBUTION.md` for source and license details. Exact content identities are
+recorded in `pack.json`; source request and response identities are retained in
+`database/build_report.json` and in the database's `metadata` schema.

smard-de-lu-2024/1.0.0/cases.jsonl

@@ -0,0 +1,12 @@
+{"answer_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"total_mwh":{"type":"number"}},"required":["total_mwh"],"type":"object"},"case_id":"smard-core-01-annual-wind-solar","case_version":"1","expected_answer":{"total_mwh":202138291.75},"metadata":{"family":"aggregation","source_level":"easy"},"question":"Use Europe/Berlin for calendar dates. Emit interval identities as ISO 8601 UTC strings ending in Z. Break ties by the earliest UTC interval or earliest local date. Consecutive intervals are adjacent fifteen-minute UTC identities. Forecast error is forecast minus actual. Round percentages to four decimal places, correlations to six decimal places, and all other calculated numbers to two decimal places, with halfway cases away from zero. Missing measurements are never imputed. Across eligible 2024 intervals, sum actual offshore wind, onshore wind, and photovoltaic generation and return total_mwh. Return JSON only according to the response schema."}
+{"answer_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"date":{"type":"string"},"interval_count":{"type":"integer"},"photovoltaic_mwh":{"type":"number"}},"required":["date","photovoltaic_mwh","interval_count"],"type":"object"},"case_id":"smard-core-02-peak-pv-day","case_version":"1","expected_answer":{"date":"2024-06-25","interval_count":96,"photovoltaic_mwh":425934.75},"metadata":{"family":"daily-aggregation","source_level":"easy"},"question":"Use Europe/Berlin for calendar dates. Emit interval identities as ISO 8601 UTC strings ending in Z. Break ties by the earliest UTC interval or earliest local date. Consecutive intervals are adjacent fifteen-minute UTC identities. Forecast error is forecast minus actual. Round percentages to four decimal places, correlations to six decimal places, and all other calculated numbers to two decimal places, with halfway cases away from zero. Missing measurements are never imputed. Find the earliest local date with the greatest total actual photovoltaic generation and return date, photovoltaic_mwh, and interval_count. Return JSON only according to the response schema."}
+{"answer_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"date":{"type":"string"},"eligible_intervals":{"type":"integer"},"grid_load_mwh":{"type":"number"},"renewable_mwh":{"type":"number"},"share_percent":{"type":"number"}},"required":["date","renewable_mwh","grid_load_mwh","share_percent","eligible_intervals"],"type":"object"},"case_id":"smard-core-03-peak-renewable-share-day","case_version":"1","expected_answer":{"date":"2024-10-13","eligible_intervals":96,"grid_load_mwh":1094219.75,"renewable_mwh":1113206.0,"share_percent":101.7351},"metadata":{"family":"daily-ratio","source_level":"medium"},"question":"Use Europe/Berlin for calendar dates. Emit interval identities as ISO 8601 UTC strings ending in Z. Break ties by the earliest UTC interval or earliest local date. Consecutive intervals are adjacent fifteen-minute UTC identities. Forecast error is forecast minus actual. Round percentages to four decimal places, correlations to six decimal places, and all other calculated numbers to two decimal places, with halfway cases away from zero. Missing measurements are never imputed. Find the earliest local date with the greatest ratio of total actual renewable generation to realized grid load over the same complete intervals. Return date, renewable_mwh, grid_load_mwh, share_percent, and eligible_intervals. Return JSON only according to the response schema."}
+{"answer_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"grid_load_mwh":{"type":"number"},"interval_start_local":{"type":"string"},"interval_start_utc":{"type":"string"}},"required":["interval_start_utc","interval_start_local","grid_load_mwh"],"type":"object"},"case_id":"smard-core-04-peak-grid-load","case_version":"1","expected_answer":{"grid_load_mwh":19140.0,"interval_start_local":"2024-01-15T11:30:00","interval_start_utc":"2024-01-15T10:30:00Z"},"metadata":{"family":"extreme-interval","source_level":"easy"},"question":"Use Europe/Berlin for calendar dates. Emit interval identities as ISO 8601 UTC strings ending in Z. Break ties by the earliest UTC interval or earliest local date. Consecutive intervals are adjacent fifteen-minute UTC identities. Forecast error is forecast minus actual. Round percentages to four decimal places, correlations to six decimal places, and all other calculated numbers to two decimal places, with halfway cases away from zero. Missing measurements are never imputed. Find the earliest interval with maximum realized grid load and return interval_start_utc, interval_start_local, and grid_load_mwh. Return JSON only according to the response schema."}
+{"answer_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"eligible_intervals":{"type":"integer"},"negative_intervals":{"type":"integer"},"percent":{"type":"number"}},"required":["negative_intervals","eligible_intervals","percent"],"type":"object"},"case_id":"smard-core-05-negative-price-prevalence","case_version":"1","expected_answer":{"eligible_intervals":35136,"negative_intervals":1828,"percent":5.2026},"metadata":{"family":"price-prevalence","source_level":"easy"},"question":"Use Europe/Berlin for calendar dates. Emit interval identities as ISO 8601 UTC strings ending in Z. Break ties by the earliest UTC interval or earliest local date. Consecutive intervals are adjacent fifteen-minute UTC identities. Forecast error is forecast minus actual. Round percentages to four decimal places, correlations to six decimal places, and all other calculated numbers to two decimal places, with halfway cases away from zero. Missing measurements are never imputed. Count strictly negative day-ahead price intervals among intervals with a non-null price and return negative_intervals, eligible_intervals, and percent. Return JSON only according to the response schema."}
+{"answer_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"duration_minutes":{"type":"integer"},"end_utc_exclusive":{"type":"string"},"interval_count":{"type":"integer"},"minimum_price_eur_per_mwh":{"type":"number"},"start_utc":{"type":"string"}},"required":["start_utc","end_utc_exclusive","interval_count","duration_minutes","minimum_price_eur_per_mwh"],"type":"object"},"case_id":"smard-core-06-longest-negative-price-run","case_version":"1","expected_answer":{"duration_minutes":1080,"end_utc_exclusive":"2024-07-07T16:00:00Z","interval_count":72,"minimum_price_eur_per_mwh":-22.37,"start_utc":"2024-07-06T22:00:00Z"},"metadata":{"family":"consecutive-intervals","source_level":"medium"},"question":"Use Europe/Berlin for calendar dates. Emit interval identities as ISO 8601 UTC strings ending in Z. Break ties by the earliest UTC interval or earliest local date. Consecutive intervals are adjacent fifteen-minute UTC identities. Forecast error is forecast minus actual. Round percentages to four decimal places, correlations to six decimal places, and all other calculated numbers to two decimal places, with halfway cases away from zero. Missing measurements are never imputed. Find the earliest longest run of adjacent strictly negative day-ahead prices and return start_utc, end_utc_exclusive, interval_count, duration_minutes, and minimum_price_eur_per_mwh. Return JSON only according to the response schema."}
+{"answer_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"bias_mwh":{"type":"number"},"eligible_intervals":{"type":"integer"},"mae_mwh":{"type":"number"}},"required":["eligible_intervals","mae_mwh","bias_mwh"],"type":"object"},"case_id":"smard-core-07-load-forecast-error","case_version":"1","expected_answer":{"bias_mwh":-99.35,"eligible_intervals":35132,"mae_mwh":494.81},"metadata":{"family":"forecast-error","source_level":"medium"},"question":"Use Europe/Berlin for calendar dates. Emit interval identities as ISO 8601 UTC strings ending in Z. Break ties by the earliest UTC interval or earliest local date. Consecutive intervals are adjacent fifteen-minute UTC identities. Forecast error is forecast minus actual. Round percentages to four decimal places, correlations to six decimal places, and all other calculated numbers to two decimal places, with halfway cases away from zero. Missing measurements are never imputed. For paired non-null forecast and realized grid load, return eligible_intervals, mae_mwh, and bias_mwh. Return JSON only according to the response schema."}
+{"answer_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"offshore_wind":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"bias_mwh":{"type":"number"},"eligible_intervals":{"type":"integer"},"mae_mwh":{"type":"number"}},"required":["eligible_intervals","mae_mwh","bias_mwh"],"type":"object"},"onshore_wind":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"bias_mwh":{"type":"number"},"eligible_intervals":{"type":"integer"},"mae_mwh":{"type":"number"}},"required":["eligible_intervals","mae_mwh","bias_mwh"],"type":"object"},"photovoltaic":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"bias_mwh":{"type":"number"},"eligible_intervals":{"type":"integer"},"mae_mwh":{"type":"number"}},"required":["eligible_intervals","mae_mwh","bias_mwh"],"type":"object"}},"required":["offshore_wind","onshore_wind","photovoltaic"],"type":"object"},"case_id":"smard-core-08-renewable-forecast-error","case_version":"1","expected_answer":{"offshore_wind":{"bias_mwh":11.36,"eligible_intervals":35136,"mae_mwh":143.88},"onshore_wind":{"bias_mwh":-5.76,"eligible_intervals":35132,"mae_mwh":285.0},"photovoltaic":{"bias_mwh":30.61,"eligible_intervals":35132,"mae_mwh":144.47}},"metadata":{"family":"forecast-error","source_level":"medium"},"question":"Use Europe/Berlin for calendar dates. Emit interval identities as ISO 8601 UTC strings ending in Z. Break ties by the earliest UTC interval or earliest local date. Consecutive intervals are adjacent fifteen-minute UTC identities. Forecast error is forecast minus actual. Round percentages to four decimal places, correlations to six decimal places, and all other calculated numbers to two decimal places, with halfway cases away from zero. Missing measurements are never imputed. For paired non-null forecast and actual offshore wind, onshore wind, and photovoltaic generation, return separate offshore_wind, onshore_wind, and photovoltaic objects with eligible_intervals, mae_mwh, and bias_mwh. Return JSON only according to the response schema."}
+{"answer_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"eligible_intervals":{"type":"integer"},"pearson_r":{"type":"number"}},"required":["eligible_intervals","pearson_r"],"type":"object"},"case_id":"smard-gen-01-price-renewables-correlation","case_version":"1","expected_answer":{"eligible_intervals":35136,"pearson_r":-0.605479},"metadata":{"family":"correlation","source_level":"hard"},"question":"Use Europe/Berlin for calendar dates. Emit interval identities as ISO 8601 UTC strings ending in Z. Break ties by the earliest UTC interval or earliest local date. Consecutive intervals are adjacent fifteen-minute UTC identities. Forecast error is forecast minus actual. Round percentages to four decimal places, correlations to six decimal places, and all other calculated numbers to two decimal places, with halfway cases away from zero. Missing measurements are never imputed. For paired complete measurements, calculate the Pearson correlation between day-ahead price and actual offshore wind plus onshore wind plus photovoltaic generation and return eligible_intervals and pearson_r. Return JSON only according to the response schema."}
+{"answer_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"eligible_intervals":{"type":"integer"},"net_export_mwh":{"type":"number"}},"required":["eligible_intervals","net_export_mwh"],"type":"object"},"case_id":"smard-gen-02-net-export-during-negative-prices","case_version":"1","expected_answer":{"eligible_intervals":1828,"net_export_mwh":2914895.71},"metadata":{"family":"conditional-aggregation","source_level":"medium"},"question":"Use Europe/Berlin for calendar dates. Emit interval identities as ISO 8601 UTC strings ending in Z. Break ties by the earliest UTC interval or earliest local date. Consecutive intervals are adjacent fifteen-minute UTC identities. Forecast error is forecast minus actual. Round percentages to four decimal places, correlations to six decimal places, and all other calculated numbers to two decimal places, with halfway cases away from zero. Missing measurements are never imputed. Over intervals with strictly negative day-ahead prices and non-null SMARD commercial net export, sum exports minus imports and return eligible_intervals and net_export_mwh. Return JSON only according to the response schema."}
+{"answer_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"eligible_intervals":{"type":"integer"},"exceeding_intervals":{"type":"integer"},"percent":{"type":"number"}},"required":["eligible_intervals","exceeding_intervals","percent"],"type":"object"},"case_id":"smard-gen-03-wind-solar-exceeds-load","case_version":"1","expected_answer":{"eligible_intervals":35136,"exceeding_intervals":278,"percent":0.7912},"metadata":{"family":"conditional-ratio","source_level":"medium"},"question":"Use Europe/Berlin for calendar dates. Emit interval identities as ISO 8601 UTC strings ending in Z. Break ties by the earliest UTC interval or earliest local date. Consecutive intervals are adjacent fifteen-minute UTC identities. Forecast error is forecast minus actual. Round percentages to four decimal places, correlations to six decimal places, and all other calculated numbers to two decimal places, with halfway cases away from zero. Missing measurements are never imputed. Among paired complete intervals, count where actual offshore wind plus onshore wind plus photovoltaic generation strictly exceeds realized grid load and return eligible_intervals, exceeding_intervals, and percent. Return JSON only according to the response schema."}
+{"answer_schema":{"$schema":"https://json-schema.org/draft/2020-12/schema","additionalProperties":false,"properties":{"from_interval_start_utc":{"type":"string"},"increase_mwh":{"type":"number"},"to_interval_start_utc":{"type":"string"}},"required":["from_interval_start_utc","to_interval_start_utc","increase_mwh"],"type":"object"},"case_id":"smard-gen-04-largest-residual-load-increase","case_version":"1","expected_answer":{"from_interval_start_utc":"2024-05-13T04:00:00Z","increase_mwh":1700.25,"to_interval_start_utc":"2024-05-13T04:15:00Z"},"metadata":{"family":"adjacent-change","source_level":"hard"},"question":"Use Europe/Berlin for calendar dates. Emit interval identities as ISO 8601 UTC strings ending in Z. Break ties by the earliest UTC interval or earliest local date. Consecutive intervals are adjacent fifteen-minute UTC identities. Forecast error is forecast minus actual. Round percentages to four decimal places, correlations to six decimal places, and all other calculated numbers to two decimal places, with halfway cases away from zero. Missing measurements are never imputed. Find the earliest greatest increase between adjacent realized residual-load intervals and return from_interval_start_utc, to_interval_start_utc, and increase_mwh. Return JSON only according to the response schema."}

smard-de-lu-2024/1.0.0/database/build_report.json

@@ -0,0 +1,71 @@
+{
+  "attribution": "Bundesnetzagentur | SMARD.de",
+  "built_at_utc": "2026-08-26T02:00:00+02:00",
+  "database_id": "smard_de_lu_2024-1.0.0",
+  "database_sha256": "249143dd8b399a6be84d666190f16b89ebc71545783362a50efaa40dc02bc45b",
+  "interval_count": 35136,
+  "license": "CC BY 4.0",
+  "period_end_local_exclusive": "2025-01-01T00:00:00",
+  "period_start_local": "2024-01-01T00:00:00",
+  "region": "DE-LU",
+  "source_files": [
+    {
+      "raw_byte_count": 1549390,
+      "raw_sha256": "c8db3ac29e3ccb3af04b812fd6e498a33233c877189a720e2fd2b2b7aaeebf4f",
+      "request_json": "{\"request_form\":[{\"format\":\"CSV\",\"language\":\"de\",\"moduleIds\":[22004629],\"region\":\"DE-LU\",\"resolution\":\"\",\"timestamp_from\":1704063600000,\"timestamp_to\":1735686000000,\"type\":\"discrete\"}]}",
+      "response_filename": "commercial_net_export.csv",
+      "retrieved_at_utc": "2026-08-26T02:00:00+02:00",
+      "row_count": 35136,
+      "source_file_id": "commercial_net_export"
+    },
+    {
+      "raw_byte_count": 1448825,
+      "raw_sha256": "fa696975e315d3fd21711cdcb60018feafe978b1603aec8c4edd9cb347286cfb",
+      "request_json": "{\"request_form\":[{\"format\":\"CSV\",\"language\":\"de\",\"moduleIds\":[8004169],\"region\":\"DE-LU\",\"resolution\":\"\",\"timestamp_from\":1704063600000,\"timestamp_to\":1735686000000,\"type\":\"discrete\"}]}",
+      "response_filename": "day_ahead_price.csv",
+      "retrieved_at_utc": "2026-08-26T02:00:00+02:00",
+      "row_count": 35136,
+      "source_file_id": "day_ahead_price"
+    },
+    {
+      "raw_byte_count": 1902059,
+      "raw_sha256": "538717a2ab47fc33344be78254d975003987460e8658eae0c0d752eccf30640e",
+      "request_json": "{\"request_form\":[{\"format\":\"CSV\",\"language\":\"de\",\"moduleIds\":[6000411,6004362],\"region\":\"DE-LU\",\"resolution\":\"\",\"timestamp_from\":1704063600000,\"timestamp_to\":1735686000000,\"type\":\"discrete\"}]}",
+      "response_filename": "forecast_load.csv",
+      "retrieved_at_utc": "2026-08-26T02:00:00+02:00",
+      "row_count": 35136,
+      "source_file_id": "forecast_load"
+    },
+    {
+      "raw_byte_count": 2039921,
+      "raw_sha256": "cbd02695e9c66243cf5f33d731761b3be0870fdbcf0fa950674ac9727f35f6ac",
+      "request_json": "{\"request_form\":[{\"format\":\"CSV\",\"language\":\"de\",\"moduleIds\":[2000123,2000125,2003791],\"region\":\"DE-LU\",\"resolution\":\"\",\"timestamp_from\":1704063600000,\"timestamp_to\":1735686000000,\"type\":\"discrete\"}]}",
+      "response_filename": "forecast_renewables.csv",
+      "retrieved_at_utc": "2026-08-26T02:00:00+02:00",
+      "row_count": 35136,
+      "source_file_id": "forecast_renewables"
+    },
+    {
+      "raw_byte_count": 4207065,
+      "raw_sha256": "aee0a2686f8e2fad9fea6d8796aab4365197c92be39f9036b26f62bd61abe2df",
+      "request_json": "{\"request_form\":[{\"format\":\"CSV\",\"language\":\"de\",\"moduleIds\":[1001224,1004066,1004067,1004068,1001223,1004069,1004071,1004070,1001226,1001228,1001227,1001225],\"region\":\"DE-LU\",\"resolution\":\"\",\"timestamp_from\":1704063600000,\"timestamp_to\":1735686000000,\"type\":\"discrete\"}]}",
+      "response_filename": "realized_generation.csv",
+      "retrieved_at_utc": "2026-08-26T02:00:00+02:00",
+      "row_count": 35136,
+      "source_file_id": "realized_generation"
+    },
+    {
+      "raw_byte_count": 1902029,
+      "raw_sha256": "4dcf8e7da20dbae6bccf0101ebfc242743d8223d1027ed3be2e50cd09699cd5c",
+      "request_json": "{\"request_form\":[{\"format\":\"CSV\",\"language\":\"de\",\"moduleIds\":[5000410,5004359],\"region\":\"DE-LU\",\"resolution\":\"\",\"timestamp_from\":1704063600000,\"timestamp_to\":1735686000000,\"type\":\"discrete\"}]}",
+      "response_filename": "realized_load.csv",
+      "retrieved_at_utc": "2026-08-26T02:00:00+02:00",
+      "row_count": 35136,
+      "source_file_id": "realized_load"
+    }
+  ],
+  "source_name": "SMARD market data",
+  "source_url": "https://www.smard.de",
+  "timezone": "Europe/Berlin",
+  "version": "1.0.0"
+}

smard-de-lu-2024/1.0.0/database/smard_de_lu_2024.duckdb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:249143dd8b399a6be84d666190f16b89ebc71545783362a50efaa40dc02bc45b
+size 25964544

smard-de-lu-2024/1.0.0/pack.json

@@ -0,0 +1 @@
+{"cases":{"case_count":12,"path":"cases.jsonl","sha256":"0bb698f392fece3e3b39b6f296087b6d6c5dc64730d03b8ed86aa35ae7305667","size_bytes":15979},"database":{"id":"smard_de_lu_2024-1.0.0","path":"database/smard_de_lu_2024.duckdb","sha256":"249143dd8b399a6be84d666190f16b89ebc71545783362a50efaa40dc02bc45b","size_bytes":25964544},"format":"dsa-evaluation-pack/v1","license":"CC-BY-4.0","pack_id":"smard-de-lu-2024","provenance":{"source_datasets":[{"case_count":8,"export_sha256":"b6d5fd6c31fa6d0b515cc0ced6cc2572ccec7003a0f39b60480b017a26e99f70","name":"smard-de-lu-2024-core-v1","version":"1"},{"case_count":4,"export_sha256":"7285de5262dceff46b28a9b9f5a9e2cc5afcdfbd7f4c3f1f0dfe61cc7e6c6e67","name":"smard-de-lu-2024-generalization-v1","version":"1"}]},"scorer":{"name":"exact-json","version":"1"},"version":"1.0.0"}