EvalMarket
Back to Explore

Everyday Math Reasoning

Ten original word problems with numeric-answer grading across ratios, time, percentages, and multi-step arithmetic.

Data analysis By Matthew Keeley

10 casesReviewed version 1CSV / JSON / deterministic runner

Public sample

1 case

Ticket Revenue

Prompt given to the agent

A museum sold 18 adult tickets at $14 and 7 child tickets at $8. What was the total revenue in dollars? Respond with only the numeric answer.

Expected response

308

How it is scored

Exact answer matchSurrounding whitespace is ignored.
See grading details

The first case demonstrates the input, required numeric response, and exact-match grading.

The first case demonstrates the input, required numeric response, and exact-match grading.

{
  "method": "exact_match",
  "expected": "308",
  "trim_whitespace": true
}

Try these samples with your agent

What you get

Contents
10 cases · CSV / JSON / deterministic runner
Made for
Builders testing an assistant's basic reasoning and response formatting.
Setup and compatibilityHow to run the pack with your agent

Download and unzip this pack. Install Node.js 22 or newer and sign in to your agent's CLI locally. Open a terminal in the unzipped folder containing collect.cjs.

node collect.cjs claude --local

This answers each case in a fresh session, grades locally, and saves the results in grading-results.json. Deterministic grading needs no npm install. Your local model usage applies.

Already have responses.json? Run node collect.cjs --local to grade it without calling your agent again.

For server grading, start a public-sample preview and follow its instructions to download the selected cases and run with --server. Acquired packs also support deterministic server grading from My library.

Model options and other tools

Append your usual CLI flags after --, for example:

node collect.cjs claude --local -- --model sonnet --effort high

Use a model available to your account, normal terminal quoting, and absolute paths for configuration files. The runner manages task prompts, fresh sessions and output capture. For another agent, follow the included Generic guide.

Promptfoo is an optional report viewer. After collecting answers, run npm --prefix optional/promptfoo install, npm --prefix optional/promptfoo run eval, then npm --prefix optional/promptfoo run view. See README.md for the current setup instructions.

Seller setup notes

These notes were supplied with this published version and may describe the seller's original workflow. Use the commands above or the downloaded README.md for the current runner.

Any text-generation agent that can read a prompt and return plain text.

Send each case's prompt to the agent. Submit its complete numeric answer for grading. Use the same model settings across cases when comparing runs.

Scope and limitationsKnow what this pack does and does not measure

Limitations

Small, public, English-only synthetic sample. Exact result checks do not measure reasoning quality; they reject alternate formats even when the underlying result is correct. This is not GSM8K and is not comparable to its published scores.

Public sample coverage

The public sample shows one task and answer format; the remaining cases use different calculations.

Review covers version 1 of this pack, not the readiness of your agent.

Reviews and questionsBuyer feedback and answers from the seller

Buyer reviews

How well does this eval measure what it claims? Reviews come from people who acquired it through EvalMarket.

Loading reviews…

Checking review access…

Questions

Ask about compatibility, coverage or setup. The seller can answer here.

Loading questions…

Checking sign-in…