README.md

metadata

task_categories:
  - question-answering
language:
  - en
tags:
  - finance
  - search
  - benchmark
size_categories:
  - n<1K
configs:
  - config_name: historical-lookup
    data_files:
      - split: public
        path: historical-lookup.jsonl
  - config_name: multi-hop-lookup
    data_files:
      - split: public
        path: multi-hop-lookup.jsonl
  - config_name: live-lookup
    data_files:
      - split: public
        path: live-lookup.jsonl

Finance Search

The public sample of the OpenBenchmarks finance search benchmark, which measures how well a search API helps an agent answer a finance professional's questions. The agent, prompt, search budget and grader are held fixed; only the search API changes. This sample is published so the question format and a harness can be checked end to end. Results on it are not leaderboard scores.

Configurations

Configuration	Rows	Task
historical-lookup	10	One published figure for a named entity and period; one well-aimed search should find it.
multi-hop-lookup	5	Several published facts combined: a peer comparison, a trend, a result against guidance.
live-lookup	10	A price at a given moment, asked shortly after it: 3 stocks, 3 coins, 4 currency pairs.

Each configuration has one split named public.

Fields

historical-lookup and multi-hop-lookup:

• id: question id.
• question: the question as asked, with named entities and absolute dates.
• answer: the reference answer. Where it ends with "Wrong:", the listed values are the expected wrong answers.
• tolerance: how close a numeric answer must be.
• area: one of company results, banks and insurers, macro and policy, small and international, markets.

live-lookup:

• id, asset (equity, crypto, fx), name, ticker.
• query_template: the query a search API receives once a date and a time in UTC are filled in, for example <name> (<ticker>) stock price {{date}} {{time}}, price for crypto and exchange rate for currency pairs.

Live lookup

Live lookup measures freshness: how soon a new price reaches a search API's results. In the benchmark each item is asked shortly after a moment that has just passed and graded against market data pulled afterwards, so a live answer only exists once that moment has passed. These rows carry the items and the query format, without answers.

Grading

An answer is correct only if every part the question asks for is right within its tolerance. A hedged answer counts only if its committed value is right.

License

No license is assigned yet.