AsapiThe AI Agent Trials
Asapi / Catalogue / Laundry‑shop operations app for owner, hosts, workers
Environment datasheet · Small business

Laundry‑shop operations app for owner, hosts, workers

ID laycode/task06 · revision 1 · tested 21 October 2026

A laundry owner asks for one web app to run her shop: host bookings, collection rounds, wash batches, photo proof, invoices, payments, supplies and bills, used by everyone at once. The agent builds it from her files and leaves a note.

Checks
161, in three weighted parts
Tested
7 models, one attempt each
Best
77% (Opus 5.5)
Median
65%
Full solves
0 of 7
Median attempt
39 min, 43 tool calls
What it tests

One app for the owner, the hosts and the workers, all at once

Unstated requirements

Prices, agreements, building rules and care terms are only in 53 plain‑text files; owner controls and safe recovery count too.

Agentic coding

The app must serve everyone at once and work with mock invoicing, payment and messaging services that the grader takes down or replays.

How it works

53 files, mock services and faults put in on purpose

The agent's world

A fresh container with the owner's 53 files, beside mock invoicing, payment and messaging services, worked through the model's usual coding agent.

Limits

Two hours, 150 model turns and a $20 spend cap at list prices, with web search off.

Grading

A tester model signs in as owner, hosts and workers through 16 stages, while the grader cuts the internet and makes services fail.

The reward

161 checks: the app's features count 60%, edge cases 20% and security 20%, from fixed trials through the app's real pages.

Results

Most attempts scored 65–77%, and three fell well below

Each dot is one attempt, and the line marks the median.

MedianFully solved
7 modelsOne attempt each65%0 of 7
ModelReward
Opus 5.5Anthropic77%
Sonnet 5.5Anthropic74%
GPT‑6 AstraOpenAI67%
GPT‑6.1 SolOpenAI65%
Fable 5.1Anthropic46%
Sonnet 5Anthropic33%
GPT‑6 LunaOpenAI16%

With an IT consultant's 70 requirements added to the request, four of the seven scored 74–81%, and the other three 13% or less.

What models get wrong

They leave the owner without controls, skip recovery, and get the money wrong

Unstated requirements

  1. No app let the owner add, edit and archive flats, buildings, prices, machines, suppliers, stock or bills.

  2. No app restored a backup, refused a damaged one, or survived an interrupted restore.

Agentic coding

  1. 71% of apps left passwords readable or had no sign‑in.

  2. No app applied payments to the right invoice with exact balances, or exported a month the accountant can reconcile.

Running it

What an attempt costs, and what the environment includes

Cost

The median attempt cost about $6 at list prices for the agent model; the tester's and judges' calls are extra.

What's in it

Her files, the mock services and their fault controls, the task in three versions, the full grader, and per‑check results for every attempt.

Format

Packaged as a Harbor task on request; grading needs Docker, a browser, and models for the tester and judges.

Fresh variants

Other files, people and facts in the same design, made on request and never published.

The files and grader aren't published; the job's row on The AI Agent Trials, our public benchmark, sums up the job and shows each model's score.

Try one of our three sample environments

Three environments are packaged as samples you can download, with their graders and reference solutions. Ask for them with your work email, or book a call about the rest of the catalogue.