--- license: cc-by-4.0 pretty_name: "Lilambd Empire Eval: grounded questions over living public ledgers" language: - en - ja task_categories: - question-answering - table-question-answering tags: - rag - evaluation - benchmark - agents - tabular - grounding - hallucination - abstention - contamination-free size_categories: - 10K#`). Use it to measure whether a RAG pipeline, a table agent or a tool-using LLM actually reads the data, and whether it says "not in the data" when the data has no answer. ## What makes it different - **Every answer is re-derivable.** `scripts/verify.py` reloads the dataset files, re-runs every recipe and re-reads every cited row. Result on 2026-11-03: **5,483 / 5,483 recipe answers reproduced exactly**. - **It tests abstention.** 173 questions have no answer in the data (a future month, a region with no station, a fictional name). The scorer reports a hallucination rate next to accuracy. - **It cannot be memorised.** 10 of the sets are fictional planets (Lilambd Worlds) simulated from plate tectonics to languages; nothing about them exists on the web, so a correct answer must come from the files. - **It renews itself.** The real-world ledgers are rebuilt from public measurement records (daily for some), and the generator scripts are included, so a fresh question set can be cut from any later revision. - **Held-out test split.** 30% of the questions are published without answers. Their answers are kept by the maintainer so the split stays clean. ## Sets | set | source dataset | dev | test | no-answer | licence of answers | |---|---|---:|---:|---:|---| | `japan-dpc-hospitals` | [Japan DPC Hospitals](https://huggingface.co/datasets/Lilambd/japan-dpc-hospitals): all acute-care hospitals: beds, discharges, case mix | 156 | 72 | 2 | CC-BY-4.0 | | `japan-prefecture-facts` | [Japan by Prefecture](https://huggingface.co/datasets/Lilambd/japan-prefecture-facts): sourced facts for all 47 prefectures | 136 | 70 | 0 | CC-BY-4.0 | | `machine-economy-daily` | [Machine Economy Daily](https://huggingface.co/datasets/Lilambd/machine-economy-daily): x402 hosts, MCP registry, agent packages | 81 | 40 | 2 | CC-BY-4.0 | | `security-daily` | [Security Daily](https://huggingface.co/datasets/Lilambd/security-daily): CVE, KEV, EPSS, breaches, joined daily | 140 | 62 | 6 | CC-BY-4.0 | | `world-biodiversity-cells` | [World Biodiversity Cells](https://huggingface.co/datasets/Lilambd/world-biodiversity-cells): GBIF records and species on a 0.5° grid | 181 | 72 | 0 | CC-BY-NC-4.0 | | `world-earth` | [World Earth Surface Ledger](https://huggingface.co/datasets/Lilambd/world-earth): forest loss, land cover, built-up area, soil per region | 220 | 110 | 18 | CC-BY-4.0 | | `world-ground` | [World Ground Ledger](https://huggingface.co/datasets/Lilambd/world-ground): people near volcanoes, coasts and plate boundaries; soil | 202 | 102 | 12 | CC-BY-4.0 | | `world-hazard` | [World Hazard Ledger](https://huggingface.co/datasets/Lilambd/world-hazard): earthquakes per region, station warming, disaster deaths | 306 | 113 | 22 | CC-BY-4.0 | | `world-marketplaces` | [World Marketplaces](https://huggingface.co/datasets/Lilambd/world-marketplaces): digital marketplaces and 409k physical markets | 250 | 101 | 5 | CDLA-Permissive-2.0 | | `world-markets` | [World Markets](https://huggingface.co/datasets/Lilambd/world-markets): 138k physical markets, bazaars, night markets | 161 | 55 | 0 | CDLA-Permissive-2.0 | | `world-prices` | [World Prices Long Record](https://huggingface.co/datasets/Lilambd/world-prices): commodity prices since 1960, CPI, energy prices | 217 | 101 | 20 | CC-BY-4.0 | | `world-pulse-daily` | [World Pulse Daily](https://huggingface.co/datasets/Lilambd/world-pulse-daily): 15 public catalogues written down daily | 235 | 94 | 2 | CC-BY-SA-4.0 | | `world-science` | [World Science](https://huggingface.co/datasets/Lilambd/world-science): OpenAlex paper counts by country, field, year | 248 | 80 | 10 | CC0-1.0 | | `world-seeds` | [World Seeds](https://huggingface.co/datasets/Lilambd/world-seeds): Wikipedia "by country" tables keyed by ISO2 | 124 | 58 | 0 | CC-BY-SA-4.0 | | `world-signals` | [World Signals](https://huggingface.co/datasets/Lilambd/world-signals): daily attention snapshot: Steam, GitHub, HF, packages, podcasts, search | 164 | 58 | 2 | CC-BY-4.0 | | `world-sound` | [World Sound](https://huggingface.co/datasets/Lilambd/world-sound): instruments, Cantometrics songs, tunings (dataset upload pending) | 241 | 127 | 12 | CC-BY-4.0 | | `worlds-*` (10 sets) | Lilambd Worlds fictional planets: Hibe, Dingongfi, Nenshee, Chortsin, Ithu, Deranhaa, Takcho, Kuyus, Sei, Dena | 1440 | 639 | 60 | CC-BY-4.0 | | **total** | | **4,502** | **1,954** | **173** | | Task types: lookup 3,635, count 769, rank 735, aggregate 322, multi_hop 286, compare 193, reasoning 179, unanswerable 173, temporal 70, derived 47, nunique 37, documentation 10. Skills tagged: `table_lookup` 2,537, `contamination_free` 2,079, `free_text` 973, `count` 769, `filter` 680, `ranking` 663, `fresh_data` 579, `temporal` 477, `multilingual_entity` 466, `id_lookup` 419, `aggregation` 293, `code_resolution` 246, `comparison` 193, `security` 191. ## One row ```json { "id": "world-hazard-0001", "dataset": "Lilambd/world-hazard", "revision": "hf-main@2026-11-03 (data as of 2026-11-01)", "task": "lookup", "question": "In the World Hazard Ledger, how many M6+ earthquakes from 1900 to 2026 are assigned to Eastern Samar (Philippines)?", "answer": "26", "answer_type": "number", "tolerance": { "abs": 0.0, "rel": null }, "evidence": [ { "row_id": "data/admin1_quake.csv#92", "file": "data/admin1_quake.csv", "row": 92, "key": "{\"a1\": \"PHL-2582\", \"name\": \"Eastern Samar\", \"country\": \"Philippines\"}" } ], "recipe": "{\"op\": \"lookup\", \"file\": \"data/admin1_quake.csv\", \"where\": [[\"name\", \"==\", \"Eastern Samar\"], [\"country\", \"==\", \"Philippines\"]], \"col\": \"m6\"}" } ``` Fields: `answer` (text form), `answer_json` (typed value: number, string, list, boolean or `null` for no answer), `answer_type`, `tolerance` (`abs` or `rel` for numbers), `evidence` (row ids with their key columns), `evidence_scope` (the filter behind counts, sums and rankings), `recipe` (JSON: the operation that produces the answer), `skills`, `difficulty` 1-3, `revision` (dataset revision the answer belongs to). ## Multilingual subset `data/multilingual/` holds the same 1,000 dev questions in `en`, `es`, `ja`, `zh`. Only the question text is translated; names, codes, numbers, column names and quoted strings are kept exactly, and the answer, evidence and recipe are the English item's. Each row keeps `source_id` and `question_en`, so a gap between languages is a reading gap, not a data gap. All recipe answers in every language file were re-derived by `verify.py`. Use the `multilingual` config. ## World Ledgers (SQL recipes) The `world-ledgers` config adds **9,999 questions** (dev 7,076, test 2,923, same hash split) over the Lilambd world ledgers of 2026-11-03: borders, calendars, cities, climate, currency, demand, food, languages, objects, population, power. 4,179 of them join two or three ledgers (city_id, ISO codes, nearest city). Here the recipe is a DuckDB query (`recipe_sql`) and the evidence is the list of files it reads (`evidence_files`); `question_ja` gives a Japanese wording. `scripts/verify_sql.py --views scripts/world_ledgers_views.json` re-runs every query; on 2026-11-03 it reproduced 9,999 / 9,999 answers (one item whose query no longer matched the ledger files was removed). Not every ledger file is on the Hub yet, so outside users can score against these answers but cannot all re-run the queries today. Licence: 1,652 of these rows (`commercial_ok: false`) rest on CShapes (non-commercial) or FAOSTAT production data; use them for non-commercial evaluation only. The rest follow CC BY 4.0 with source attribution (`licence_note`). ## Score a system ```bash pip install pandas pyarrow pycountry python scripts/make_prompts.py --qa data/dev --out prompts.jsonl --with-context-hint # run your system, write {"id": ..., "answer": ...} per line to predictions.jsonl, or use the bundled runner: ANTHROPIC_API_KEY=... python scripts/run_eval.py --qa data/dev --provider anthropic --out results/claude/predictions.jsonl --concurrency 8 OPENAI_API_KEY=... python scripts/run_eval.py --qa data/dev --provider openai --base-url http://localhost:8000/v1 --model MODEL --out results/MODEL/predictions.jsonl python scripts/score.py --qa data/dev --pred predictions.jsonl --out score.json ``` The scorer accepts numbers with commas or units, lists as JSON or comma-separated, and treats `null`, empty, "not in the dataset", "unknown" and similar as abstaining. It reports accuracy, answerable accuracy, abstention accuracy, hallucination rate and false-abstention rate, broken down by set, task, difficulty and skill. Use `--lenient` for free-text answers (the fictional-planet questions written by the world generator). Baselines on dev (from `results/baseline_*/score.json`): oracle (reads the answers) accuracy 1.0000, hallucination rate 0.00; always abstain accuracy 0.0362, hallucination rate 0.00; random guess accuracy 0.0003, hallucination rate 1.00. `leaderboard/build_leaderboard.py` turns any number of `score.json` files into `leaderboard.json` and a static `leaderboard.html`; `leaderboard/EVAL_CARD.md` has a model-index snippet for reporting scores on a model card. **Test split:** send `predictions.jsonl` for the test ids to submissions@lilambd.com and you get the same report back. ## Rebuild or extend `scripts/fetch_hf.py` downloads the source files pinned to their current commit, `scripts/build.py` generates questions (fixed seed), `scripts/verify.py` re-derives them, `scripts/package.py` makes the dev/test split. A private set over your own tables can be built the same way. ## Limits - Answers belong to the stated `revision`. The daily ledgers move on; re-verify against the same revision, or rebuild. - Many questions name a table or a column (for example `features_country.csv`, `m6_in`). They test grounded reading, not trivia. - Counting and ranking questions over large tables (hundreds of thousands of rows) are meant for systems that can run code or SQL; pure retrieval will not see every row. - About 970 fictional-planet questions come from the world generator's own question list: free-text answers, checked by the generator's consistency tests, not by recipes (`recipe` is empty, `verified_by` says so). - In the `worlds-*` sets, `source_qid` (for example `021-000`) is the world generator's internal question id, not a Wikidata QID. - Source keys that repeat with different values (31 country CPI series in world-prices `monthly.csv`, 26 region rows in world-earth `features_region.csv`) are left out of question generation, so no question has two possible answers. - Entity names are as written in the source; country names in questions come from ISO 3166 where the data has only a code. ## Licence Questions, recipes and scripts: CC BY 4.0, Lilambd. Answers are facts taken from the source datasets and keep their licence (column above): most are CC BY 4.0, World Science is CC0, the market sets are CDLA-Permissive-2.0. Cite the source dataset alongside this one. ```bibtex @misc{lilambd_empire_eval_2026, title = {Lilambd Empire Eval: grounded questions over living public ledgers}, author = {Lilambd}, year = {2026}, url = {https://huggingface.co/datasets/Lilambd/empire-eval} } ``` Need an evaluation set over your own data, a private held-out split, or a set in another language? submissions@lilambd.com --- **Missing something?** Tell us in one line what you'd gladly pay for, if someone made it — one-line form: https://docs.google.com/forms/d/1tzMQtHsEbMg7JyMk7YGqslHijfNTciByi8wrwjQpUPY/viewform (or email submissions@lilambd.com, subject "Data wish"). If enough people ask, we build it, and you get it free. **Used this?** Tell us what you used it for, what was missing, and what you used before: https://docs.google.com/forms/d/1tzMQtHsEbMg7JyMk7YGqslHijfNTciByi8wrwjQpUPY/viewform (one minute, no email asked). **Need a specific file or dataset made?** Ask here (no promise, quick form): https://docs.google.com/forms/d/e/1FAIpQLSf6V7YALfwE42SaF1re_cLzAoP96TpRg61jE3Nf7UTWEtWRxA/viewform